School‐based self‐management interventions for asthma in children and adolescents: a mixed methods systematic review

Summary of findings for the main comparison. Effects of school‐based asthma interventions compared to usual care for asthma among children and adolescents

Effects of school‐based asthma interventions compared to usual care for asthma among children and adolescents
Patient or population: asthma among children and adolescents Setting: primary/elementary schools through to high/senior schools Intervention: effects of school‐based asthma interventions Comparison: usual care
Outcomes	*Anticipated absolute effects (95% CI)**		Relative effect (95% CI)	No. of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with usual care	Risk with effect of school‐based asthma interventions
Exacerbations leading to hospitalisation (hospitalisations) assessed with RCT Follow‐up: range 1 week to 12 months		Mean level of hospitalisation at post‐treatment in the intervention group was 0.19 standard deviations lower than in the control group (0.35 to 0.04 lower)	‐	1873 (6 RCTs)	⊕⊕⊕⊝ MODERATE^a	Meta‐analysis based on SMD including data transformed from OR (data on median level from Gerald 2006 not included)
Asthma symptoms leading to emergency hospital visits (ED visits) Follow‐up: range 1 week to 12 months	Less than 10% experience ED visit annually		OR 0.70 (0.53 to 0.92)	3883 (13 RCTs)	⊕⊕⊝⊝ LOW^b	Data from Gerald 2006 on median visits not combined Assumed risk based on rates over 12 months < 10% based on Horner 2008, McGhan 2010, Velsor‐Friedrich 2005 ≥ 10% based on Cicutto 2013, McGhan 2003
	75 per 1000	54 per 1000 (41 to 69)
	Over 10% experience ED visit annually
	281 per 1000	215 per 1000 (172 to 264)
Unplanned visit to hospital or GP due to asthma symptoms (unplanned medical visits) Follow‐up: range 1 week to 12 months	Unplanned visits over 6 to 9 months		OR 0.74 (0.60 to 0.90)	3283 (5 RCTs)	⊕⊕⊕⊝ MODERATE^c	Unplanned visits over 6 to 9 months based on McGhan 2003, Splett 2006; unplanned visits over 12 months based on Cicutto 2013, McGhan 2010
	264 per 1000	210 per 1000 (177 to 244)
	Unplanned visits over 12 months
	318 per 1000	257 per 1000 (219 to 296)
Absence from school Follow‐up: range 1 week to 15 months	Mean absence from school was 4.3 school days missed annually	MD 0.399 school days missed annually lower (1.254 lower to 0.456 higher)	‐	4609 (10 RCTs)	⊕⊕⊝⊝ LOW^d	Meta‐analysis based on SMD including data transformed from OR; transformation to mean difference undertaken based on data from Cicutto 2005
Experience of daytime and night‐time symptoms ‐ daytime symptoms (daytime symptoms) Follow‐up: range 2 months to 12 months	Mean experience of daytime and night‐time symptoms ‐ daytime symptoms was 3.3 days experienced in past 2 weeks	MD 0.377 days experienced in past 2 weeks lower (0.828 lower to 0.05 higher)	‐	1065 (5 RCTs)	⊕⊕⊕⊝ MODERATE^e	CI for this pooled estimate crossed the line of no effect by a small margin. Original meta‐analysis based on SMDs, including transformations from ORs. SMD to MD based on Bruzzese 2011
Use of reliever therapies, e.g. beta₂‐agonists (reliever therapies) Follow‐up: range 1 week to 15 months	Study population		OR 0.52 (0.15 to 1.81)	437 (2 RCTs)	⊕⊝⊝⊝ VERY LOW^f
	228 per 1000	133 per 1000 (42 to 349)
Health‐related quality of life Follow‐up: range 1 week to 12 months	Mean health‐related quality of life was 4.96 Paediatric Asthma Quality of Life Questionnaire points	MD 0.36 Paediatric Asthma Quality of Life Questionnaire points higher (0.06 higher to 0.64 higher)	‐	2587 (7 RCTs)	⊕⊕⊕⊝ MODERATE^g	Two studies provided information on change in QoL. Both showed positive intervention effects. Risk with usual care based on follow‐up scores
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; ED: emergency department; GP: general practitioner; MD: mean difference; OR: odds ratio; QoL: quality of life; RCT: randomised controlled trial; RR: risk ratio; SMD: standardised mean difference.
GRADE Working Group grades of evidence. High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.
^aStudies with high or unclear risk of bias contribute the least to the overall effect size. Hospitalisations may be due to reasons other than asthma (‐1 for indirectness). ^bFour studies had high risk of bias around allocation concealment; four also had high risk of bias around attrition; many other studies had unclear risks of bias. However, these risks did not appear to inflate the effect size nor systematically influence the effect. A high degree of inconsistency was evident, as measured by heterogeneity statistics in the meta‐analysis, which was partially explained by subgroup analyses. A large degree of variation was evident in measurement of the outcome, prompting concerns about indirectness; similarly, wide confidence intervals were detected (0.53 to 0.95). Study results led to concerns that not all ED visits may be due to asthma (‐1 for inconsistency; ‐1 for indirectness). ^cNo guarantee that unplanned medical visits were due to asthma (‐1 for indirectness). ^dSchool absences could be due to causes other than asthma; heterogeneity statistics suggested a large degree of statistical inconsistency (‐1 for indirectness; ‐1 for inconsistency). ^eHigh risk of bias detected in at least one domain for two out of five studies, which accounted for around a third of the pooled effect size. This included high risk of bias suspected for attrition bias in one study (‐1 for risk of bias). ^fRisk of bias deemed high for attrition and reporting bias for one of the two studies included in the meta‐analysis; very wide confidence interval; although both studies were consistent in the direction of effect, they showed large differences in the magnitude of effect (‐1 for risk of bias; ‐1 for inconsistency; ‐1 for imprecision). ^gImprecision was deemed to be serious based on the nature of the outcome; five of the seven studies were deemed to have high risk of bias in at least one domain. This included three studies deemed to have high risk of bias for allocation concealment. However, these did not appear to differentially influence the effect size (‐1 for imprecision).

Background

Description of the condition

Asthma is a chronic respiratory condition characterised by bronchoconstriction, airway inflammation, and mucus hypersecretion leading to variable airflow limitation. Resulting symptoms include wheeze, dyspnoea, cough, and tightness in the chest. No single definitive diagnostic 'test' for asthma is available; instead asthma is diagnosed clinically upon assessment of respiratory symptoms and clinical response to inhaled therapy, and review of evidence of reversible airflow limitation or airway hyper‐responsiveness ‐ as in BTS 2016 and Levy 2014 ‐ and elevated exhaled breath nitric oxide ‐ as in NICE 2017. Asthma is the most common chronic disease among children (Neuzil 2000; To 2012), with more than a million children in the UK living with this chronic condition (Asthma UK 2013). Many countries report high prevalence rates of childhood asthma. The International Study of Asthma and Allergy in Children (ISAAC) study, for example, found high prevalence in Australasia and the United Kingdom (Asher 2006). Much of the evidence on non‐pharmacological interventions derives from North America, where prevalence is among the highest globally, at 21.5% and 16.7% for six‐ to seven‐year‐old boys and girls, respectively, and 19.8% and 23.3% among children 13 to 14 years of age (Mallol 2013).

In the UK, children from black and white ethnic backgrounds have higher levels of asthma symptoms compared with children from South Asian backgrounds (Netuveli 2005), although substantial variation in the risk of developing asthma has been found within these broad ethnic groups (Kneale 2010). Successful management of asthma among UK children is associated, in part, with social position and socio‐economic status. For example, although South Asian children are at lower risk of asthma, they, along with black children, are at higher risk than white children of admission following asthma complications (Netuveli 2005). Indeed a systematic review of socio‐economic status and health outcomes found evidence to suggest that the risk of developing asthma is highest among children in the UK from lower‐income families (Spencer 2012). Overall, the UK government estimates that a billion pounds is spent annually through the National Health Service (NHS) on treatment and prevention of asthma among adults and children (Department of Health 2012). Thus population‐based interventions that improve asthma control have the potential to generate significant savings for the UK NHS.

Description of the intervention

Globally, a large proportion of people with asthma do not receive adequate self‐management education and training in primary care, and in England in 2014, more than a quarter of people (adults and children) living with asthma had not undergone an asthma review in the previous 15 months (HSCIC 2014). Moreover, inadequate knowledge of the condition and patient non‐adherence with clinician recommendations for asthma treatment (e.g. overuse of long‐acting beta₂‐agonists, under‐use of inhaled corticosteroids) may contribute towards poor asthma management among children (Piecoro 2001; Walsh 1999).

Children who experience an asthma exacerbation are at risk of hospitalisation and death (Bush 2017). Of the 65,000 hospitalisations for asthma occurring in 2011‐2012 in the UK, more than one‐third (38%) occurred in children (aged birth to 14 years); moreover, in an in‐depth study of asthma deaths, 14% (28 of 195) of confirmed deaths from asthma in the UK over a year occurred among children and young people 20 years of age and younger (Levy 2014). Effective self‐management of asthma could reduce levels of hospitalisation, which may reduce the financial implications of asthma and improve outcomes for children and adults with asthma, while reducing asthma‐related deaths in children.

Living with asthma can impact many other child health and social outcomes, and asthma, particularly severe asthma, is associated with a range of developmental, emotional, and behavioural problems (Blackman 2007). Some studies suggest that children with asthma are disadvantaged in terms of their peer relationships, and other studies report that some children with asthma are bullied (Harris 2017; Wildhaber 2012). Moreover, children with asthma are more likely to limit participation in activities as the result of dyspnoea and other asthma‐related symptoms (Van Den Bemt 2011).

Children with asthma tend to have poorer school attendance rates than their peers (Rodriguez 2013). For example, one US study reported that children living with asthma miss an average of 1.5 additional days of school annually compared with their peers, and that increased asthma severity was associated with an increase in the number of days absent from school (Moonie 2006). Furthermore, average school days missed masks large heterogeneity in experience, with some children missing many school days as a result of asthma. A school‐based survey, conducted by two members of the review team (KH, JG), assessed current levels of asthma control and school attendance in a sample of 766 children with asthma attending London secondary schools (Harris 2017). Overall, 20.9% of London school children self‐reported at least one school absence due to asthma over a four‐week period. Moreover, children with poor asthma control (n = 350) had greater rates of school absence compared to their peers with good asthma control (32.7% vs 10.9%) (Harris 2017). Fowler 1992 found that grade failure is more frequent among children with asthma.

Self‐management consists of educating and enabling children to achieve good control over their own asthma symptoms, thereby preventing future exacerbations (Kotses 2010);self‐management is viewed as a cornerstone of asthma treatment and care (Bateman 2008; BTS 2016; GINA 2018). Asthma control refers to the degree to which asthma symptoms can be observed and subsequently improved with treatment (GINA 2018). Well‐controlled asthma is associated with reduced daytime and night‐time symptoms, decreased long‐term morbidity, and diminished risk of life‐threatening asthma attacks (Juniper 2006). Asthma control tends to improve with age among children; one study reported excellent or satisfactory control in 38% of children four to six years of age and in 66% of children 13 to 16 years of age (Kuehni 2002).

For chronic respiratory diseases, self‐management is defined by the British Thoracic Society (BTS) as "the tasks that individuals must undertake to live with chronic conditions, including have the confidence to deal with medical management, role management and emotional management of their conditions" (BTS 2016). For asthma, successful self‐management skills include good inhaler technique and ability to recognise and respond to asthma symptoms. Self‐management also encourages an alliance between the physician or healthcare professional and the patient for the purpose of managing asthma (Kotses 2010). For the purposes of the present review, we have included only self‐management studies that provided education on asthma symptoms and their avoidance and management, omitting studies that provided education solely on the nature of asthma.

One main indirect cost of childhood asthma is absence from school, and costs of hospitalisation and of asthma medication drive most of the direct costs of this condition (Bahadori 2009). Although delivery of an asthma self‐management intervention in schools has the potential to reduce asthma burden, the effectiveness of this approach across various "proximal" (e.g. improvement in asthma symptoms), "intermediate" (e.g. healthcare usage), and "distal" outcomes (e.g. school achievement) remains unclear (Figure 1). Even when interventions are delivered in similar school settings, several factors can influence success, including variation in treatment settings, study populations, and ways in which school‐based asthma self‐management interventions and intervention components are delivered, in addition to the role of intervention mediators such as changes in school‐level policies around asthma or asthma medication (Al Aloola 2014).

Figure 1

Logic model of school‐based asthma interventions.

How the intervention might work

Self‐managment works by enabling patients to control their asthma symptoms, thereby preventing future exacerbations and improving their quality of life. Schools are a familiar environment for children's learning, and interventions provided at school have the potential to include large numbers of children with asthma at a single location (Ahmad 2011; Bruzzese 2009; Coffman 2009).

A previous systematic review of self‐management interventions delivered in clinic, home, and school environments for children with asthma found that these were positively associated with moderate improvements in lung function, school absenteeism, emergency visits to hospital, and self‐efficacy (Guevara 2003). A separate Cochrane Review reported that targeted self‐management interventions can lead to reduced hospital admissions among those at risk of hospitalisation (Boyd 2009). Participants included in both reviews were children from birth to 18 years of age with a diagnosis of asthma. Guevara 2003 excluded children with a pulmonary diagnosis other than asthma. Neither review noted participant comorbidities. Other reviews of self‐management interventions for children with asthma suggest that educational interventions delivered to children with asthma can be effective; however, these reviews have considered interventions delivered within schools alongside those delivered in other settings, including the clinic and the home (e.g. Smith 2005; Wolf 2002). Indeed Welsh 2011 points to lack of consensus around the optimal setting for asthma interventions. To date, only two systematic reviews have evaluated the evidence for interventions delivered exclusively within schools. These reviews reported a positive impact on school absenteeism but provided less conclusive evidence on the impact on health outcomes such as hospitalisations (Ahmad 2011; Coffman 2009). Notably, both reviews used a narrative approach to synthesis (Ahmad 2011; Coffman 2009). Another review examined outcomes for primary school age children only (Al Aloola 2014).

To date, few reviews have included analyses of accompanying "process‐level" measures, such as changes in school policy. Pinnock 2015 is one exception. These review authors explored how asthma self‐management interventions should be implemented, although they did not focus on school interventions alone. Nevertheless, based on analysis of two studies conducted in schools, they identified high school turnover and lack of parental involvement as challenges to implementation. Analysis of such process factors would further illuminate the modifiable components of interventions that may be most critical in determining the success (or failure) of interventions, and in mapping out the diverse processes undertaken as part of the intervention.

Systematic reviews of self‐management interventions in adults with asthma highlight the importance of gaining a deeper understanding of intervention characteristics and implementation processes. For example, Denford 2013 found that active involvement was associated with greater effect size, but that focus on stress management techniques was potentially counterproductive. Previous studies of self‐management in children have focused on child‐level moderators. Consequently, the effectiveness of different aspects of school‐based interventions for children with asthma is currently unclear.

Background to the methods used in this review

In this review, we aim to synthesise the evidence for school‐based interventions by addressing asthma self‐management, for the first time, using a mixed methods approach. Mixed methods involves synthesising qualitative and process evaluation evidence, as well as quantitative evidence, in an integrated way. Process evaluation studies explore the implementation, receipt, and setting of an intervention. Although "process" and "qualitative" are often mistakenly used interchangeably, data for process evaluation can be both quantitative and qualitative (Oakley 2006). Process evaluations can be used to develop mechanistic theories around how interventions work, although no universally agreed definition is available for what a process evaluation is and which core components it should include.

Investigators in one study defined a process evaluation as evaluating the quality of the intervention and measuring the disparity between the way in which an intervention was intended to be implemented and the way in which it is actually implemented (Shepherd 2010). This focus on evaluating the processes of delivery and the factors "responsible for successful outcomes, implementation of the intervention, and intervention integrity" is also shared elsewhere (Waters 2006). Meanwhile, UK Medical Research Council guidance on how to conduct process evaluations states that core components of process evaluations include (I) clear description (and evaluation) of implementation and processes of implementation; (ii) clear analysis of the mechanism of impact (participant responses to and interactions with the intervention); and (iii) clear description of context and analysis of how contextual factors affect mechanisms and implementation (Moore 2015).

Qualitative comparative analysis (QCA)

Although other reviews have set out to apply a mixed methods approach (albeit applied to other health topics) (Hurley 2013; Husk 2016), we sought to review the literature using both meta‐analyses of quantitative studies to assess the effectiveness of interventions and qualitative comparative analysis (QCA) to discern the importance of different configurations of intervention features. QCA has its basis in set‐theoretic logic, and is a well‐placed method for synthesising data from a small number of studies with complex characteristics. This approach aims to uncover the degree of overlap between a set of studies that are successful in their implementation and sets of studies that share different configurations of intervention characteristics. In pursuing the aim described above, we used a logic model to help structure and synthesise review findings (Figure 1), in accordance with the practices described in previous reviews (Glenton 2013).

Logic models

Logic models are tools that can be used to evaluate the effectiveness of a programme and/or to guide programme planning and implementation (NHS Scotland 2014). The protocol authors developed a logic model to outline some school‐based asthma self‐management intervention components that may be influential (Figure 1). We developed the logic model from the outcomes to be included in this review, and we worked backwards, theorising the causal chain necessary to lead to these outcomes. We developed the logic model using published literature and systematic reviews, including existing logic models used in studies and policy documents. Use of a logic model assisted us in identifying the types of data that may need to be captured if we are to gain an understanding of intervention components and implementation processes (Kneale 2015). The underlying idea behind a logic model is that a target or final goal is identified, and the pre‐conditions needed to reach this goal are hypothesised as different steps, building up a theorised chain of intervention actions and how they may impact outcomes. The logic model in Figure 1 shows the steps needed to reach the distal (long‐term) outcome of improvement in general health, well‐being, and educational outcomes among children with asthma; to achieve this long‐term outcome, we hypothesise that improvement in more intermediate outcomes such as episodes of healthcare usage and school absences is needed; to achieve improvement in these outcomes, we would expect improvement in asthma symptoms and lung function to be a necessary pre‐condition, and, in turn, to improve these, we theorise that children need better knowledge about asthma and improved skill in using inhalers. Changes in children's knowledge and skills follow from exposure to the intervention, although several modifiable intervention design characteristics may cause the intervention to have a differential impact, and may influence the characteristics of children themselves and the context in which the intervention takes place. Each intervention however includes various core elements (reflecting our definition of self‐management), as well as a set of resources and theories underlying its delivery. In addition, the logic model recognises that interventions can fail to effect change in children's outcomes because of issues of design or implementation, and a box on 'process metrics' incorporates ways of understanding the success of intervention implementation.

Why it is important to do this review

Educational impacts attributable to asthma are larger among children from lower socio‐economic groups and/or ethnic minority groups (Milton 2004), with children from ethnic minorities more likely than others to report asthma‐related hospitalisations (Netuveli 2005). Such differentials may, in part, reflect the failure of existing intervention models to deliver asthma self‐management training equitably to children across socio‐demographic groups. Given that the school environment offers a platform by which children from all socio‐economic backgrounds can receive the same asthma self‐management interventions, delivery of asthma self‐management interventions at this level could reduce inequalities in self‐management. Indeed, schools were previously identified as effective sites for the delivery of asthma self‐management interventions because the school environment is commonly associated with learning of new skills. Schools also provide access to large numbers of children with asthma, including those who do not have a general practitioner (GP) and those who do not regularly attend GP appointments. However, 'school age' (usually five to 18 years old) spans a wide spectrum of child development stages and consequently represents different teaching needs and various responses to self‐management interventions. Therefore, an understanding of the processes of implementation (and their success) is essential for the development of mechanistic theories of how and why interventions work that can be understood in the context of the child's characteristics.

In planning the current review, we placed strong emphasis on documenting and understanding the different processes that occur during school‐based asthma self‐management interventions. We envisaged that this approach would help us to understand the different mechanisms involved and would allow future trialists to evaluate the generalisability of processes and outcomes described and measured. The focus on delivery of interventions to help children self‐manage their own chronic condition is encouraged by advisory groups to UK policy‐makers. They view the integration of health and educational (and social care) services as critical in improving the quality of life of children with chronic conditions such as asthma, and in reducing differentials in outcomes such as school attendance (Lewis 2012). This systematic review draws on a mixed methods approach, looking at different sets of literature that evaluate intervention implementation and effectiveness, and using different methods to combine this literature. This approach will provide a rich account of school‐based asthma interventions by examining whether these interventions are effective in changing children's outcomes and by discerning how they effect change.

Objectives

This review has two primary objectives.

To identify the intervention features that are aligned with successful intervention implementation.
To assess effectiveness of school‐based interventions provided to improve asthma self‐management among children.

Methods

Criteria for considering studies for this review

Types of studies

We addressed our first objective (to identify intervention components and processes that are aligned with successful intervention implementation) by exploring process evaluation reports. We pursued the second objective (to assess the effectiveness of school‐based interventions for improvement of asthma self‐management) by examining outcome evaluation reports (i.e. randomised parallel‐group design involving individual or cluster randomisation).

Identifying the intervention components and processes aligned with intervention success in process evaluation studies

In this review, we identified process evaluations as involving systematic measurements to determine the extent to which a particular programme was implemented, in keeping with the guidance described above. Measures of implementation were focused on fidelity and specifically on attrition, adherence, and dosage. To capture the breadth of evidence about implementation, we identified a process evaluation study as (I) a study that was a self‐defined "process evaluation"; or (ii) a study that included the elements of a process evaluation as defined in a section of an outcome evaluation; or (iii) a study in which researchers integrated process evaluation data within an outcome evaluation but provided within the results measures around processes that were detailed and extractable. Studies not self‐identified as process evaluation studies must have contained (I) an assessment of core components (implementation, mechanisms, context); (ii) clear research questions guiding the process evaluation; and (iii) use of recognised evaluation methods (described by Moore 2015). We also included studies with a focus on the presence/development of school asthma policies (as represented in the logic model (Figure 1)); we expanded this to include studies measuring broader school‐level commitment. In this way, use of a logic model explicitly impacted study selection decisions (Kneale 2015).

Previous systematic reviews of process evaluation studies have tended to include only process evaluation studies linked to an outcome evaluation (e.g. Murta 2007). In this review, we have linked included process evaluation studies to randomised controlled trials (RCTs) assessing the effectiveness of the intervention; we have also included trials evaluating the implementation of a variety of study designs, provided they met our other inclusion criteria. This allowed us to use process evaluation data for theory development and testing within a mixed method framework.

Publication date and language

We imposed criteria around the date on which studies were published to help ensure that the content of self‐management interventions was broadly reflective of today's recommendations. Recommendations around the management of asthma in the UK were first developed in 1990 on the basis of articles that had appeared in British Medical Journal and Archives of Diseases in Childhood, from 1989 onwards (British Asthma Guidelines 1997); recommendations were developed in the USA around the same time (National Institute of Health 1997). Therefore, we excluded studies that pre‐dated the impetus around development of guidelines for the management of asthma, and we included only studies published from 1995 onwards (corresponding with publication of the first Global Initiative for Asthma (GINA) guidelines, which provided a foundation for asthma guidelines globally). We included only studies published in English.

Types of participants

We included school‐aged children and young people (five to 18 years old) with asthma. When the intervention included young people and adults (e.g. when provided in colleges with students 16 to 24 years of age), we intended to include these studies only if most participants were 18 years of age or younger (although we observed no such instance). We also included interventions if they incorporated some components that were delivered to peers, teachers, and/or parents and families, although only when they involved at least partial delivery of the intervention to school‐aged participants with asthma within school environments. We included studies reporting on interventions among children and young people with intermittent or mild to severe or persistent asthma.

We did not impose criteria regarding the types of schools that we included in our scope, as long as schools represented the physical location where intervention participants usually received most of their education.

Types of interventions

We included asthma self‐management interventions delivered at school. Eligible interventions aimed to develop and enhance self‐management of asthma among children by achieving the following.

Increasing knowledge of asthma self‐management.
Enhancing self‐management skills.
Improving self‐management behaviours and practice.

Eligible interventions must have included the active transfer of information around at least one of the aspects of asthma self‐management outlined below. However, we recognise that for asthma self‐management to be effective, a combination of these must be incorporated into the interventions.

Reinforcement of regular monitoring of lung function.
Emphasis on the importance of self‐management practice and behaviour.
Development of a partnership/alliance between patient and primary care/healthcare practitioners (including school nursing staff) for the management of asthma.
Instruction on inhaler techniques.
Reinforcement/provision of an individualised written asthma management plan.
Emphasis on the importance and appropriate use of reliever therapies such as beta₂‐agonists (BTS 2016).
Emphasis on the importance and appropriate use of regular preventer therapies such as inhaled corticosteroids and combination inhaled corticosteroid and long‐acting beta₂‐agonist therapies (BTS 2016).
Non‐pharmacological self‐management strategies focused on avoiding or reducing the risk of experiencing asthma or asthma attacks, including lifestyle and behavioural modifications (as set out in BTS 2016).

Interventions that focused only on treating children's asthma in schools, and not on enhancing self‐management skills, were not eligible. For example, interventions that provided directly observed therapy but did not seek to actively improve children's self‐management skills inside and outside school were not eligible for inclusion. This included studies in which we determined that most of the self‐management component of the intervention had not occurred in the school environment. This led to the omission of some studies that otherwise met the inclusion criteria and have been included in previous reviews (e.g. Halterman 2011; Halterman 2012).

Interventions may focus on improving the climate for asthma self‐management within schools, for example, by changing school policies around the way that teaching staff may assist in asthma self‐management. However, studies that did not also include the development and evaluation of asthma self‐management skills and behaviours among children were not eligible. We included self‐management interventions if they fit the definition given in the guidelines produced by the British Thoracic Society/Scottish Intercollegiate Guidelines Network, or in the GINA guidelines (BTS 2016; GINA 2018), as described in the Background section. We excluded studies that concentrated on breathing exercise methods (including yoga interventions) if they did not directly focus on the other aspects of self‐management listed above.

The intervention could be provided by a trained educator, nurse (including school, practice, or community nurse), doctor or physician, peer, or social worker, and most delivery or access must have been provided on the premises of the school attended by the children. Interventions for which the school setting was not involved in delivery were not eligible for inclusion.

Comparisons

For outcome evaluation studies, comparison groups were restricted to usual care or to a self‐management or health intervention with a focus other than asthma (placebo).

For process evaluation studies, a comparison group could have received another asthma intervention, or the study may not have included a comparison group at all; all process evaluation studies must have included other parameters as described above in terms of study population, study setting, and contents of the asthma intervention.

Types of outcome measures

Outcomes for meta‐analyses

Our primary outcomes were based on those identified as indicators of good asthma control (BTS 2016), represented as intermediate outcomes in Figure 1. We were also interested in several secondary outcomes (represented as proximal and intermediate outcomes in Figure 1, as well as a measure of acceptability/implementation in withdrawal from the intervention).

Primary outcomes

Asthma symptoms or exacerbations leading to admission to hospital (children with one or more admissions or high admission rates)
Asthma symptoms or exacerbations leading to emergency department visits
Parent‐reported absence from school
Days of restricted activity

Secondary outcomes

Unplanned visit to hospital or GP due to asthma symptoms
Experience of daytime and night‐time symptoms (*these were differentiated from 'any' symptomatology by stating that symptoms occurred either in the daytime or at night‐time)
Lung function (e.g. forced expiratory volume in one second (FEV₁) in clinic, peak flow at home)
Use of reliever therapies such as beta₂‐agonists
Corticosteroid dosage and/or use of add‐on therapies (e.g. long‐acting beta₂‐agonists (LABAs), leukotriene receptor antagonists (LTRAs))
Health‐related quality of life (HRQoL) as measured by a validated questionnaire
Withdrawal from the study

We extracted data for all points at which the outcomes above were measured and pooled data as appropriate.

Outcomes for qualitative comparative analysis (QCA): defining a successful intervention

Qualitative comparative analysis (QCA) as used in this review and described in further detail below, is a method of evidence synthesis that enables understanding of which configurations of intervention components and processes trigger successful outcomes. QCA is predicated upon set theory, and in this context essentially involves exploring the degree of overlap between a set of successfully implemented studies and a set of studies with a particular range of intervention components and processes.

A first step in our use of QCA was deciding how 'successful' implementation could be identified. Currently, no approach has been established for categorising the implementation of an intervention as 'successful' or 'not successful' (Schellenberg 2012). We began by examining aspects of intervention implementation that were related to intervention fidelity as well as evidence around attrition, dosage, and adherence. A literature review of implementation scoring methods for public health interventions ‐ Schellenberg 2012 ‐ included one study that examined the implementation of a complex intervention that included a school component (Rosecrans 2008). Study authors used the following criteria: "process indictors for which standards were set, such as fidelity (e.g. % of minimum foods stocked) or dose received (e.g. % of family pack cards completed and returned), were assigned to categories of implementation as follows: low (0–49%), moderate (50–74%) or high (75–100%)" (Rosecrans 2008; p75). This 75% threshold also corresponds with the 25% attrition rate that is often incorporated within study sample size calculations for public health trials involving children (Berry 2013; Bruzzese 2011; Clark 1986).

A 75% threshold formed the basis of our coding scheme for the outcome, by which 75% was used as a cross‐over point for a 'high' or 'successful' implementation score. Implementation reflected reports of attrition, dosage, and adherence. For each of these indicators, we set values by using a blend of direct and transformational assignment (see Table 1), whereby we assigned values to qualitative data and then calibrated all data using transformational assignment. This blended approach was necessary to combine qualitative and quantitative data. To derive an outcome variable that reflected intervention implementation more holistically, we aggregated the three separate indicators into a single outcome variable by adding each separate value and calibrating the summed score. This outcome value reflected the mainstay of the analyses and distinguished our successfully implemented intervention set.

Table 1. Detailed coding framework for conditions and outcomes

	Field	Instructions for extractors	Coding values and method
	Setting and participants
1	Number of children	Recorded total number of children involved in intervention	Transformational assignment implemented to condition, reflecting whether it was a ‘large intervention’. Interventions with 15 or fewer children = 0; interventions with 90 children = 0.5; interventions with 300 or more children = 1. Other values fell between 0 and 1
2	Multiple settings	Evidence if delivered at more than 1 school	Direct assignment: yes (mentioned) = 1; no evidence = 0
3	Single sex school	Evidence if delivered at a single sex school	Direct assignment: yes (mentioned) = 1; no evidence = 0
4	Type of school	High school; primary/elementary; junior/middle; other Variable transformed to reflect whether the intervention took place at a high school	Direct assignment: high school = 1; middle/junior = 0.66; elementary/primary = 0.33; missing = 0.5; mixture of high schools and middle schools = 0.75
5	Ethnicity of children	Whether minority ethnic children were targeted/represented. Actual proportions recorded where possible	Transformational assignment Interventions with 25% or fewer children from ethnic minority = 0; interventions with 33.3% of children from ethnic minority = 0.5; interventions with 50% or more children from ethnic minority = 1 When value is missing (and no qualitative statement supports assumption of targeting), assume that this is ‘probably not’ – i.e. probably not targeted – input value of 0.25
6	Socio‐economic status of children	Whether children from lower socio‐economic groups were targeted/represented Actual proportions recorded where possible Indicators included parents with low levels of education; low household income; receipt of free school meals	Transformational assignment Interventions with 25% or fewer children from low socio‐economic groups = 0; interventions with 33.3% of children from low socio‐economic groups = 0.5; interventions with 50% or more children from low socio‐economic groups = 1 Where value is missing (and no qualitative statement supports assumption of targeting), assume that this is ‘probably not’ – i.e. probably not targeted – input value of 0.25
7	Child age	Age groups/classes targeted: ages 5 to 10	Direct assignment: yes (mentioned) = 1; no evidence = 0
8		Age groups/classes targeted: ages 11 to 14	Direct assignment: yes (mentioned) = 1; no evidence = 0
9		Age groups/classes targeted: ages 15 to 18	Direct assignment: yes (mentioned) = 1; no evidence = 0
10	Direct recipients	Children directed recipients	Direct assignment: yes (mentioned) = 1; no evidence = 0
11		Teachers directed recipients	Direct assignment: yes (mentioned) = 1; no evidence = 0
12		Parents directed recipients	Direct assignment: yes (mentioned) = 1; no evidence = 0
13		School nurses directed recipients	Direct assignment: yes (mentioned) = 1; no evidence = 0
	Programme design
14	Theory driven	Did the study name a theoretical framework that underpins the intervention design or delivery style?	Direct assignment: yes (mentioned) = 1; no evidence = 0
15	Intensity of the programme	Coded initially as follows: high intensity = 6+ sessions (group and individual); medium intensity = 3 to 5 sessions; low intensity/no evidence of med/high = 1 to 2 sessions; unclear. Variable transformed to reflect whether the intervention was of high intensity	Direct assignment: high intensity = 1, medium intensity = 0.66; low intensity = 0.33. When no evidence on intensity of intervention was included (1 study = (Richmond 2011)), this was coded as 0.33 (no evidence of high intensity) – interpreted as no evidence of high intensity; for Splett (Splett 2006), such is the degree of personalisation/tailoring that 0.5 was selected as the intensity – each individual session was personalised and lengthy
16	Personalisation/tailoring	Did the programme include individual sessions or use personalisation in any way to alter curriculum to individual students’ needs?	Direct assignment: yes, all sessions implemented were personalised/tailored = 1; some sessions were personalised/tailored = 0.66; personalisation/tailored sessions account for only a minor component = 0.5; no evidence, only generic group sessions implemented = 0 Note that this was personalised by or individual sessions were held with an instructor (included guided online sessions); self‐study components including homework were not included here
17	Timing of the intervention	Did the intervention interfere with the child’s own time (during lunch or after school)?	Direct assignment: yes, all sessions did = 1; yes, but not all sessions = 0.75; missing data = 0.5; described as not interfering with child’s own time = 0
18		Did the intervention interfere with the child’s lessons/other education?	Direct assignment: yes, all sessions did = 1; yes, but not all sessions = 0.75; missing data = 0.5; described as not interfering with child’s lessons/other education = 0
19	Information about control condition	Described whether trialists were also providing a control for the main intervention (intended to capture complexity of running an intervention and a control)	Direct assignment: yes, an equivalent control = 1; yes, but not an equivalent = 0.66; no control described = 0
20	Instructor or facilitator	Teacher	Direct assignment: yes, main instructor = 1; secondary instructor or facilitator = 0.66; not mentioned as an instructor/facilitator = 0
21		Peer	Direct assignment: yes, main instructor = 1; secondary instructor or facilitator = 0.66; not mentioned as an instructor/facilitator = 0
22		School nurse	Direct assignment: yes, main instructor = 1; secondary instructor or facilitator = 0.66; not mentioned as an instructor/facilitator = 0
23		Self‐directed/child‐directed	Direct assignment: yes, main instructor = 1; secondary instructor or facilitator = 0.66; not mentioned as an instructor/facilitator = 0
24		Parent	Direct assignment: yes, main instructor = 1; secondary instructor or facilitator = 0.66; not mentioned as an instructor/facilitator = 0
25		Other	Direct assignment: yes, main instructor = 1; secondary instructor or facilitator = 0.66; not mentioned as an instructor/facilitator = 0
	Programme content
26	Curriculum	Lung physiology/asthma biology	Direct assignment: yes (mentioned) = 1; no evidence = 0
27		Asthma acceptance/asthma into identity	Direct assignment: yes (mentioned) = 1; no evidence = 0
28		Symptom monitoring and correct medication use	Direct assignment: yes (mentioned) = 1; no evidence = 0
30		Avoiding triggers	Direct assignment: yes (mentioned) = 1; no evidence = 0
31		General health including exercise	Direct assignment: yes (mentioned) = 1; no evidence = 0
32		Strengthening alliances including asthma action plans with primary care providers	Direct assignment: yes (mentioned) = 1; no evidence = 0
33		Specific focus on smoking	Direct assignment: yes (mentioned) = 1; no evidence = 0
34		Personalised/tailored (individualised)	Direct assignment: yes (mentioned) = 1; no evidence = 0
35		School performance	Direct assignment: yes (mentioned) = 1; no evidence = 0
36		Emergencies	Direct assignment: yes (mentioned) = 1; no evidence = 0
37		Unknown	Direct assignment: yes (mentioned) = 1; no evidence = 0
38		Specific focus on breathing/relaxation techniques	Direct assignment: yes (mentioned) = 1; no evidence = 0
39	Learning styles	Problem‐solving component	Direct assignment: yes (mentioned) = 1; no evidence = 0
40		Self‐directed (including homework) component	Direct assignment: yes (mentioned) = 1; no evidence = 0
41		Peer delivery component	Direct assignment: yes (mentioned) = 1; no evidence = 0
42		Interactive (non‐didactic) components	Direct assignment: yes (mentioned) = 1; no evidence = 0
43		Didactic components	Direct assignment: yes (mentioned) = 1; no evidence = 0
44		Other style/unclear	Direct assignment: yes (mentioned) = 1; no evidence = 0
45	Programme ethos/aims	Emphasis on social benefit	Direct assignment: yes (mentioned) = 1; no evidence = 0
46		Emphasis on improving well‐being	Direct assignment: yes (mentioned) = 1; no evidence = 0
47		Emphasis on having fun	Direct assignment: yes (mentioned) = 1; no evidence = 0
48		Emphasis on fostering independence/personal responsibility	Direct assignment: yes (mentioned) = 1; no evidence = 0
49		Emphasis on developing children's knowledge	Direct assignment: yes (mentioned) = 1; no evidence = 0
50		Emphasis on collaboration	Direct assignment: yes (mentioned) = 1; no evidence = 0
51		Emphasis on tailoring for specific group needs	Direct assignment: yes (mentioned) = 1; no evidence = 0
52		Emphasis on breathing technique	Direct assignment: yes (mentioned) = 1; no evidence = 0
53		Unclear	Direct assignment: yes (mentioned) = 1; no evidence = 0
54	Additional components – school asthma policy	Additional support provided for developing school policy	Direct assignment: yes (mentioned) = 1; no evidence = 0
55		School asthma policy developed organically	Direct assignment: yes (mentioned) = 1; no evidence = 0
	Additional processes undertaken – planned and unplanned
56	Recruitment methods ‐ school	Ad hoc/convenience sample of schools	Direct assignment: yes (mentioned) = 1; no evidence = 0
57		Census of school district (all schools invited and potentially eligible)	Direct assignment: yes (mentioned) = 1; no evidence = 0
58		Unspecified methods of school recruitment	Direct assignment: yes (mentioned) = 1; no evidence = 0
59	Additional processes to improve/attenuate attrition/enrolment	Marketing materials sent to parents	Direct assignment: yes (mentioned) = 1; no evidence = 0
60		Low motivation of students acknowledged and addressed	Direct assignment: yes (mentioned) = 1; no evidence = 0 Note that 1 study received a value of 0.75, as low motivation was acknowledged but was not explicitly described as being addressed (Magzamen 2008)
61		Incentives used (child or parent)	Direct assignment: yes (mentioned) = 1; no evidence = 0 Incentives for teachers and no evidence for children/teachers coded as 0.5
62		Make‐up/catch‐up sessions provided	Direct assignment: yes (mentioned) = 1; no evidence = 0
63		Reminders sent to parents/children	Direct assignment: yes (mentioned) = 1; no evidence = 0
64	Relationships/engagement	Did teachers engage or participate in the way they were expected to?	Direct assignment: yes, good reported throughout = 1; yes, some weaker evidence of good relationships or evidence that relationships improved during the course of the intervention = 0.75; missing, not applicable, or undetermined = 0.5; no, some weaker evidence of poorer relationships or evidence that relationships deteriorated during the course of the intervention = 0.25; evidence of poor relationships throughout = 0
65		Did parents engage or participate in the way they were expected to?	Direct assignment: yes, good reported throughout = 1; yes, some weaker evidence of good relationships or evidence that relationships improved during the course of the intervention = 0.75; missing, not applicable, or undetermined = 0.5; no, some weaker evidence of poorer relationships or evidence that relationships deteriorated during the course of the intervention = 0.25; evidence of poor relationships throughout = 0 One study described good levels of engagement, but review authors assigned value of 0.25 as a third of parents did not engage as expected (Kintner 2012); similar rationale for Mujuru 2011
66		Did school nurses engage or participate in the way they were expected to?	Direct assignment: yes, good reported throughout = 1; yes, some weaker evidence of good relationships or evidence that relationships improved during the course of the intervention = 0.75; missing, not applicable, or undetermined = 0.5; no, some weaker evidence of poorer relationships or evidence that relationships deteriorated during the course of the intervention = 0.25; evidence of poor relationships throughout = 0
67		Did other relevant stakeholders engage or participate in the way they were expected to?	Direct assignment: yes, good reported throughout = 1; yes, some weaker evidence of good relationships or evidence that relationships improved during the course of the intervention = 0.75; missing, not applicable, or undetermined = 0.5; no, some weaker evidence of poorer relationships or evidence that relationships deteriorated during the course of the intervention = 0.25; evidence of poor relationships throughout = 0
	Process outcomes
68	Child satisfaction	Put in level of satisfaction (%) or record qualitative statement on child satisfaction with the intervention experience. Indicators of satisfaction include children reporting that they enjoyed the intervention; whether the children would recommend the intervention to others; whether children found the intervention helpful. Knowledge development should not be included here	Elements of direct and transformational assignment included here [First] Direct assignment: where there is a qualitative statement indicating positive agreement, assign value of 0.66; where a qualitative statement indicating negative agreement, assign value of 0.33; where no child satisfaction data were collected or data were missing, assign value of 0.5 [Second; including of direct above] Transformational assignment implemented to condition reflecting whether children were satisfied. Interventions with 25% or fewer children satisfied = 0; interventions with 50% of children satisfied = 0.5; missing data coded as 0.5; interventions with 75% or more children satisfied See text for further justification on use of the 75% threshold
69	Child attrition (overall level)	Put in level of completion (%) or record qualitative statement on child completion rate	Elements of direct and transformational assignment here. Note thresholds were higher than for satisfaction, as fewer data were missing [First] Direct assignment: where there is a qualitative statement indicating high level of completion, assign value of 0.83; where a qualitative statement indicating problematic completion, assign value of 0.66. Where data are missing, assign value of 0.75 [Second; including of direct above] Transformational assignment implemented to condition reflecting level of completion. Interventions with 66% or fewer children completing the intervention = 0; interventions with 75% of children completing the intervention = 0.5; interventions with 83% or more children completing the intervention = 1. Missing data coded as 0.5 See text for further justification on the use of thresholds
70	Child dosage level	Did the children receive the intended dosage of the intervention? Put in level of dosage (%) or record qualitative statement on child dosage.	Elements of direct and transformational assignment here. Note thresholds are higher than for satisfaction, as fewer data are missing [First] Direct assignment: where there is a qualitative statement indicating high level of dosage, assign value of 0.83; where a qualitative statement indicating problematic dosage, assign value of 0.66. Where data are missing, assign value of 0.75 [Second; including of direct above] Transformational assignment implemented to condition reflecting level of dosage. Interventions with 66% or fewer children receiving the full dosage = 0; interventions with 75% of children receiving the full dosage = 0.5; interventions with 83% or more of children receiving the full dosage = 1. Missing data coded as 0.5 See text for further justification on the use of thresholds
71	Child adherence	Did the children adhere to the intervention instructions, e.g. students being compliant with paperwork; completing homework; going to visit PCPs as instructed, etc. Put in level of adherence (%) or record qualitative statement on child dosage	Elements of direct and transformational assignment here. Note thresholds are higher than for satisfaction as fewer data are missing [First] Direct assignment: where there is a qualitative statement indicating high level of adherence, assign value of 0.83; where a qualitative statement indicating problematic adherence, assign value of 0.66. Where data are missing, assign value of 0.75 [Second; including of direct above] Transformational assignment implemented to condition reflecting level of adherence. Interventions with 66% or fewer children adherent = 0; interventions with 75% of children adherent = 0.5; interventions with 83% or more children adherent = 1. Missing data coded as 0.5 See text for further justification on the use of thresholds
72	Consolidated process variable	Summation of attrition, adherence, and dosage scores as a marker of implementation success	Transformational assignment Score of 0 = 0 implementation not successful; score of 1.5 = mid point between successful and unsuccessful implementation; score of 3 = full implementation success

Search methods for identification of studies

Electronic searches

We searched the Cochrane Airways Group Specialised Register (see Appendix 1) for trials, using the strategy presented in Appendix 2, which was developed by the Cochrane Airways Information Specialist (Liz Stovold). We conducted searches in April 2015 and updated them in April 2016. We conducted further searches on 25 August 2017.

We searched the databases below for process evaluations for our qualitative comparative analyses, using the search criteria identified in Appendix 1, although we modified these criteria to account for the different search syntax/parameters used in additional databases (see Appendix 3,Appendix 4,Appendix 5,Appendix 6, and Appendix 7 for example search strategies).

Database of Promoting Health Effectiveness Reviews (DoPHER).
Cochrane Database of Systematic Reviews (CDSR).
Database of Abstracts of Reviews of Effects (DARE).
The Campbell Library.
National Institute for Health Research (NIHR) Health Technology Assessment (HTA) Programme website/journals library.
Health Technology Assessment (HTA) database.

We applied search strategies to a comprehensive search of the following clinical, public health, psychology, and social care databases from 1995 to the present*.

Allied and Complementary Medicine Database (AMED)
Applied Social Sciences Index and Abstracts (ASSIA).
Bibliomap (EPPI‐Centre Database of Health Promotion Research).
ClinicalTrials.gov
Cochrane Database of Systematic Reviews (CDSR).
Cochrane Central Register of Controlled Trials (CENTRAL).
Cumulative Index to Nursing and Allied Health Literature (CINAHL).
Excerpta Medica dataBASE (EMBASE)
Health Management Information Consortium (HMIC).
International Bibliography of the Social Sciences (IBSS).
National Health Service Economic Evaluation Database (NHS EED).
PsychInfo.
PubMed.
Sociological Abstracts (SocAbs).
Social Policy and Practice (SPP).
Social Services Abstracts
Web of Knowledge.

*MEDLINE, the Cumulative Index to Nursing and Allied Health Literature (CINAHL), Embase, the Allied and Complementary Medicine Database (AMED), and PsycINFO are included within the Cochrane Airways Group Specialised Register search.

Searching other resources

We handsearched Google Scholar, Social Policy Digest (for content up to 2014), and other sources such as the British Thoracic Society and Asthma UK for further studies.

We initially identified integral process evaluations (sibling studies) through backwards and forwards citation searches. As expected, we identified multiple process evaluations for some intervention studies; our strategy also allowed for inclusion of process evaluations without linkage to a trial included for quantitative analyses.

Data collection and analysis

Selection of studies

We piloted criteria for title and abstract screening on a random subset of studies for which the review authors who were involved in screening (DK, KH) took part in moderation exercises; we resolved disagreements by discussion and developed a shared understanding of the inclusion criteria. We achieved an agreement rate exceeding 90% in three consecutive samples before we proceeded to independent screening (DK, KH). We also employed priority screening (text mining) for independent title and abstract screening (Thomas 2011), after achieving a sufficiently high agreement rate, to locate likely included studies more quickly. However, both review authors (DK, KH) screened all abstracts.

We applied inclusion criteria successively to titles and abstracts, and to full reports. We obtained full‐text reports when studies appeared to meet the criteria for title and abstract, or when information was insufficient for a decision. For outcome evaluation studies, screening criteria covered populations (children five to 18 years of age), disease status (asthma), interventions (school‐based and focused on self‐management), comparators (usual care or placebo), study design (randomised controlled trials or cluster randomised controlled trials), date (publication year after 1995), and language (English language). We entered full‐text reports into EPPI‐Reviewer and reapplied the inclusion criteria (Thomas 2010); we included studies that met these study design criteria (irrespective of the actual outcomes collected). We developed a similar set of inclusion criteria for process evaluation studies covering populations, disease status, interventions, date, and language; additional criteria stipulated that studies must include the core components expected within a process evaluation and must use structured or recognised tools to collect data.

Data extraction and management

Data management

We uploaded records identified by searches to the specialist systematic review software EPPI‐Reviewer 4 for duplicate stripping and screening (Thomas 2010). This software recorded the bibliographic details of each study considered in the review, the origins of all studies (including search strings), and reasons for their inclusion or exclusion. We first extracted all data into EPPI‐Reviewer 4 and later exported them, as appropriate, into other software for synthesis (RevMan 2014; StataCorp 2013; Thiem 2013).

Extraction and management of data from process evaluation studies

Process evaluation measures ‐ data selection

Overall approach

The primary aim of exploring process evaluations using QCA was to identify the combinations of components and processes undertaken for interventions that were associated with successful intervention implementation. QCA is based on set theory, and, in this review, we explored the extent of overlap between a set of studies with successful implementation (our process outcome) and sets of studies that share combinations of different intervention components and processes. We presented extracted intervention components and processes (equivalent to antecedents and referred to as conditions from hereon in, in line with QCA terminology) as modifiable design characteristics in the logic model (Figure 1).

Extracting data and building the data table: initial data reduction and assignment of values

Two review authors (DK, KH) independently extracted the conditions (process evaluation measures) of interest from eligible studies. We developed an extensive data table of information supporting over 90 conditions for each study. These data represented quantitative indicators (showing the level of presence of a condition (e.g. the proportion of children from an ethnic minority recruited into an intervention)); binary indicators (representing whether or not a condition was present (e.g. study authors reported that the asthma curriculum contained information on lung physiology)); or qualitative statements (e.g. when study authors published quotes illustrative of child satisfaction with the intervention). In accordance with guidance provided by Rihoux and De Meur (Rihoux 2009), we developed a set of rules for assigning values to conditions (Table 1); these rules reflect a mixture of direct and transformational assignment (we have provided further explanation and an example in Appendix 8).

Reduction of data on conditions

We extracted more data than any QCA model could support ‐ a problem referred to as 'limited diversity in QCA terminology'. Recognising that many of the conditions extracted were binary indicators of constructs related to the same underlying condition, we implemented cluster analyses of linked items (e.g. elements of the curriculum) to create natural groupings and to reduce the number of conditions included in some models (Thomas 2014). We have displayed original and reduced data for these conditions in Table 2. In addition, we used the logic model presented in Figure 1 to guide our analysis, to rationalise and prioritise the conditions entered into models, and to limit the number of conceptually similar conditions that were entered into models.

Table 2. Original and reduced conditions for curriculum content, delivery style, and programme emphasis

Curriculum – original conditions	Curriculum – reduced conditions^a
I. Lung physiology ii. Asthma acceptance iii. Symptom monitoring and treatment iv. Trigger avoidance v. General health vi. Forming alliances vii. Smoking viii. Tailored/personalised ix. School performance x. Emergencies xi. Unknown content	I. Symptom monitoring and alliances ii. Lung physiology and general health iii. Symptom monitoring and trigger avoidance iv. Other various foci v. Unknown
Pedagogical delivery style – original conditions	Pedagogical delivery style – reduced conditions^b
I. Problem‐solving ii. Self‐direct iii. Peer delivery iv. Interactive v. Didactic vi. No information/other focus	I. Interactive focused style ii. Diverse style iii. Unknown style
Intervention emphasis – original conditions	Intervention emphasis – reduced conditions^c
I. Emphasis on social benefit ii. Emphasis on well‐being iii. Emphasis on having fun iv. Emphasis on personal responsibility v. Emphasis on children’s knowledge vi. Emphasis on collaboration vii. Emphasis on tailoring/personalisation viii. Emphasis unclear	I. Emphasis on tailoring/personalisation ii. Emphasis on personal responsibility iii. Diffuse emphasis/other

^aPseudo‐F index = 5.66.
^bPseudo‐F index = 8.36.
^cPseudo‐F index = 6.50.

Reduction of cases

Although cluster analysis reduced the number of conditions examined, we made the decision to focus on cases (studies) that were coded as providing high‐ or medium‐intensity interventions. We did not explicitly mention this in the protocol (therefore it is reported as a deviation), although this approach is congruent with indicators such as attrition and dosage.

Extraction and management of data from outcome evaluation studies (RCTs)

Outcome measures ‐ data extraction

Two review authors (DK, KH) independently extracted study characteristics and numerical outcome data from studies meeting the eligibility criteria of the review. In agreement meetings, review authors resolved discrepancies by discussion; we encountered no disagreements that needed resolution through arbitration by senior members of the review team. When we encountered missing data, we recorded these instances and contacted study authors for further information.

Assessment of risk of bias in included studies

Assessment of risk of bias in included RCTs

We assessed how the following sources of bias may affect the results of an individual study.

Sequence generation: we deemed that studies that used a computer‐generated allocation procedure, a random number table, or other recognised low‐risk means were at low risk of bias (as advised by the Cochrane tool for assessing risk of bias). We deemed that studies that used items such as clinic visit date or date of birth when the order of treatment group assignment was predictable or open to external influence were at high risk of bias. We described studies for which we were unable to ascertain methods of randomisation and allocation as having unclear risk of bias. Given the potential impact of socio‐economic imbalance between cluster sites within the same study, we also considered whether study authors had stratified socio‐economic variables.
Allocation concealment: we deemed that studies for which researchers took measures to prevent disclosure of treatment group assignment, such as off‐site allocation or allocation by a third party not involved in the study, were at low risk of bias. For cluster randomised studies, an additional consideration was timing of recruitment into the study in relation to assignment.
Blinding (performance bias and detection bias): we deemed that studies for which investigators took measures to ensure that personnel collecting data were unaware of participants' treatment group assignment were at low risk of bias. However, given the nature of the intervention and the difficulty involved in blinding recipients, a degree of performance bias may have impacted some outcomes, particularly patient‐reported outcomes, and this was unavoidable.
Handling of missing data and attrition: we deemed that studies for which data sets were complete, or for which reasons for missing data were not related to treatment, were at low risk of bias. When attrition rates were particularly high or imbalanced and unexplained, and only an available case set was presented, we deemed that the study was at high risk of bias. We deemed that studies for which study authors did not report the attrition rate separately for treatment and control groups, and for which we were unable to determine satisfactorily the reasons for withdrawal, were at high risk of bias.
Selective reporting: we restricted assessments of selective reporting to examination of available data related to outcomes included in the 'Summary of findings' table.
Other bias: we examined baseline imbalances in the characteristics of participants (see also the first point around stratification) for potential bias. We also looked for evidence of contamination between intervention and control groups. We restricted sensitivity analysis to primary outcomes of the review, and we derived overall judgements for each study at the outcome level.

Assessment of risk of bias in included process evaluation studies

We assessed the quality of process evaluation studies using elements of two tools. The first tool was developed at the EPPI‐Centre to assess the methodological rigour of 'views' studies that aimed to collect information on people's experiences during trials (Harden 2004). This tool considers seven criteria, including (I) whether the study includes an explicit theoretical framework and/or literature review; (ii) clearly stated aims and objectives; (iii) a clear description of context; (iv) a clear description of the sample and how it was recruited; (v) a clear description of methods used to collect and analyse data; (vi) attempts made to establish the reliability or validity of data analysis; and (vii) inclusion of sufficient original data to mediate between evidence and interpretation. The second tool, which was developed by the EPPI‐Centre to assess the quality of process evaluation data (O'Mara‐Eves 2013), assesses (I) methods of data collection; (ii) process evaluation participants as described; (iii) timing of the process evaluation with respect to the intervention; (iv) process evaluation data collection methods; (v) process evaluation data analysis methods; (vi) whether findings were supported by data; (vii) breadth and depth of findings; (viii) the extent to which the process evaluation gave privilege to the views of participants; (ix) reliability of findings; and (x) usefulness of process evaluation. As some of these domains overlap, we combined elements from both tools to assess the quality of process measures. This strategy also covers the main domains that had been set out in the Cochrane Qualitative Methods Group guidance that was current at the time (Hannes 2011).

Assessment of bias in conducting the systematic review

We conducted the review according to the published protocol (Harris 2015), and we reported deviations from it under Differences between protocol and review.

Measures of treatment effect

Continuous data

We planned to calculate mean differences (MDs) when continuous data were measured by the same scale or unit; however, this did not occur for most outcomes (one MD model had been constructed to explore quality of life as an outcome). Instead, when similar outcomes were measured by different scales or units, we used standardised mean differences (SMDs) (Hedges' (adjusted) g).

Dichotomous data

For dichotomous data, we calculated odds ratios (ORs), and, when appropriate, we combined results from different trials.

Ordinal data

We planned to analyse ordinal outcomes (such as quality of life scales) as continuous variables; when appropriate thresholds were identified, we analysed these as dichotomous variables.

Count data

We planned to calculate rate ratios for any count data that we encountered that represented the ratio of events experienced between two groups, such as episodes of hospitalisation or absences from school.

Unit of analysis issues

Cluster randomised studies

We included cluster randomised controlled trials in which schools or classes within schools rather than individuals with asthma were the unit of allocation. As variation in response to treatment between clusters may also be influenced by cluster membership, meaning that cluster members' data can no longer be considered independent of one another, we extracted data when study authors had undertaken analysis that properly adjusted for a clustered design. When study authors provided no intracluster correlation coefficient (ICC), we intended to estimate the ICC and the design effect according to methods recommended in Chapter 16 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). No study provided a direct estimate of ICC; however, we selected 0.05 based on the ICC estimate used in one of the included papers to calculate the sample size (McCann 2006). We adjusted effect estimates using methods described in Higgins 2011.

Choice of measurement point

For trials that reported outcomes at multiple time points, such as at post‐test with longer follow‐up, we extracted all data and combined in meta‐analyses the follow‐up points most consistently reported among trials.

Dealing with missing data

When study characteristics and numerical outcome data were missing from studies, we contacted study authors to request missing information. For quantitative aspects of process evaluations, such as satisfaction or participation data, we applied the same procedure. Recording of the 'missingness' of qualitative data in the process evaluations that we include is more oblique, although we recorded instances in which investigators indicated that the data collected were not reported upon as part of the quality assessment.

Assessment of heterogeneity

We assessed statistical heterogeneity by using the I² measure (Higgins 2003). We explored possible sources of variation by conducting prespecified sensitivity and subgroup analyses and performing meta‐regression analyses. These included those set out in the protocol (Harris 2015), as well those that we developed from QCAs.

We intended to construct random‐effects multi‐variate meta‐regression models using STATA, which would allow us to model the impact of different covariates simultaneously after first exploring the impact of these potential effect size study‐level moderators in bivariate models. However, a relatively small number of studies (our largest meta‐analysis model included 13 studies) meant that we were unable to extend the modelling in this way without compromising the underlying assumptions.

Assessment of reporting biases

We recorded the number of studies for which we were not able to ascertain the analysis of data related to our primary outcomes. We also recorded the number of studies for which we were not able to extract process measures, and we assessed the breadth and depth of those studies in terms of information on processes. We selected all process evaluation studies conditionally on addressing process‐related research questions, although the core process outcomes included within these did not always match our own selected process outcomes (e.g. some studies addressed different recruitment techniques as a central process of interest, although this focus did not match our own focus).

We plotted the distribution of effect sizes for each (outcome) study against study standard errors as a funnel plot for primary outcomes and based our assessment of publication bias on visual inspection (if 10 or more studies contributed to the outcome); we also undertook formal tests for small‐study publication bias using Egger's test (Harbord 2009).

Data synthesis

Data synthesis ‐ adopting a mixed methods approach

In the first strand of analyses, we explored which intervention features (components and processes) are associated with successful implementation of an intervention. This first strand involved undertaking qualitative comparative analysis (QCA) to uncover which configurations of these features (known as 'conditions' in QCA terminology) are aligned with successful intervention implementation. The QCA served to generate hypotheses about the importance of different intervention components and processes that were tested in meta‐analyses (below). Conditions identified through QCA helped us to identify which conditions matter for implementing an intervention, and structuring the meta‐analysis helped us to identify their potential impact on the overall effectiveness of interventions. The possibility that hypotheses were generated and tested on the same dataset was avoided due to very little overlap between studies included in the QCA synthesis and studies included in the meta‐analyses.

To examine the effectiveness of school‐based asthma self‐management interventions in improving children's outcomes, we undertook meta‐analyses. We performed subgroup analyses based upon results of the QCA described above.

We undertook the synthesis of process evaluations performed before the RCTs were conducted to remain blinded to the possible impact of specific measures. We further examined the link between implementation and effectiveness by estimating whether interventions defined as 'successful' in terms of their implementation were those with greater effect sizes. These analyses focused on a subgroup of studies adopting diverse designs (as outlined below).

Data synthesis part 1 ‐ using process evaluation studies for qualitative comparative analysis (QCA) of determinant conditions for successful intervention implementation

Qualitative comparative analysis (QCA) is used to identify configurations of conditions associated with successful intervention implementation. QCA takes a study‐based approach (accounting for several of the study's observed characteristics simultaneously) rather than a variable‐based approach, so that the focus is on different configurations of conditions (Thomas 2014). As this approach is relatively novel to systematic reviews, we have provided further information on the underlying principles and operationalisation of the approach in Appendix 8. The QCA approach used here aimed to generate theories about components 'sufficient' for triggering successful implementation; 'sufficient' relationships signify that an outcome is triggered in the presence of a sufficient condition or a sufficient condition set, but that other pathways to triggering the outcome may also exist. Here the outcome is successful implementation, and conditions are intervention characteristics and processes. In analysing our data, we followed the steps laid out by others (Ragin 2009; Thomas 2014).

We began by operationalising our data and creating a set of rules on how data should be coded for creating a data table of intervention characteristics (known as 'conditions' in QCA terminology) and the extent to which an intervention was successfully implemented (the outcome in this case). In the section titled Secondary outcomes, we have described the way in which we derived our outcome variable, and in the section titled Data extraction and management, we have described our coding framework for other intervention characteristics of interest. Two review authors (DK, KH) coded data for each study and grouped the information into separate data tables reflecting different domains of an intervention (i.e. conditions): setting and participants (Table 3); recruitment and retention processes (Table 4); curriculum and pedagogical factors (Table 5); modifiable intervention design features (Table 5); and stakeholder involvement (Table 6). We adopted this strategy to avoid 'limited diversity', whereby too many possible combinations of intervention characteristics are unsupported by observed studies.
We constructed truth tables that move beyond examining individual studies (i.e. one row per study) to examining configurations of conditions. Configurations could be supported by no studies, one study, or multiple studies. Truth tables also show the extent to which a 'set' of studies belonging to a configuration overlap with the outcome set.
We checked the quality of the truth tables. For each truth table, we considered whether a spread of positive and negative outcomes was triggered; whether configurations were supported by (multiple) cases (especially for configurations triggering a successful outcome); whether some configurations were counterintuitive and whether some conditions showed identical patterns; and whether some conditions occurred too infrequently. Our most important check involved whether we observed contradictory configurations when evidence suggested that configurations triggered positive and negative outcomes. When we were unable to resolve these issues according to guidance provided in Thomas 2014, the analysis progressed no further (see Appendix 8).
We then implemented Boolean minimisation to identify the most logically simple expression of a 'pathway' to a successful outcome. A pathway in this case represents a configuration of conditions that is observed to sufficiently trigger an outcome. This solution is based on observed configurations of conditions only and is known as a 'complex solution'.
When we detected logical remainders, we incorporated these into further models as 'intermediate solutions' to simplify the solution and maintain its theoretical coherence (see Appendix 8). For intermediate solutions, review authors (DK, KH) set expectations on whether the conditions entered were likely to lead to success.
A sixth stage involved interpretation, when review authors considered the plausibility of the solution and determined whether conclusions were consistent with evidence obtained from individual cases. We constructed a consolidated model, using evidence from preceding models. We checked the quality of the overall solution to ensure that it did not trigger negation of the outcome; we also assessed the parameters of fit and the validity of simplifying assumptions.

Table 3. Data table for QCA model 1 ‐ setting and participants

	Successful intervention	School‐based health centre	High school	Parents directly involved	Teachers received training	School nurses or other stakeholders received training
Joseph 2010	0.52	0.55	1	0	0	0
Kouba 2012	0.33	0.33	1	1	0	0
Dore‐Stites 2007	0.67	0.66	0	1	0	0
Joseph 2013	1.00	0.55	1	0	0	0
Mujuru 2011	0.67	0.66	0	0	1	0
Henry 2004	0.83	0.33	1	0	1	0
Pike 2011	0.67	0.33	0	0	1	0
Spencer 2000	0.33	0.66	0	1	0	0
Engelke 2013	0.50	0.66	0.5	1	1	1
Splett 2006	0.50	1.00	0.5	0	1	1
Kintner 2012	0.83	0.66	1	1	0	1
Berg 2004	0.83	0.66	1	0	0	0
Howell 2005	0.33	0.75	0	1	0	0
Gerald 2006	0.33	0.55	0	0	0	0
Langenfeld 2010	0.33	0.66	0	0	1	0
Al‐Sheyab 2012	0.83	0.33	1	0	0	0
Levy 2006	0.52	0.33	0	0	1	0
Terpstra 2012	1.00	0.66	0.66	1	0	0
Horner 2015	0.67	0.66	0	0	0	0
Bruzzese 2008	0.94	0.66	0.66	1	0	0
Lee 2011	0.50	0.66	0	0	0	0
Bruzzese 2004	0.33	0.55	1	0	0	1
Cicutto 2013	0.67	0.33	0	0	0	1
Brasler 2006	0.00	0.66	0.66	1	0	0
Crane 2014	0.50	0.33	0	0	0	0
Bruzzese 2011	0.88	0.55	1	0	0	1
Magzamen 2008	0.19	0.55	0.75	0	1	0

QCA: qualitative comparative analysis.

Table 4. Data table for QCA model 2 ‐ recruitment and retention processes

	Successful intervention	Provision of additional marketing materials	Provision of incentives	Make‐up sessions provided	Reminders provided for attendance at activity
Joseph 2010	0.52	1	1	0	0
Kouba 2012	0.33	1	0	1	0
Dore‐Stites 2007	0.67	1	1	0	0
Joseph 2013	1.00	1	1	0	0
Mujuru 2011	0.67	0	0	0	1
Henry 2004	0.83	0	0	0	0
Pike 2011	0.67	0	0.5	0	0
Spencer 2000	0.33	1	0	0	0
Engelke 2013	0.50	0	0	0	0
Splett 2006	0.50	0	0	0	0
Kintner 2012	0.83	1	1	1	0
Berg 2004	0.83	0	1	0	0
Howell 2005	0.33	0	1	1	1
Gerald 2006	0.33	0	0	0	0
Langenfeld 2010	0.33	0	1	0	0
Al‐Sheyab 2012	0.83	0	0	0	0
Levy 2006	0.52	0	0	0	0
Terpstra 2012	1.00	1	1	1	1
Horner 2015	0.67	0	0	0	0
Bruzzese 2008	0.94	0	0	1	0
Lee 2011	0.50	0	0.75	0	0
Bruzzese 2004	0.33	0	1	0	1
Cicutto 2013	0.67	0	0	1	0
Brasler 2006	0.00	1	1	1	1
Crane 2014	0.50	0	0	0	0
Bruzzese 2011	0.88	0	0	1	0
Magzamen 2008	0.19	1	1	0	1

QCA: qualitative comparative analysis.

Table 5. Data table for QCA model 4 ‐ modifiable design features

	Successful intervention	Theory driven	Personalised or individual sessions	Intervention takes place during lesson time	Intervention takes place during students’ own free time	School nurse involved in delivery of the intervention
Joseph 2010	0.52	1	1	1	0.33	0
Kouba 2012	0.33	1	1	0	1	0
Dore‐Stites 2007	0.67	1	0	0.33	0.33	0.66
Joseph 2013	1.00	1	1	0.75	0.75	0
Mujuru 2011	0.67	0	0	1	0	0
Henry 2004	0.83	0	0	1	0	0
Pike 2011	0.67	0	0	1	0	0
Spencer 2000	0.33	0	1	0.33	0.33	0.66
Engelke 2013	0.50	0	0.66	0.33	0.33	1
Splett 2006	0.50	0	1	0.33	0.33	1
Kintner 2012	0.83	1	0	1	1	0.66
Berg 2004	0.83	1	0.66	0.33	0.33	0.66
Howell 2005	0.33	1	1	0.33	0.33	0.66
Gerald 2006	0.33	0	0	1	0.33	0
Langenfeld 2010	0.33	0	1	0.33	0.33	1
Al‐Sheyab 2012	0.83	1	0	0.33	0.33	0
Levy 2006	0.52	0	0.66	0.33	0.33	1
Terpstra 2012	1.00	1	0	0	1	0.66
Horner 2015	0.67	1	0	0	1	0
Bruzzese 2008	0.94	1	0	0.33	0.33	0.66
Lee 2011	0.50	1	0	1	0	0.66
Bruzzese 2004	0.33	1	1	0.75	0.75	0
Cicutto 2013	0.67	1	0	0	1	0
Brasler 2006	0.00	0	0	0.75	0.75	0.66
Crane 2014	0.50	1	0	0	1	0.66
Bruzzese 2011	0.88	1	1	0.33	0.33	0
Magzamen 2008	0.19	0	0	0	1	1

QCA: qualitative comparative analysis.

Table 6. Data table for QCA model 5 ‐ stakeholder involvement and engagement

	Successful intervention	School asthma policy	Good relationships/engagement with parents	Good relationships/engagement with school nurses	Child Satisfaction	School asthma policy
Joseph 2010	0.52	0	0	0	0	0
Kouba 2012	0.33	0	0	0	0	0
Dore‐Stites 2007	0.67	0	0.75	1	1	0
Joseph 2013	1.00	0	1	0	0	0
Mujuru 2011	0.67	0	0.25	0	0	0
Henry 2004	0.83	1	0	0	0	1
Pike 2011	0.67	0	0	0	0	0
Spencer 2000	0.33	0	1	1	0	0
Engelke 2013	0.50	1	1	0	0	1
Splett 2006	0.50	1	0	1	0	1
Kintner 2012	0.83	0	0.25	0	1	0
Berg 2004	0.83	0	0	0	1	0
Howell 2005	0.33	0	0.75	0.75	0.633333	0
Gerald 2006	0.33	0	0	0	0	0
Langenfeld 2010	0.33	1	0	1	0	1
Al‐Sheyab 2012	0.83	0	0	0	0.633333	0
Levy 2006	0.52	1	0	0	0	1
Terpstra 2012	1.00	0	0.25	0	0	0
Horner 2015	0.67	0	0	0	0	0
Bruzzese 2008	0.94	0	1	0	1	0
Lee 2011	0.50	0	0	0	0	0
Bruzzese 2004	0.33	0	0	0	0.633333	0
Cicutto 2013	0.67	1	0	0	0	1
Brasler 2006	0.00	1	0	1	0.633333	1
Crane 2014	0.50	0	0	1	0	0
Bruzzese 2011	0.88	0	0	0	0	0
Magzamen 2008	0.19	0	0	0	0	0

QCA: qualitative comparative analysis.

We constructed all QCA models using R and a package developed by Thiem and Dusa (Thiem 2013). We have outlined further details of all steps, as well as the background to the method, in Appendix 8.

Data synthesis part 2 ‐ using RCTs for meta‐analyses of effectiveness

We combined data in Review Manager 5.3 (RevMan 2014), and we conducted some analyses and data transformations in STATA (when we encountered cluster randomised trials, we converted our standard errors using EPPI‐Reviewer functions (Thomas 2010)). We expected outcomes to be reported as similar units of analysis, although we encountered several variations and used Chinn's formulae for converting effect sizes and standard errors between SMDs and ORs (Chinn 2000), according to direction provided in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011). In addition, although we had originally specified daytime and night‐time symptoms as a single outcome, we split this into two separate outcomes to maintain conceptual coherence.

Occasionally, we could not incorporate some data into the meta‐analyses because of methodological difficulties in combining these data (including data based on rank (e.g. median)). Other changes and forms of imputation for missingness included the following: (I) basing the effect size for quality of life from Al‐Sheyab 2012 on the P value because of uncertainty regarding the effect size derived from point estimates and the precision provided; (ii) basing effect sizes for Cicutto 2013 on approximations of the numbers of participants in control and treatment groups; and (iii) estimating the numbers in treatment and control arms for Clark 2005 (assuming equal distribution of the overall sample size); we also imputed an OR of 0.996 for a value reported as 1.00 for Clark 2005 for ED visits, so we could combine the information from different models.

Data synthesis part 3: adjunct meta‐analyses exploring the link between implementation and effectiveness of school‐based asthma self‐management interventions

Methods used by review authors for the adjunct meta‐analyses followed the same processes as were used for the main meta‐analysis (part 2) in terms of the approaches taken in extracting effect sizes and combining data. The difference between analyses is that results of part 3 are based both on RCTs included in the main analyses (part 2) and on studies included in part 1 that allow for calculation of an effect size for school absences and/or emergency department visits. All studies included here must have included a control group and must have allowed for calculation of successful implementation, which we defined in the same way as our QCA analysis (part 2), and represented a combined indicator around attrition, adherence, and dosage.

Rating the certainty of the evidence

The certainty of evidence rating reflects the extent to which we can be confident that results for review outcomes reflect the true effect (Guyatt 2008). We rated the certainty of evidence for our main outcomes using methods developed by the GRADE (Grades of Recommendation, Assessment, Development and Evaluation) Working Group (http://www.gradeworkinggroup.org/publications/JCE_series.htm). We considered the possible impact of each of the following factors on our outcomes of interest.

Risk of bias.
Imprecision.
Inconsistency.
Indirectness.
Publication bias.

We attempted to identify a representative control group risk to illustrate the effects of our meta‐analysis results in absolute terms. We tabulated GRADE ratings alongside absolute and relative effects in a 'Summary of findings' (SoF) table for the following outcomes.

Asthma symptoms or exacerbations leading to admission to hospital.
Asthma symptoms or exacerbations leading to emergency department (ED) visits.
Unplanned visit to hospital or GP due to asthma symptoms.
School absence.
Experience of daytime symptoms.
Use of reliever therapies such as beta₂‐agonists.
(Health‐related) quality of life.

We generated the SoF table using the GRADE Guideline Development tool (GDT). We have described elsewhere further analyses undertaken to explore heterogeneity in effect size.

Subgroup analysis and investigation of heterogeneity

We conducted a statistical test for heterogeneity across subgroups using an I² statistic. We planned to construct a multi‐variate meta‐regression model based on our results for different outcomes. However, the small number of included studies precluded this possibility. We undertook prespecified subgroup analyses to investigate heterogeneity on the basis of the following characteristics, which are represented in our logic model as child‐level, school‐level, and contextual moderators, as well as modifiable design characteristics of the intervention itself, which we identified on the basis of QCA.

Setting: elementary/primary school versus secondary/high school.
Age: five to 10 years; 11 to 15 years; 16 years and older.
Socio‐economic level: low or mixed/high/unclear.
Delivery of intervention: healthcare provider (e.g. health educator, school nurse, other healthcare professional) versus other professional (e.g. teacher, mixture) versus other model of delivery (e.g. peer led).
Other (prespecified): intervention moderators developed from hypotheses generated through syntheses of process evaluation data including whether the intervention was theory driven, whether parents were actively involved, and the timing of the intervention during the school day. We entered these as single conditions and as groups reflecting configurations.

We measured some indicators, such as socio‐economic status, very differently, and we used broad groupings based on income, social class, or other indicators of social position, such as having received means tested benefits.

Sensitivity analysis

We undertook sensitivity analyses on the basis of the following.

Risk of bias assessment: we included all studies in the primary analysis and restricted included studies to those that were not classed as having high risk of bias for any single domain.
Fixed‐effect modelling.
Exclusion of cluster study data from outcomes (originally intended when external or imputed ICCs had been used, although this applied to most included cluster RCTs).

We did not plan to apply an equivalent for QCA modelling, although we did conduct robustness checks, including whether solutions predicted negation of the outcome.

We had intended to run sensitivity analyses based on the severity of children's asthma; however, no intervention specifically targeted children at particular levels of asthma severity, and inconsistent and low levels of reporting of asthma severity meant that we did not conduct these analyses. We have reported elsewhere other deviations from the protocol.

Results

Description of studies

We have reported the characteristics of all included studies in the Characteristics of included studies section; Table 7 presents an additional summary of how process evaluations met the review inclusion criteria.

Table 7. Included process evaluation studies: methodological characteristics and processes described

Study	Type of study	Approach	Process evaluation elements
Al‐Sheyab 2012a	Feasibility study	Qualitative	Thematic analyses of student perceptions
Berg 2004	Outcome and process evaluation	Qualitative and quantitative	Thematic analyses of student perceptions
Bignall 2015	Feasibility study	Qualitative and quantitative	Thematic analyses of student perceptions
Brasler 2006	Feasibility/case study of implementation	Quantitative data and trialist reports	Implementation challenges and facilitators identified
Bruzzese 2004	Feasibility study	Qualitative and quantitative	Section evaluating intervention reach, dosage, and student satisfaction
Bruzzese 2011	Outcome evaluation with section on process evaluation	Quantitative	Section evaluating intervention reach (dosage)
Bruzzese 2008	Feasibility study	Qualitative and quantitative	Stand‐alone section on process evaluation results assessing implementation and student perceptions
Carpenter 2016	Outcome and process evaluation	Qualitative and quantitative	Thematic analyses of student perceptions
Cicutto 2013	Outcome and process evaluation	(Mainly) Quantitative	In addition to information on other processes of interest, provided a description of wider school support through policy changes (process of interest included in the logic model)
Crane 2014	Feasibility study	Quantitative	Study was included as it represented an implementation study (through focus on the impact of changing dosage schedule)
Dore‐Stites 2007	Feasibility study	Quantitative	In addition to information on other processes of interest, provided information on student satisfaction
Engelke 2013	Feasibility study	Quantitative	Detailed process/implementation information was provided
Gerald 2006	Outcome and process evaluation	(Mainly) Quantitative	In addition to information on other processes of interest, provided a description of implementation challenges
Henry 2004	Outcome and process evaluation	(Mainly) Quantitative	In addition to information on other processes of interest, provided a description of wider school support through policy changes (process of interest in the logic model) and assessment of sustainability
Horner 2015	Outcome evaluation with process evaluation information	Quantitative	Included detailed information on attrition and cost‐effectiveness
Howell 2005	Outcome and process evaluation	Quantitative	In addition to information on other processes of interest, provided information on student satisfaction
Jackson 2006	Outcome evaluation with process evaluation information	Quantitative	In addition to information on other processes of interest, provided information on student satisfaction
Joseph 2010	Outcome and process evaluation	Quantitative	In addition to information on other processes of interest, provided detailed information on non‐adherence
Joseph 2013	Outcome and process evaluation	Quantitative	Included detailed studies of non‐adherence and relationship with student characteristics
Kintner 2012	Feasibility study	Quantitative	In addition to information on other processes of interest, provided information on student satisfaction
Kouba 2012	Outcome evaluation with process evaluation information	Quantitative	In addition to information on other processes of interest, provided detailed information on dosage (and dose‐response)
Langenfeld 2010	Implementation study	Quantitative	In addition to information on other processes of interest, provided detailed information on dosage (and dose‐response)
Lee 2011	Implementation study	Qualitative	In addition to information on other processes of interest, provided detailed information on instructor experiences
Levy 2006	Outcome evaluation with process evaluation information	Quantitative	In addition to information on other processes of interest, provided information on parental adherence to intervention protocol
Magzamen 2008	Outcome evaluation with process evaluation information	Quantitative	In addition to information on other processes of interest, provided information on attrition
McCann 2006	Outcome evaluation with process evaluation information	Quantitative	In addition to information on other processes of interest, provided information on teacher adherence/school level commitment
Mickel 2016	Outcome and process evaluation	Qualitative and quantitative	Thematic analyses of student perceptions
Mujuru 2011	Outcome and process evaluation	(Mainly) Quantitative	In addition to information on other processes of interest, provided a description of parental satisfaction
Pike 2011	Outcome and process evaluation	(Mainly) Quantitative	In addition to information on other processes of interest, provided information on teacher adherence/school level commitment
Richmond 2011	Outcome and process evaluation	(Mainly) Quantitative	Included detailed information on adherence and awareness
Spencer 2000	Outcome and process evaluation	Quantitative	In addition to information on other processes of interest, provided information on instructor satisfaction and school level commitment
Splett 2006	Outcome and process evaluation	Quantitative	In addition to information on other processes of interest, provided information on adherence and school level commitment
Terpstra 2012	Outcome and process evaluation	Quantitative	In addition to information on other processes of interest, represented an implementation study by including a focus on the impact of parental involvement/increasing parental awareness

Results of the search

We performed the first search in April 2015, and an updated search in April 2016. We conducted further searches on 25 August 2017. Two members of the review team (KH, DK) conducted the searches for process evaluation studies (see Figure 2). The Cochrane Airways Information Specialist, Liz Stovold, conducted the searches for outcome evaluation studies (see Figure 3). Review team members (KH, DK) performed initial automated checks for duplication using EPPI‐Reviewer software during the data screening and extraction process. After de‐duplication, we (KH, DK) screened 29,384 titles and abstracts of potential process evaluation studies, facilitated by text mining, as well as 350 title and abstracts for eligibility as outcome evaluations. Following application of inclusion criteria to review of titles and abstracts, KH and DK independently assessed the remaining 1066 full‐text process evaluation records and 105 full‐text outcome evaluation records for eligibility for inclusion. We included 54 papers, from 33 different studies, for further analysis as process evaluation studies, and 44 papers, from 33 different studies, for further analysis as outcome evaluation studies.

Figure 2

Process evaluation study flow diagram.

Figure 3

Outcome evaluation study flow diagram.

We identified several potential additional sources as ongoing studies (n = 4; see Characteristics of ongoing studies) and other studies as awaiting classification (n = 5; see Characteristics of studies awaiting classification).

Included studies

We included in the review 33 process evaluation studies and 33 outcome evaluation studies that met the inclusion criteria. We have described the characteristics of process and outcome evaluation studies separately below. We noted little overlap between the 33 studies included in both sets of studies, with Bruzzese 2004,Bruzzese 2008,Bruzzese 2011,Cicutto 2013,Gerald 2006,Henry 2004,Horner 2015,Howell 2005,Levy 2006,McCann 2006, and Splett 2006 (11/33) common to both sets of studies, although Bruzzese 2004 and McCann 2006 did not contribute data to the meta‐analyses.

Characteristics of process evaluation studies

Study population and intervention characteristics

Process evaluations of asthma self‐management interventions in schools reported on a diversity of intervention models. Nine studies included evaluations of the effectiveness of Open Airways for Schools (OAS) (American Lung Association 2018), or modifications to this programme (see Table 8). OAS consists of six 40‐minute sessions, aimed at groups of children aged eight to 11 who learn about different topics including general information about asthma, how to recognise and manage asthma symptoms, and problem‐solving and decision‐making about asthma medication. Authors of process evaluation studies described other intervention models (e.g. PowerBreathing (Berg 2004); Staying Healthy‐Asthma Responsible and Prepared (SHARP; Kintner 2012); Asthma Self‐Management for Adolescents (ASMA; Bruzzese 2004; Bruzzese 2008)), although these were diffuse across studies and were common to no more than two included studies.

Table 8. Process evaluation studies ‐ summary of intervention characteristics

	Named theoretical framework	Aim	Intervention type	Control	Intensity	Included in QCA
Al‐Sheyab 2012a	Developmental stages (not named)	To assess feasibility in the Jordanian context of a peer‐led, school‐based asthma education programme	Triple A. Children received education through interactive teaching and learning activities	N/A	14 hours over 6 days	Setting and participants; further modifiable design features; stakeholder involvement and engagement
Berg 2004	Social learning theory	To evaluate effects of the Power Breathing programme and individual coaching sessions on asthma knowledge and functional health status	Power Breathing. Children received education in a group session on asthma management	N/A	2 weeks	Stakeholder involvement and engagement
Bignall 2015	None	To test the feasibility and preliminary efficacy of a school‐based RCT on breathing retraining for asthma outcomes and anxiety symptoms	Single workshop for children. Children received information on relaxation/breathing techniques	30 minutes of standard asthma education	2 face‐to‐face visits 1 month apart	None
Brasler 2006	None	To provide adolescents with knowledge and skills to take control of their asthma; to enhance knowledge and skills of school staff, health professionals, and parents	Power Breathing. Children received basic asthma education and addressed social/lifestyle concerns	N/A	3× 90‐minute or 6× 45‐minute sessions	None
Bruzzese 2004	Self‐regulation theory	To help students weave asthma and management strategies into their self‐identity	ASMA. Students were taught how to manage their asthma to prevent symptoms and reduced quality of life. Continued medical education was also offered to medical providers	Usual care	3 workshops 2 or 3 weeks apart for 8 weeks	Stakeholder involvement and engagement
Bruzzese 2011	Social cognitive theory	To test the efficacy of ASMA	ASMA; academic detailing. Students attended workshops to empower them to manage their asthma. Parents received training on how to support their child's need to manage his or her asthma	Usual care	8‐week programme/3× 45‐minute sessions and individual coaching sessions once a week for 5 weeks	Further modifiable design features
Bruzzese 2008	Social cognitive theory; cognitive‐behavioural therapy	To test the feasibility and short‐term outcomes of asthma: it’s a family affair!	OAS and ASMA; caregiver education. Intervention students received education about asthma, based on existing materials, from coping with asthma at home and at school; OAS and ASMA	Usual care	6× 75‐minute group sessions once a week for 6 weeks; caregiver 5× 90‐minute sessions once a week	Setting and participants; further modifiable design features; stakeholder involvement and engagement
Carpenter 2016	None	To test whether a tailored inhaler technique video intervention could be feasibly implemented by school nurses; to improve the inhaler technique of children with asthma	Multiple sessions for children. Children watched a tailored video and demonstrated their inhaler technique before and after	N/A	6 weeks or less	None
Cicutto 2013	Social cognitive theory	To prepare and support children with asthma to be successful managers of their asthma, thereby reducing school absenteeism, interrupted activity, and health service use	Roaring Adventures of Puff. Workshops included goal‐setting and self‐monitoring, trigger identification, control and avoidance, basic pathophysiology, medication use, symptom recognition, and the asthma action plan, using interactive techniques	Usual care	Unclear	Setting and participants
Crane 2014	Educational theory of Jean Piaget	To pilot a shorter, condensed OAS education programme as an alternative, yet still effective, delivery approach compared to the lengthier original programme	OAS. Children received education from OAS	Non‐equivalent intervention	10 weeks	Setting and participants; further modifiable design features
Dore‐Stites 2007	None	Unclear	OAS; Quest for the Code. Children received a computer game, home activities, and caregiver information	N/A	20 minutes a week for 8 to 9 weeks	Further modifiable design features
Engelke 2013	Case management theory	To identify the process of case management used by school nurses, and when they provide case management to students with asthma. The second aim was to identify the impact of case management on parent perception of how well the child manages illness; parent perception of how well the child keeps up with school work; quality of life and academic achievement of children	Case management; nurse meetings; multiple sessions for children; multiple sessions for staff. Children received education and counselling, and parent/family education was delivered, as well as education and healthcare co‐ordination for teachers/staff	N/A	Unclear	None
Gerald 2006	None	To evaluate a comprehensive school‐based asthma management programme in an inner city, largely African American school system	OAS. The intervention included 3 educational programmes and medical management for children, as well as education for school staff	Usual care	Unclear	None
Henry 2004	Unclear	To determine whether an asthma education programme in schools would have a direct impact on student knowledge and attitudes toward asthma and quality of life of students with asthma; an indirect impact on teacher knowledge and attitudes on asthma and on school policies about asthma; and a sustainable programme after resources were withdrawn	Asthma education. A package about asthma was taught within the PD/H/PE (Personal Development, Health and Physical Education) strand of the school curriculum	Usual care	Unclear	Setting and participants
Horner 2015	Bruhn’s theoretical model of asthma self‐management	To test effects of 2 modes of delivering an asthma educational intervention on health outcomes and asthma management	7‐topic curriculum. The intervention was designed for children in rural areas and included asthma information	In‐school asthma classes	16× 15‐minute sessions for 5 weeks	None
Howell 2005	Learning theory and behaviour modification	To examine whether it was feasible to implement an interactive computer game at school health centres. Second, to examine whether exposure to the game was effective in increasing asthma knowledge, reducing asthma symptoms, and reducing unnecessary healthcare use compared with no exposure to the game	Quest for the Code. Computer game	Usual care	4× 30‐minute sessions	None
Jackson 2006	None	To evaluate knowledge and attitude outcomes of an educational asthma programme for third grade children with and without asthma	Single sessions for children. Children completed an educational programme. Teachers were also encouraged to attend	N/A	3 classes per session for 11 sessions	None
Joseph 2010	None	To develop and evaluate a multi‐media, web‐based asthma management programme	Puff City. A web‐based programme was delivered to children to focus on adherence, inhaler availability, and smoking cessation/reduction	Generic asthma websites	Unclear	Further modifiable design features; stakeholder involvement and engagement
Joseph 2013	Behavioural theory	To evaluate a school‐based RCT to evaluate Puff City	Adapted version of the Puff City computer programme	Generic asthma education	4× 15‐minute sessions	None
Kintner 2012	Lifespan development perspective	To evaluate the feasibility of the SHARP programme for students, their family, school personnel, and community partners	SHARP; Community Coalition component	N/A	Once a week for 10 weeks plus a 3‐hour community component	Setting and participants; further modifiable design features; stakeholder involvement and engagement
Kouba 2012	Orem’s self‐care deficit theory	To determine the effectiveness of the ICAN programme for nutrition knowledge and dietary behaviours	Single workshop for staff; multiple sessions for children; Quest for the Code; Fight Asthma Now; additional nurse meetings; combined education	N/A	8 weeks	None
Langenfeld 2010	None	Unclear	OAS; case management; stand‐alone respiratory therapy. Children received the OAS curriculum and case management asthma strategies developed with teachers	N/A	6× 40‐minute sessions for 1 school year	None
Lee 2011	The functional context approach	To evaluate the effectiveness and feasibility of using undergraduate nursing students as facilitators to deliver an asthma management programme	OAS. Children received the OAS curriculum	N/A	Unclear	Further modifiable design features
Levy 2006	None	To evaluate the effectiveness of a school‐based nurse case management approach to asthma in students with poor control	OAS; monitoring of students; health status. Students received OAS education and weekly monitoring of their health status	Usual care	1 school term	None
Magzamen 2008	None	To evaluate the implementation of Kickin’ Asthma	Multiple sessions for children; Kickin’ Asthma. Educational sessions, similar to the OAS curriculum. Customised letters were also sent home to describe health needs and goals for each child	N/A	3 months	None
McCann 2006	None	To assess whether a school‐based intervention would produce clinical and psychological benefits for children with asthma	Education; role‐play. The intervention focused on describing the respiratory condition through a role‐play	Respiratory education	45‐minute session	None
Mickel 2016	None	To provide Iggy education to more than 75% of children with asthma; To increase asthma knowledge; increase families’ awareness of asthma; and cultivate collaboration between school nurses and asthma providers	Iggy and the Inhalers intervention. Children received an asthma education video, poster, comic book, sticker, and trading card programme	N/A	Unclear	None
Mujuru 2011	None	To demonstrate the feasibility of a school‐based asthma education programme for students and to evaluate parents’ perspectives on the intervention	OAS. Children received the OAS curriculum	N/A	40‐minute session once a week for 2 months	None
Pike 2011	None	To assess student asthma knowledge gain, teacher acceptance, and grade appropriateness after an intervention	Multiple sessions for children; integrated into the curriculum. Teachers taught lessons with information about asthma	Usual care	7 lesson plans	Setting and participants
Richmond 2011	None	To increase the number of current provider‐written asthma action plans submitted to the school nurse at the beginning of the school year	Breathe Your Best. Students were encouraged to receive an asthma action plan from their doctor and to collect their prescriptions	N/A	Unclear	None
Spencer 2000	None	To evaluate the OAS programme for children	OAS. Children received the OAS curriculum	N/A	6× 40‐minute sessions	None
Splett 2006	None	To evaluate the effectiveness and sustainability of the Healthy Learners Asthma Initiative	Children received training on asthma self‐management. Licensed nurses and healthcare assistants received coaching and reinforcement from asthma resource nurses	Usual care	Varied according to asthma severity and need	None
Terpstra 2012	Social cognitive theory	To test a version of an intervention with a caregiver newsletter vs no newsletter	Multiple sessions for children; materials for parents. Children received skills training on how to use a peak flow meter. Parents received a newsletter about an important theme from the research	Intervention or intervention with a newsletter	6‐week training	Setting and participants; further modifiable design features

ASMA: Asthma Self‐Management for Adolescents.

ICAN: I Can Control Asthma and Nutrition Now.

N/A: not applicable.

OAS: Open Airways for Schools.

RCT: randomised controlled trial.

SHARP: Staying Healthy–Asthma Responsible & Prepared.

Triple A: Adolescent Asthma Action.

Across all studies, investigators taught a diverse curriculum. Although most studies mentioned that the intervention involved developing knowledge and skills around asthma physiology and monitoring and treatment of symptoms, fewer included studies explicitly mentioned that investigators aimed to develop alliances between children/parents and their care provider(s) (Dore‐Stites 2007; Gerald 2006; Richmond 2011; Terpstra 2012), and a greater number did involve parents in the intervention in other ways. Most interventions were reliant on trialists, research staff, and others from outside schools to deliver the intervention, although some interventions were primarily delivered, or supported pivotally, by school nurses (Engelke 2013; Langenfeld 2010; Levy 2006; Magzamen 2008; Splett 2006), teachers (Henry 2004; Mujuru 2011; Pike 2011), or children's peers (Magzamen 2008).

Several studies explicitly drew on social cognitive theory (Bruzzese 2008; Bruzzese 2011; Cicutto 2013; Terpstra 2012). Two studies from the same research team drew upon the Health Belief Model (Joseph 2010; Joseph 2013). Other theoretical models featured in only a single study included self‐regulation theory (Bruzzese 2004), learning or social learning theory (Berg 2004; Howell 2005), Piaget's pedagogical theory (Crane 2014), Orem's self‐care deficit theory (Kouba 2012), attribution theory (Joseph 2013), miscellaneous theoretical concepts that contributed to a theoretical framework (Al‐Sheyab 2012a), biopsychosocial theory (Dore‐Stites 2007), a transtheoretical model (Joseph 2010), and a functional context model (Lee 2011). A small minority of studies named a theoretical framework that was specific to asthma, with Horner 2015 employing Bruhn's theoretical model of asthma self‐management to underpin an intervention (Bruhn 1983), and Kintner 2012 drawing upon an asthma acceptance model (alongside a life course development perspective). These theoretical frameworks also differed in their use and in whether they supported the premise and emphasis of the intervention in a holistic manner, or whether they supported a particular pedagogical technique that was favoured in delivery of the intervention; this distinction was not clear in some studies. Few studies presented a clear logic model or theory of change to describe the underlying conceptual framework (Kneale 2015).

Five studies evaluated implementation of interventions involving delivery of self‐management education in part or mainly through electronic games or training provided through computers (Dore‐Stites 2007; Howell 2005; Joseph 2010; Joseph 2013; Kouba 2012). In two of these interventions (Joseph 2010; Joseph 2013), the information provided was tailored to children based on their input. In total, nine interventions had components that tailored content towards the needs of an individual child through delivery on a one‐to‐one basis or through delivery of personalised content (Bruzzese 2004; Bruzzese 2008; Howell 2005; Joseph 2010; Joseph 2013; Langenfeld 2010; Spencer 2000; Splett 2006).

Most studies took place in the USA (29/33 studies); several of these US‐based studies explicitly mentioned that the intervention took place in an urban or inner city area, or explicitly made reference to the diverse socio‐economic or ethnic background of participants (Berg 2004; Bignall 2015; Brasler 2006; Bruzzese 2004; Bruzzese 2010; Bruzzese 2011; Gerald 2006; Joseph 2010; Joseph 2013; Kouba 2012; Levy 2006; Magzamen 2008; Mickel 2016; Pike 2011; Richmond 2011; Splett 2006); in contrast, just two studies specifically explored implementation in rural areas (Horner 2015; Mujuru 2011). Fewer studies took place in high schools (14 studies) than in junior, middle, or elementary/primary schools (see Table 9).

Table 9. Process evaluation studies ‐ summary of study design, setting, and population

	Study design	Number of children	Country	Type of School	Recipients	Age of children (years)	Representation of children from BME backgrounds	Representation of children from low SES backgrounds
Al‐Sheyab 2012a	Case study	31	Jordan	High	Children	11 to 18	Unclear	Unclear
Berg 2004	Quasi‐experimental	13	USA	High	Children	15 to 18	46.2% African American	Unclear
Bignall 2015	Parallel‐group RCT	33	USA	High	Children	11 to 18	100% Black or African American	Unclear
Brasler 2006	Case study	342	USA	Junior/middle	Children; teachers; parents	11 to 14	Unclear	Unclear
Bruzzese 2004	Parallel‐group RCT	45	USA	High	Children	11 to 18	Unclear	Unclear
Bruzzese 2011	Parallel‐group RCT	345	USA	High	Children	11 to 18	45.5% Hispanic; 37.7% African American; 11.6% mixed; 5.2% other	75% free school meals
Bruzzese 2008	Parallel‐group RCT	24	USA	Junior/middle	Children; parents	11 to 14	41% Hispanic; 17% White; 8% African American; 34% other	8% unemployed; 21% part‐time employment; 71% full‐time employment
Carpenter 2016	Quasi‐experimental	25	USA	All school types	Children; nurses	Unclear	72% White; 12% Hispanic; 8% African American; 8% Black	Unclear
Cicutto 2013	Cluster RCT	1316	Canada	Primary/elementary	Children; school board; head teacher; teachers; peers	5 to 10	Unclear	25% to 50% deprived
Crane 2014	Quasi‐experimental	45	USA	Primary/elementary	Children	5 to 10	Unclear	Unclear
Dore‐Stites 2007	Quasi‐experimental	32	USA	Primary/elementary	Children; parents	5 to 10	39% African American; 28.6% Caucasian; 14.3% Hispanic; 18% biracial	34.6% < $20,000; 53.8% $21,000 to $40,000
Engelke 2013	Quasi‐experimental	143	USA	All school types	Children; teachers; parents; nurses	Unclear	40.6% Caucasian; 37.8% African American; 7% Latino; 14% other	63.6% on Medicaid
Gerald 2006	Cluster RCT	736	USA	Primary/elementary	Children; teachers	5 to 10	97% African American	Unclear
Henry 2004	Cluster RCT	4161	Australia	High	Children; teachers	11 to 14	Predominantly Caucasian	Unclear
Horner 2015	Cluster RCT	292	USA	Primary/elementary	Children	5 to 10	21.2% African American; 25% Spanish speaking	30.7% low SES
Howell 2005	Cluster RCT	24	USA	Primary/elementary	Children; parents	5 to 10	75% African American	Unclear
Jackson 2006	Quasi‐experimental	943	USA	Primary/elementary	Children	5 to 10	Unclear	Unclear
Joseph 2010	Parallel‐group RCT	314	USA	High	Children	11 to 18	Unclear	52% eligible for free school meals
Joseph 2013	Parallel‐group RCT	422	USA	High	Children	11 to 18	98% African American	73% on Medicaid
Kintner 2012	Quasi‐experimental	28	USA	High	Children; peers; families; teachers	11 to 14	53.6% African American; 32.1% White; 3.6% American; 10.7% biracial	35.7% low SES; 42.9% low middle SES; 17.8% upper middle SES; 3.6% high SES
Kouba 2012	Quasi‐experimental	25	USA	High	Children	11 to 18	92% African American; 4% Hispanic; 4% mixed	25% to 50% deprived
Langenfeld 2010	Quasi‐experimental	286	USA	Primary/elementary	Children; teachers	5 to 10	63% African American; 23.9% Hispanic; 6.4% White; 2.6% Asian	High percentage on free school meals
Lee 2011	Quasi‐experimental	827	USA	Primary/elementary	Children	5 to 10	Unclear	Unclear
Levy 2006	Cluster RCT	243	USA	Primary/elementary	Children; teachers	5 to 10	97% African American	80% on Medicaid
Magzamen 2008	Quasi‐experimental	845	USA	High; junior/middle	Children	11 to 18	Unclear	Unclear
McCann 2006	Parallel‐group RCT	219	UK	Primary/elementary	Children; teachers	5 to 10	Unclear	< 25% deprived
Mickel 2016	Quasi‐experimental	173	USA	Primary/elementary	Children	5 to 10	63.6% African American; 13.3% Hispanic; 20.2% White	> 50% deprived
Mujuru 2011	Quasi‐experimental	18	USA	Primary/elementary	Children; parents	5 to 10	Unclear	39% Medicaid
Pike 2011	Quasi‐experimental	236	USA	Primary/elementary	Children; teachers	5 to 10	75% African American (approx.)	80% free school meals (approx.)
Richmond 2011	Narrative	Unclear	USA	Primary/elementary	Children	5 to 10	100% African American	80% free school meals
Spencer 2000	Quasi‐experimental	369	USA	Primary/elementary	Children; parents	5 to 14	Unclear	34% free school meals
Splett 2006	Cluster RCT	1561	USA	All school types	Children; school staff	Unclear	66% African American; 6% Hispanic; 5% American Indian; 3% Asian; 20% White	73% free school meals
Terpstra 2012	Quasi‐experimental	58	USA	Junior/middle	Children; parents	11 to 14	> 50% BME	> 50% deprived

BME: black and minority ethnicity.

RCT: randomised controlled trial.

Time of assessment of process outcome measurements

Twenty‐one process evaluation studies collected pre‐ and post‐hoc data. Four studies collected post‐hoc data only (Al‐Sheyab 2012a; Berg 2004; Bruzzese 2004; Richmond 2011). Several studies collected data immediately after the intervention or within three months of cessation of the intervention (Bignall 2015; Bruzzese 2004; Bruzzese 2008; Carpenter 2016; Crane 2014; Gerald 2006; Howell 2005; Jackson 2006; Kintner 2012; Kouba 2012; Magzamen 2008; Mickel 2016; Mujuru 2011; Pike 2011; Spencer 2000; Splett 2006). The longest follow‐up data collection period lasted for 12 months post testing (Bruzzese 2011; Cicutto 2013; Horner 2015; Joseph 2010; Joseph 2013; McCann 2006). In a small number of studies, the follow‐up duration was unclear (Al‐Sheyab 2012a; Dore‐Stites 2007; Engelke 2013; Langenfeld 2010; Levy 2006; Richmond 2011; Terpstra 2012).

Measurement of process outcomes

We included 33 process evaluation studies, most of which adopted a quantitative approach to analyses. Process evaluation elements across these studies included thematic analysis of student perceptions, identification of implementation challenges and facilitators, reach of the intervention, and student satisfaction. We have provided further details of inclusion criteria and process evaluation elements for all process evaluation studies in Table 7 The descriptions below refer to all studies included as process evaluation studies, although we included in QCAs only those that we deemed to be of moderate or high intensity (see section on reduction of cases). Similarly, we transformed the data and ratings described below using direct and indirect transformations (see earlier methods).

Attrition

A total of 18 studies provided evidence that attrition was low. Five studies showed substantial attrition (Bruzzese 2004; Gerald 2006; Levy 2006; Magzamen 2008; Richmond 2011), with levels of attrition exceeding 20% and/or reported by trial authors as a substantial challenge.

Adherence to the intervention

A total of 21 studies reported child adherence. 'Child adherence' broadly referred to the extent to which children followed directions of the intervention, for example, in completing homework assignments, undertaking and completing intervention modules, or completing evaluation instruments. Fourteen studies presented evidence that child adherence with the intervention was good. Six studies highlighted evidence that adherence was not problematic among other stakeholders (Bruzzese 2011; Cicutto 2013; Jackson 2006; Joseph 2013; Kintner 2012; Splett 2006). Child adherence was problematic in eight studies (Brasler 2006; Gerald 2006; Howell 2005; Joseph 2010; Kouba 2012; Magzamen 2008; Richmond 2011; Spencer 2000); these judgements were based on reports from trialists and on reports of completion rates of intervention modules and/or completion of evaluation instruments.

Dosage of intervention received

'Dosage' broadly referred to the extent to which children received the intervention as intended, for example, in attending the expected number of sessions. This differed from attrition, in that children could have received a low dosage but may have not permanently dropped out; this also differed from adherence, in that children could have received a low dosage but were otherwise adherent. Participants received the intended dose of the intervention in nine studies (Bignall 2015; Bruzzese 2011; Jackson 2006; Joseph 2013; Kintner 2012; McCann 2006; Mickel 2016; Pike 2011; Terpstra 2012). In one study, researchers noted a dose‐response relationship (Kouba 2012). Seven studies reported that the intended dose was not achieved (Brasler 2006; Bruzzese 2008; Gerald 2006; Howell 2005; Joseph 2010; Langenfeld 2010; Magzamen 2008), with substantial numbers not receiving the intended intervention. In one study (Gerald 2006), this finding was based on reports of shortening of sessions. In another study, in which parental involvement was an integral component, study authors reported additional problems with dosage received (Bruzzese 2008). One study comparing an individualised intervention model versus a generic intervention model reported that the individualised model had higher levels of dosage, although both models showed relatively low levels of completion of all modules (Joseph 2010).

Combined indicator of 'successful' implementation

We combined data from process evaluation studies on attrition, adherence, and dosage into a single indicator. We summed scores across the three indicators and calibrated them to fall between zero and one, with 0.5 the point of maximum ambiguity and values over 0.5 indicating partial membership of the successful implementation set, up to a maximum possible value of one, which indicated full membership of the successful implementation set, values under 0.5 indicating more out of than in the set, and a value of 0 indicating full non‐membership of the successful implementation set. Eight studies were either fully or strongly within the successful implementation set (Al‐Sheyab 2012a; Berg 2004; Bruzzese 2008; Bruzzese 2011; Henry 2004; Joseph 2013; Kintner 2012; Terpstra 2012), and another five studies had scores that were mainly within the successful implementation set (Cicutto 2013; Dore‐Stites 2007; Horner 2015; Mujuru 2011; Pike 2011). A further 14 studies provided scores that were ambiguous or low implementation scores (Brasler 2006; Bruzzese 2004; Crane 2014; Engelke 2013; Howell 2005; Gerald 2006; Joseph 2010; Kouba 2012; Langenfeld 2010; Lee 2011; Levy 2006; Magzamen 2008; Spencer 2000; Splett 2006).

Characteristics of outcome evaluation studies (RCTs)

We have included in Table 10 further details of studies that met the criteria for study design, but from which we did not include data in the meta‐analysis.

Table 10. Outcome evaluation studies not included in the analyses

Study included as outcome	Reason data not included in quantitative analysis
Bruzzese 2004	Feasibility study uses randomised controlled trial (RCT) design with no quantitative data presented
Bruzzese 2010	Abstract only located and outcomes were not presented in an extractable format
Clark 2004	Published effect sizes that were extractable but of a different effect size from other studies
Clark 2010	No outcome measured in the study matched the review protocol
McCann 2006	Outcomes were not presented in an extractable format (disaggregated data for asthmatic children unavailable)
Monforte 2012	Abstract only located and outcomes were not presented in an extractable format
Mosnaim 2011	No outcome measured in the study matched the review protocol
Praena‐Crespo 2010	Abstract only located and outcomes were not presented in an extractable format
Pulcini 2007	No outcome measured in the study matched the review protocol
Srof 2012	Outcomes were not presented in an extractable format (data on overall quality of life were not presented in full; only subdomains of quality of life are available)

Study population and intervention characteristics

Most studies took place in the USA (22/33 studies), with fewer taking place in high schools (eight studies) than in junior, middle, or elementary/primary schools (see Table 11). Study reports showed substantial variation in the types of interventions that were trialled, although nine studies included evaluations of the effectiveness of Open Airways for School, or modifications to this programme (see Table 12). Study reports also showed substantial variety in the ways in which asthma self‐management interventions were delivered. Children received long programmes of sessions in some interventions, with 16 sessions delivered in two studies (Horner 2008; Horner 2015), and 10 sessions and eight sessions delivered in others (Kintner 2009; Patterson 2005, respectively). In contrast, researchers delivered three interventions in a single group session to children (Gerald 2006; Howell 2005; McCann 2006), although these interventions were supported by other activities including nurse visits or staff training. The number of sessions was not always commensurate with the quantity of content delivered however; for example, the intervention delivered in Atherly 2009 amounted to 4.5 hours of instruction over three sessions, and Horner 2015 delivered 4 hours of content over 16 sessions.

Table 11. Outcome evaluation studies ‐ summary of study design, setting, and population

	Study design	Number of children	Country	Type of school	Recipients	Age of children (years)	Representation of children from BME backgrounds	Representation of children from low SES backgrounds
Al‐Sheyab 2012	Clustered parallel RCT	261	Jordan	4 public high schools	Children	11 to 15	Unclear	Unclear
Atherly 2009	Clustered parallel RCT	524	USA	Junior and high schools	Children	11 to 15	Unclear	Unclear
Bartholomew 2006	Clustered parallel RCT	948	USA	Elementary schools	Children; care providers; parents/carers	5 to 10	45% African American; 51% Hispanic; 3% Caucasian	Deprived individuals > 50%
Bruzzese 2004	RCT	45	USA	2 public high schools	Children	Unclear	Unclear	Unclear
Bruzzese 2008	Clustered parallel RCT	24	USA	1 middle school	Children; caregivers	11 to 15	41% Hispanic; 17% African American	71% parents full‐time employment
Bruzzese 2010	Clustered parallel RCT	Unclear	USA	25 public schools	Children; caregivers	Mean age, 12.8	Unclear	Unclear
Bruzzese 2011	Clustered parallel RCT	340	USA	5 high schools	Children	Mean age, 15	> 80% BME	Unclear
Cicutto 2005	Clustered parallel RCT	256	Canada	26 elementary schools	Children	5 to 10	Unclear	Average household income $53,000
Cicutto 2013	Clustered RCT	1316	Canada	170 primary/elementary schools	Children; families	5 to 10	Unclear	Deprived individuals 25% to 50%
Clark 2004	Clustered parallel RCT	835	USA	14 public high schools	Children; parents/carers; classmates; school personnel	5 to 10	98% African American	45% annual income < $15,000
Clark 2005	Clustered parallel RCT	639	China	21 elementary schools	Children	7 to 11	Unclear	Unclear
Clark 2010	Clustered parallel RCT	1292	USA	19 middle schools	Children	Mean age, 11.6	93% African American	48% annual income < $15,000
Gerald 2006	Parallel‐group RCT	736	USA	54 elementary schools	Children	Mean age, 11	97% Black	Unclear
Gerald 2009	Parallel‐group RCT	290	USA	Unclear	Children	5 to 10	91% Black	Unclear
Henry 2004	Clustered parallel RCT	Unclear	Australia	Secondary schools	Children	11 to 15	< 50% BME	Unclear
Horner 2008	Clustered parallel RCT	183	USA	18 elementary schools	Children	5 to 10	47% Hispanic; 30% White; 22% African American	Unclear
Horner 2015	Clustered parallel RCT	196	USA	3 elementary schools	Children	5 to 10	> 50% BME	Deprived individuals 25% to 50%
Howell 2005	Clustered parallel RCT	25	USA	4 elementary schools	Children; families	5 to 10	75% African American	Unclear
Kintner 2009	Clustered parallel RCT	59	USA	5 schools	Children	9 to 12	30% Black; 36% White; 18% biracial	Deprived individuals 25% to 50%
Levy 2006	Clustered parallel RCT	243	USA	14 elementary schools	Children	5 to 10	98% African American	85% TennCare
McCann 2006	Clustered parallel RCT	229	England	24 primary/junior schools	Children; parents	5 to 10	Unclear	Deprived individuals < 25%
McGhan 2003	Clustered parallel RCT	162	Canada	18 elementary schools	Children	5 to 10	< 50% BME	Deprived individuals 25% to 50%
McGhan 2010	Clustered parallel RCT	206	Canada	Elementary schools	Children; parents/carers; teachers	Mean age, 8.6	Unclear	Unclear
Monforte 2012	Clustered parallel RCT	Unclear	USA	8 elementary schools	Children	5 to 10	Unclear	Unclear
Mosnaim 2011	Clustered parallel RCT	344 youth; 192 teens	USA	Elementary schools	Children	Median age 10	> 50% BME	Deprived individuals > 50%
Patterson 2005	Clustered parallel RCT	175	Ireland	Primary schools	Children	7 to 11	Unclear	Deprived individuals 25% to 50%
Persaud 1996	Parallel‐group RCT	36	USA	10 schools	Children	Mean age, 10.2	69% African American	69% received Medicaid
Praena‐Crespo 2010	Clustered parallel RCT	279	Spain	16 high schools	Children	11 to 15	Unclear	Unclear
Pulcini 2007	Clustered parallel RCT	40	USA	Middle schools	Children	11 to 15	Unclear	Unclear
Shah 2001	Clustered parallel RCT	272	Australia	High schools	Children	11 to 15	Unclear	Unclear
Splett 2006	Clustered parallel RCT	1561	USA	K‐8 schools	Children	5 to 15	66% African American	73% free school meals
Srof 2012	Parallel group RCT	39	USA	High schools	Children	14 to 18	Unclear	Unclear
Velsor‐Friedrich 2005	Clustered parallel RCT	52	USA	4 elementary schools	Children	Mean age, 10.1	100% African American	Unclear

BME: black and minority ethnicity.

RCT: randomised controlled trial.

Table 12. Outcome evaluation studies ‐ summary of intervention characteristics

	Named theoretical framework	Aim	Intervention type	Control	Intensity	Outcomes Included in meta‐analysis
Al‐Sheyab 2012	Self‐efficacy	To test the impact of the Triple A programme on health‐related outcomes in high school students	Triple A. Bilingual health workers trained peer leaders from year 11 to deliver 3 Triple A lessons	Unclear	3× lessons	HRQoL
Atherly 2009	None	To describe an analysis and results of the cost‐effectiveness of the Power Breathing programme	Power Breathing. This intervention focussed on education about asthma, asthma control strategies, and psychosocial concerns	Unclear	3× 90‐minute lessons	Hospitalisations; ED visits; Experience of daytime and night‐time symptoms
Bartholomew 2006	Social cognitive theory	To describe the evaluation of a school‐based intervention to improve asthma self‐management, medical care, the school environment, symptoms, and the functional status of children	Multi‐component intervention involving direct delivery to children, care providers, and parents/guardians. Children received education through the Watch, Discover, Think and Act interactive computer programme	Unclear	Unclear	Withdrawal
Bruzzese 2004	None	Unclear	ASMA. Continued medical education was also offered to medical providers	Usual care	3× lessons	None
Bruzzese 2008	Social cognitive theory; cognitive‐behavioural theory	To describe asthma: it’s a family affair; to present feasibility and preliminary outcome data from a pilot RCT	Elements of OAS and ASMA were provided to students; caregivers also received education	Usual care	6× lessons	Experience of daytime and night‐time symptoms; Withdrawal
Bruzzese 2010	None	To test the efficacy of an RCT: it’s a family affair, a school‐based, family‐focussed intervention to improve asthma outcomes in pre‐adolescents	ASMA and academic detailing. Students received workshops to empower them to manage their asthma. Parents received training to support their child’s need to manage their asthma	Unclear	Children: 6× lessons; caregivers: 5× lessons	Withdrawal
Bruzzese 2011	Social cognitive theory	Unclear	ASMA. Students received group sessions and individual tailored coaching sessions, delivered by trained health educators	Wait‐list control	3× group sessions; individual coaching sessions	Hospitalisations; ED visits; School absence; Restricted activity days; Unplanned GP or hospital visits; Experience of daytime and night‐time symptoms; Use of corticosteroids; Withdrawal
Cicutto 2005	Social cognitive theory; self‐regulation theory	To evaluate an asthma education programme for children with asthma	Roaring Adventures of Puff. Children received group sessions on asthma and goal‐setting	Usual care	6× lessons	Hospitalisations; ED visits; School absence; Restricted activity days
Cicutto 2013	Social cognitive theory	To implement an elementary school‐based asthma self‐management education programme for children with asthma; to work with schools to create an asthma‐friendly supportive school environment; to evaluate the programme	Roaring Adventures of Puff. Children received group sessions on asthma and goal‐setting	Usual care	6× lessons	ED visits; School absence; Restricted activity days; Unplanned GP or hospital visit; HRQoL; Withdrawal
Clark 2004	None	To assess the impact of a comprehensive school‐based asthma programme	OAS; control strategies for schools	Wait‐list control	6× lessons and 2× classroom sessions	School absence
Clark 2005	Social cognitive theory	To assess effectiveness in children in China of an asthma education programme adapted from a model developed in the USA	OAS; intervention directed at children only	Unclear	5× lessons	Hospitalisations; ED visits
Clark 2010	None	To assess self‐management and self‐management plus peer involvement	OAS; peer component. In the first treatment arm, an adapted form of OAS was delivered to children. In the second treatment arm, a peer education component was added	Usual care	6× lessons	Experience of daytime and night‐time symptoms
Gerald 2006	None	Unclear	OAS. The intervention included educational programmes and medical management for children, as well as education for school staff	Usual care	6× lessons	Hospitalisations; ED visits; School absence
Gerald 2009	None	To determine the effectiveness of school‐based supervised asthma therapy in improving asthma control	Children received asthma education, including a discussion of trigger avoidance (not manualised)	Usual care	1× lesson; multiple supervisions	School absence; Lung function; Use of reliever therapies; Withdrawal
Henry 2004	None	To determine whether an asthma education programme in schools would have a direct impact on student knowledge and attitudes on asthma and an indirect impact on teacher knowledge and attitudes	Asthma education. A package about asthma was taught within the PD/H/PE strand of the school curriculum	Usual care	3× lessons	HRQoL
Horner 2008	Asthma health education model	To examine changes in rural children’s asthma self‐management after they received classes, but before they received the family education session	Asthma self‐management. The curriculum included a 7‐step asthma self‐management plan	Health promotion education	16× lessons	Hospitalisations; Withdrawal
Horner 2015	Bruhn’s theoretical model of asthma self‐management	To test effects of 2 modes of delivering an asthma educational intervention on health outcomes and asthma self‐management in school‐aged children living in rural areas	7‐topic curriculum. The intervention was designed for children in rural areas and included asthma information	Health promotion education	16× lessons	Hospitalisations; ED visits; Withdrawal
Howell 2005	Social learning theory	To examine the feasibility of an interactive computer game in school‐based health centres; to test whether exposure to the game was effective in improving knowledge and reducing symptoms and healthcare use	Quest for the Code computer game. The caregiver also participated in medication interviews and received a home visit	Usual care	30‐minute session	ED visits; Experience of daytime and night‐time symptoms; HRQoL; School absence; Corticosteroid dosage
Kintner 2009	Lifespan development perspective	To evaluate the preliminary efficacy of SHARP	SHARP. Students worked through the SHARP curriculum. Caregivers also received a 3‐hour information sharing programme	Usual care	10× lessons	HRQoL; Withdrawal
Levy 2006	None	To evaluate the effectiveness of a school‐based nurse case management approach to asthma in students with poor control	OAS; monitoring of students; health status. Students received OAS education and weekly monitoring of their health status	Usual care	Weekly group sessions and weekly individual sessions	Hospitalisations; ED visits; Withdrawal
McCann 2006	None	To assess whether schools are an appropriate context for an intervention designed to produce clinical and psychological benefits for children with asthma	Education; role‐play. The intervention focussed on describing the respiratory condition through a role‐play	Education about the respiratory system	1× workshop	None
McGhan 2003	Social cognitive theory	To determine whether an interactive childhood asthma education programme improved asthma management behaviours, health status, and quality of life in elementary school children	Roaring Adventures of Puff. Children received education on asthma in a group setting. Parents and teachers were invited to participate in a school‐based asthma awareness event	Usual care	6× lessons	ED visits; School absence; Unplanned GP or hospital visit; Experience of daytime and night‐time symptoms; Withdrawal
McGhan 2010	Social cognitive theory; self‐regulation theory	To assess the feasibility and impact of the Roaring Adventures of Puff programme	Roaring Adventures of Puff delivered to children. Parents and teachers participated in an asthma awareness event.	Usual care	6× lessons	ED visits; School absence; Unplanned GP or hospital visit; Experience of daytime and night‐time symptoms; Withdrawal
Monforte 2012	None	To evaluate the implementation of OAS	OAS. No further information was given	Unclear	Unclear	HRQoL
Mosnaim 2011	None	To assess the impact of the Fight Asthma Now educational programme among 2 populations of predominantly low‐income minority students	One‐to‐one training on spacer technique, peak flow meter use, and use of an asthma action plan. Teens also received education on tobacco avoidance and peer pressure	Usual care	4× sessions	None
Patterson 2005	PRECEDE model	To evaluate the effectiveness of a programme of asthma clubs in improving quality of life for primary school children with asthma	SCAMP. Children used a workbook during sessions to learn about asthma	Wait‐list control	8× sessions	Restricted activity days; Lung function; HRQoL; Withdrawal
Persaud 1996	None	To assess the effectiveness of an intervention on knowledge, locus of control, attitudes towards asthma, functional status, school attendance, and ED visits	Individualised education sessions. Children had a personal peak flow meter in the school health office. The school nurse also reviewed the student asthma diary and discussed this with them	Usual care	3× lessons and weekly education sessions	ED visits; School absence
Praena‐Crespo 2010	None	To verify whether an asthma education program in schools would have direct benefit for student knowledge and attitudes towards asthma and quality of life for students with asthma	Asthma programme. No further information was given (abstract only)	Unclear	3× lessons	None
Pulcini 2007	None	To determine the effectiveness of an intervention to increase the number of AAPs in schools	Peak flow education. Children were given a peak flow meter and were educated on the correct technique to measure lung function	Unclear	Daily for 2 weeks	None
Shah 2001	None	To determine the effects of a peer‐led programme for asthma education on quality of life and related morbidity in adolescents with asthma	Triple‐A: asthma education and empowerment. Students learnt how to educate their peers about asthma. Peers also led 3 health lessons for classes in school	Wait‐list control	3× sessions	Experience of daytime and night‐time symptoms; Lung function; HRQoL; Withdrawal
Splett 2006	None	To improve asthma management among school children and reduce asthma‐related school absences, hospitalisations, and ED visits	Children received training on managing their asthma. Licensed nurses and healthcare assistants received coaching and reinforcement from asthma resource nurses	Usual care	Unclear	School absence; Unplanned GP or hospital visit
Srof 2012	Health promotion model	To determine effects of coping skills on asthma self‐efficacy, social support, quality of life, and peak flow among adolescents	Asthma diary; 5× coping skills sessions. Students received coping skills training and completed diary entries	Usual care	Sessions over 5 weeks	None
Velsor‐Friedrich 2005	Self‐care deficit theory	To test a 2‐part intervention on selected psychosocial and health outcomes for children with asthma	OAS; nurse practitioner visits. Children received the OAS education curriculum and nurse practitioner visits to assess asthma health and further education	Usual care	6× group sessions; individual nurse sessions	ED visits; Experience of daytime and night‐time symptoms; Lung function

AAP: XXX.

ASMA: Asthma Self‐Management for Adolescents.

ED: emergency department.

GP: general practitioner.

HRQoL: health‐related quality of life.

ICAN: I Can Control Asthma and Nutrition Now.

OAS: Open Airways for Schools.

PD/H/PE: personal development/health/physical education.

PRECEDE: Predisposing, Reinforcing, and Enabling Causes in Educational Diagnosis and Evaluation.

RCT: randomised controlled trial.

SCAMP: School Care and Asthma Management Project.

SHARP: Staying Healthy–Asthma Responsible & Prepared.

Triple A: Adolescent Asthma Action.

Several studies collected outcome data immediately after the intervention or within three months (Atherly 2009; Bruzzese 2004; Bruzzese 2008; Gerald 2006; Horner 2008; Howell 2005; Kintner 2009; Mosnaim 2011; Patterson 2005; Persaud 1996; Shah 2001; Srof 2012), or they appeared to collect data concurrently with intervention delivery (Splett 2006). The longest period between the end of the intervention and data collection was 36 months in Bartholomew 2006, and 24 months in Clark 2004 and Clark 2010, although for a minority of studies, the length of follow‐up was not clear (Levy 2006; Monforte 2012; Pulcini 2007). We included many studies on the basis of study design, although these studies did not contribute to the meta‐analyses, as they did not collect data on the outcomes of interest or did not collect these data in an extractable format (see Table 10).

Primary outcomes

Asthma symptoms or exacerbations leading to admission to hospital

Six outcome studies provided data on asthma exacerbations leading to admission to hospital that were combined in meta‐analyses (Atherly 2009; Bruzzese 2011; Clark 2005; Horner 2008; Horner 2015; Levy 2006). One study collected information on hospitalisations but did not disaggregate the information by treatment status (Bartholomew 2006), and another study provided disaggregated information on median hospitalisations that could not be combined in meta‐analyses (Gerald 2006). Two studies assessed hospitalisations using hospital or school medical records (Gerald 2006; Levy 2006); three studies assessed hospitalisations using parent reports (Clark 2005; Horner 2015; Horner 2008); and two studies used child reports (Atherly 2009; Bruzzese 2011). Of the six studies included in the meta‐analyses, most collected outcome data on hospitalisations after a substantial period between receipt of the intervention and assessment of the outcome had elapsed (12 months in the case of Bruzzese 2011; Clark 2005; and Horner 2015; and seven months in the case of Horner 2008); less time had elapsed in the case of Atherly 2009 and Levy 2006, in which assessment took place within three months of receipt of the intervention. Studies in which a longer time had elapsed between intervention and assessment tended to be those with a longer exposure time over which the outcome was measured.

Asthma symptoms or exacerbations leading to emergency department visits

Fifteen outcome evaluation studies collected data on asthma symptoms or exacerbations leading to an emergency department (ED) visit (Atherly 2009; Bartholomew 2006; Bruzzese 2011; Cicutto 2005; Cicutto 2013; Clark 2005; Gerald 2006; Horner 2008; Horner 2015; Howell 2005; Levy 2006; McGhan 2003; McGhan 2010; Persaud 1996; Velsor‐Friedrich 2005). However, we did not use data from Bartholomew 2006 because study authors did not disaggregate the data by treatment status, and we could not combine data from Gerald 2006 because of incompatibility in the unit of assessment. Three studies used school or hospital administrative records to assess ED visits, with records provided by the medical hospital (Gerald 2006; Levy 2006; Persaud 1996). Parents were frequently the sources of ED data: one study collected these data using tracking sheets of ED attendance provided by parents (Cicutto 2013); another study collected data through parent interviews (Cicutto 2005); six studies used various parent self‐completion questionnaires (Clark 2005; Horner 2015; Horner 2008; Howell 2005; McGhan 2003; McGhan 2010), and one specifically used the Usherwood symptom questionnaire (Bartholomew 2006). One study collected data from children's asthma diaries (Velsor‐Friedrich 2005), and others collected data from children's reports (Atherly 2009; Bruzzese 2011).

Of the 13 studies included in the meta‐analyses, most collected outcome data on ED visits after a substantial period had elapsed between receipt of the intervention and assessment of the outcome (12 months in the case of Bruzzese 2011,Cicutto 2005,Cicutto 2013,Clark 2005,Horner 2015,McGhan 2003,McGhan 2010; seven months in the case of Horner 2008; and 20 weeks in the case of Persaud 1996); less time had elapsed in the case of Atherly 2009,Howell 2005, and Levy 2006, which performed assessment within three months of receipt of the intervention. As was the case above, studies in which a longer time had elapsed between intervention and assessment were those with a longer exposure time over which the outcome was measured (see Table 11 for full details).

Absence from school

Twelve outcome evaluation studies assessed school absence or attendance (Bartholomew 2006; Bruzzese 2011; Cicutto 2005; Cicutto 2013; Clark 2004; Gerald 2006; Gerald 2009; Howell 2005; McGhan 2003; McGhan 2010; Persaud 1996; Splett 2006).

Four studies used administrative school records (Bartholomew 2006; Gerald 2006; Persaud 1996; Splett 2006). One study collected school absenteeism data from parents/guardians using tracking sheets (Cicutto 2013), and five studies used parental interviews or questionnaires (Cicutto 2013; Clark 2004; Howell 2005; McGhan 2003; McGhan 2010). In another study, school staff entered absence data into an intervention tracking system (Gerald 2009). Bruzzese 2011 was the only study that collected self‐reported absence data directly from children.

Bartholomew 2006 did not present disaggregated information, and we will not consider this study further here. Clark 2004 presented information on effectiveness of the intervention in terms of school absence in the form of a risk difference, which was not combined in the meta‐analyses, although researchers showed a significant intervention effect in reducing absences at three months and 12 months.

We included data from 10 studies in meta‐analysis models. Six of these studies considered long‐term impact of the intervention, with follow‐up data from nine months or longer collected and included in the meta‐analysis (Bruzzese 2011; Cicutto 2005; Cicutto 2013; Gerald 2009; McGhan 2003; McGhan 2010). However, three studies collected follow‐up data after three months or sooner (Persaud 1996; Howell 2005; Splett 2006), and one study provided unclear information on this (Gerald 2006). Differences in the exposure period over which absences were considered ranged from a year in three studies ‐ as in Cicutto 2005,Cicutto 2013, and McGhan 2010 ‐ to two weeks in one study ‐ as in Bruzzese 2011. Three studies considered any instance of recorded absence from school (Cicutto 2013; McGhan 2003; McGhan 2010), and the remaining seven studies measured mean number of days of absence or attendance at school. Most studies included in the meta‐analysis collected data on any form of absence, with only Gerald 2009 collecting data on absence related to asthma/respiratory illness.

Days of restricted activity

Three outcome evaluation studies reported days of restricted activity (Bruzzese 2011; Cicutto 2005; Cicutto 2013). One study used parent tracking sheets/diaries to record days of interrupted activity due to asthma (Cicutto 2013), another study used data from parent interviews (Cicutto 2005), and another study collected information directly from children (Bruzzese 2011). We included data from all three studies in the meta‐analyses, and all three studies collected data at 12 months' follow‐up. Two studies collected data on the mean number of days of restricted activity (Bruzzese 2011; Cicutto 2005), and Cicutto 2013 collected data on any instance of a day of restricted activity.

Secondary outcomes

Unplanned visit to a hospital or GP due to asthma symptoms

Five outcome evaluation studies reported on unplanned visits to a hospital or GP due to asthma symptoms (Bruzzese 2011; Cicutto 2013; McGhan 2003; McGhan 2010; Splett 2006). One study recorded unplanned visits using tracking sheets provided to parents (Cicutto 2013); two studies used a parental questionnaire (McGhan 2003; McGhan 2010); one study collected data directly from children (Bruzzese 2011); and a final study collected information on episodic asthma‐related visits to a school‐based health facility from administrative data (Splett 2006).

We included data from all five studies in the meta‐analyses. One study originally collected information on the mean number of unscheduled visits (Bruzzese 2011), and the remaining studies collected information on any instances of unscheduled visits to a medical provider (not captured in hospitalisation or ED utilisation data (above)). All studies collected data after substantial time had elapsed since the intervention began; this extended to nine to 12 months in four studies (Bruzzese 2011; Cicutto 2013; McGhan 2003; McGhan 2010), and in Splett 2006, longitudinal data collection occurred concurrently alongside delivery of the intervention over a period of six months.

Experience of daytime and night‐time symptoms

Nine outcome evaluation studies assessed children's experiences of daytime and night‐time symptoms (Atherly 2009; Bruzzese 2008; Bruzzese 2011; Clark 2004; Clark 2010; Howell 2005; McGhan 2003; Shah 2001; Velsor‐Friedrich 2005). These studies specifically reported on symptoms occurring during the day or during the night. Data were not combined in meta‐analyses for either Clark 2004 or Clark 2010. Clark 2004 collected data on daytime and night‐time symptoms as a risk difference, which indicated that the intervention had a positive effect in reducing daytime symptoms for all children but reduced the incidence of night‐time symptoms only for children with severe or persistent asthma (yielding a negative effect on night‐time symptoms for children with mild asthma). We did not include this in the meta‐analyses as it was incompatible with other units of analysis. Meanwhile, Clark 2010 collected information on a change in daytime symptoms, which indicated that the intervention had a positive, but non‐statistically significant, impact in terms of a drop in daytime symptoms (an effect size was extractable for one of the treatment arms only, although it was not used in meta‐analyses because of statistical and conceptual differences between post‐test data and changes in post‐test outcome data).

Among the seven studies included in the meta‐analysis, five studies reported on the incidence of daytime symptoms (Atherly 2009; Bruzzese 2008; Bruzzese 2011; Shah 2001; Velsor‐Friedrich 2005), and in the case of Shah 2001, researchers reported the incidence of daytime symptoms specifically occurring within school; four studies reported on night‐time awakenings (Bruzzese 2008; Bruzzese 2011; Howell 2005; McGhan 2003), with two studies reporting on both daytime and night‐time symptoms (Bruzzese 2008; Bruzzese 2011). Four studies reported on intervention effects six to 12 months after the intervention (Bruzzese 2011; McGhan 2003; Shah 2001; Velsor‐Friedrich 2005), and the remaining three studies included in the meta‐analyses information collected from children or parents two to three months post intervention. Similarly, data show a relatively even split between studies reporting on the mean level of asthma symptoms occurring in the daytime/at night‐time ‐ Atherly 2009,Bruzzese 2008,Bruzzese 2011,Howell 2005 ‐ and those focused on measuring any reported incidence of daytime/night‐time symptoms ‐ McGhan 2003, Shah 2001, and Velsor‐Friedrich 2005.

Lung function

Five outcome evaluation studies assessed lung function (Gerald 2009; Horner 2015; Patterson 2005; Shah 2001; Velsor‐Friedrich 2005), although studies measured this in different ways. One study assessed lung function using the peak expiratory flow rate (PEFR) and specifically focused on the occurrence of poor readings (red and yellow readings defined as less than 80% of best value) (Gerald 2009). A second study measured spirometry by measuring the percentage predicted change in forced expiratory volume in one second (FEV₁) (Patterson 2005). Shah 2001 reported forced vital capacity (FVC) before use of a bronchodilator. Velsor‐Friedrich 2005 measured peak flow increases as a percentage of pretest peak (i.e. change in peak flow); Horner 2015 measured airway inflammation by measuring exhaled nitric oxide as a biomarker of airway inflammation.

Because of conceptual differences in the outcomes collected, we did not combine these in meta‐analyses. Table 13 shows that the individual effects extracted exhibited considerable heterogeneity in the direction and magnitude of effect, confirming that meta‐analysis was not desirable due to statistical heterogeneity.

Table 13. Details of data transformations and adjustments made for meta‐analyses

Study	Indicator	Collection/reporting point	Mean cluster size (if applicable)	Intracluster correlation coefficient applied (if applicable)	Data transformation	Original effect size and standard error (with adjustment for clustering if applicable)	Final or transformed effect size and standard error (with adjustment for clustering if applicable)
Hospitalisations
Atherly 2009	Instances of hospitalisation in previous 4 weeks	Post intervention (3‐month follow‐up)	45.8	0.05	Yes – transformed from odds ratio to SMD	OR (0.7736); SE (lnOR) (1.385)	SMD (‐0.141); SE (0.764)
Bruzzese 2011	Hospitalisations in the past 2 months	Post intervention (12‐month follow‐up)	N/A	N/A	No	N/A	SMD (‐0.219); SE (0.120)
Clark 2005	Hospitalisations	Post intervention (12‐month follow‐up)	Deemed that analysis methods accounted for clustering	Deemed that analysis methods accounted for clustering	Yes – transformed from odds ratio to SMD	OR (1.43); SE (estimated from P value (lnOR)) 0.39	SMD (‐0.197); SE (0.215)
Gerald 2006	Median hospitalisations (not combined)	N/A	N/A	N/A	N/A	N/A	N/A
Horner 2008	Any hospital stays in the past 12 months (based on parents reporting any stay)	Post intervention (7‐month follow‐up)	10.1 (reported by study authors)	0.05	Yes – transformed from odds ratio to SMD	OR (0.882); SE (lnOR) (0.791)	SMD (‐0.069); SE (0.436)
Horner 2015	Mean number of hospitalisations since the previous data collection (at 8 months)	Post intervention (12‐month follow‐up)	8.9 (approx.)	0.05	No	N/A	SMD (‐0.057); SE (0.169)
Levy 2006	Mean hospital days	Post test (at intervention end)	17.36	0.05	No	N/A	SMD (‐0.293); SE (0.174)
Emergency department visits
Atherly 2009	Instances of ED visits in previous 4 weeks	Post intervention (3‐month follow‐up)	45.8	0.05	No	N/A	OR (1.036); SE (lnOR) (0.916)
Bruzzese 2011	Mean ED visits in the past 2 months	Post intervention (12‐month follow‐up)	N/A	N/A	Yes ‐ transformed from SMD to OR	SMD (‐0.289); SE (0.120)	OR (0.592); SE (lnOR) (0.218)
Cicutto 2005	ED visits in the past year	Post intervention (12‐month follow‐up)	9.85	0.05	No	N/A	OR (0.697); SE (lnOR) (0.407)
Cicutto 2013	ED visits in the past year (reports of)	Post intervention (12‐month follow‐up)	7.7	0.05	No	N/A	OR (0.318); SE (lnOR) (0.317)
Clark 2005	ED visits	Post intervention (12‐month follow‐up)	Deemed that analysis methods accounted for clustering	Deemed that analysis methods accounted for clustering	No, but see notes	N/A	OR (1.002); SE (estimated from P value (lnOR)) 0.072 Note that the OR was reported as 1.00 in the paper with a P value of 0.98. So information could be used and an SE extracted, a small correction to an OR of 1.002 was applied
Gerald 2006	Median ED visits (not combined)	N/A	N/A	N/A	N/A	N/A	N/A
Horner 2008	Any ED visits in the past 12 months (based on parents reporting any stay)	Post intervention (7‐month follow‐up)	10.1 (reported by study authors)	0.05	No	N/A	OR (0.857); SE (lnOR) (0.461)
Horner 2015	Mean number of ED visits since the previous data collection (8 months)	Post intervention (12‐month follow‐up)	8.9 (approx.)	0.05	Yes ‐ transformed from SMD to OR	SMD (0); SE (0.169)	OR (1.00); SE (0.306)
Howell 2005	Mean number of ED visits in the past 6 weeks	Post intervention (3‐month follow‐up)	4.25	0.05	Yes ‐ transformed from SMD to OR	SMD (‐0.331); SE (0.578)	OR (0.549); SE (1.049)
Levy 2006	Mean urgent care or emergency visits	Post test (at intervention end)	17.36	0.05	Yes ‐ transformed from SMD to OR	SMD (‐0.286); SE (0.174)	OR (0.595); SE (0.318)
McGhan 2003	ED visits in the past year (any)	Post intervention (9‐month follow‐up)	9	0.05	No	N/A	OR (1.283); SE (lnOR) (0.649)
McGhan 2010	ED visits in the past year (any)	Post intervention (12‐month follow‐up)	8.3	0.05	No	N/A	OR (2.64); SE (lnOR) (0.707)
Persaud 1996	Children with ED Visits (20‐week period post intervention)	Post intervention (events in 20‐week period post intervention)	N/A	N/A	No	N/A	OR (0.286); SE (lnOR) (0.737)
Velsor‐Friedrich 2005	Urgent doctor visits (any in the past 12 months)	Post intervention (12‐month follow‐up)	13	0.05	No	N/A	OR (0.683); SE (lnOR) (0.933)
Absence from school
Bruzzese 2011	Mean self‐reported absence in past 2 weeks (any absence)	Post intervention (12‐month follow‐up)	N/A	N/A	No	N/A	SMD (‐0.382); SE (0.121)
Cicutto 2005	Parent‐reported absence (any absence) over a year	Post intervention (12‐month follow‐up)	9.85	0.05	No	N/A	SMD (‐0.256); SE (0.151)
Cicutto 2013	Parent‐reported absence (any absence) over a year	Post intervention (12‐month follow‐up)	7.7	0.05	Yes – transformed from odds ratio to SMD	OR (0.660); SE (lnOR) (0.129)	SMD (‐0.229); SE (0.071)
Gerald 2006	Absences recorded on school records	Post test (unclear duration)	Clustering accounted for in analytical strategy	Clustering accounted for in analytical strategy	No	N/A	SMD (‐0.199); SE (0.084)
Gerald 2009	Absence from school due to respiratory illness/asthma *December measure used	Post intervention (15‐month follow‐up)	N/A	N/A	Yes – transformed from odds ratio to SMD	OR (1.1667); SE (lnOR) (0.364)	SMD (0.085); SE (0.227)
Howell 2005	School days missed in past 6 weeks	Post intervention (3‐month follow‐up)	3.25	0.05	No	N/A	SMD (0.152); SE (0.635)
McGhan 2003	Any missed school days	Post intervention (9‐month follow‐up)	9	0.05	Yes – transformed from odds ratio to SMD	OR (0.720); SE (lnOR) (0.413)	SMD (‐0.181); SE (0.227)
McGhan 2010	(No) Missed school days (any) over past 12 months	Post intervention (12‐month follow‐up)	8.3	0.05	Yes – transformed from odds ratio to SMD	OR (0.640); SE (lnOR) (0.353)	SMD (0.246); SE (0.195) (note: inverse taken as the intervention favours control)
Persaud 1996	Mean school days of absence based on school records	Post intervention (immediately afterwards)	N/A	N/A	No	N/A	SMD (‐0.236); SE (0.335)
Splett 2006	Mean percentage of days attended	Post intervention (12‐month follow‐up)	Deemed that analysis methods accounted for clustering	Deemed that analysis methods accounted for clustering	No	N/A	SMD (0.019); SE (0.051)
Days of restricted activity
Bruzzese 2011	Mean self‐reported days of restricted activity in past 2 weeks	Post intervention (12‐month follow‐up)	N/A	N/A	No	N/A	SMD (‐0.349); SE (0.120)
Cicutto 2005	Days of limited activity due to asthma	Post intervention (12‐month follow‐up)	9.85	0.05	No	N/A	SMD (‐0.318); SE (0.151)
Cicutto 2013	Percentage of students reporting days of restricted activity	Post intervention (12‐month follow‐up)	7.7	0.05	Yes – transformed from odds ratio to SMD	OR (0.612); SE (lnOR) (0.130)	SMD (‐0.271); SE (0.072)
Unplanned visits to medical providers
Bruzzese 2011	Mean acute care visits in the past 2 months	Post intervention (12‐month follow‐up)	N/A	N/A	Yes – transformed from SMD to OR	SMD (‐0.283); SE (0.120)	OR (0.598); SE (0.217)
Cicutto 2013	Unscheduled care in the past year (reports of)	Post intervention (12‐month follow‐up)	7.7	0.05	No	OR (0.703); SE (lnOR) (0.143)	SMD (‐0.194); SE (0.079)
McGhan 2003	Any unscheduled doctor visits	Post intervention (9‐month follow‐up)	9	0.05	No	OR (0.886); SE (lnOR) (0.426)	SMD (‐0.067); SE (0.235)
McGhan 2010	Unscheduled GP visits (any) over past 12 months	Post intervention (12‐month follow‐up)	8.3	0.05	No	OR (1.169); SE (lnOR) (0.397)	SMD (0.086); SE (0.219)
Splett 2006	Episodic asthma visits to school health office (over 6 months following start of intervention)	Over 6 months following start of intervention	97.6	0.05	No	OR (0.913); SE (lnOR) (0.282)	SMD (‐0.046); SE (0.156)
Daytime symptoms
Atherly 2009	Mean number of days with asthma symptoms	Post intervention (3‐month follow‐up)	45.8	0.05	No	N/A	SMD (‐0.026); SE (0.168)
Bruzzese 2008	Mean days last 2 weeks with asthma symptoms	Post intervention (2‐month follow‐up)	N/A	N/A	No	N/A	SMD (‐0.151); SE (0.418)
Bruzzese 2011	Mean days last 2 weeks with asthma symptoms	Post intervention (12‐month follow‐up)	N/A	N/A	No	N/A	SMD (‐0.210); SE (0.120)
Shah 2001	Number of students reporting attacks in school at follow‐up	Post intervention (6‐month follow‐up)	41.8	0.05	Yes – transformed from odds ratio to SMD	OR (0.647); SE (lnOR) (0.488)	SMD (‐0.240); SE (0.269)
Velsor‐Friedrich 2005	Symptom days in past 2 weeks	Post intervention (12‐month follow‐up)	13	0.05	Yes – transformed from odds ratio to SMD	OR (0.846); SE (lnOR) (0.705)	SMD (‐0.030); SE (0.413)
Night‐time symptoms
Bruzzese 2008	Mean nights woken last 2 weeks with asthma symptoms	Post intervention (2‐month follow‐up)	N/A	N/A	No	N/A	SMD (‐0.433); SE (0.423)
Bruzzese 2011	Mean self‐reported night‐time awakenings in past 2 weeks	Post intervention (12‐month follow‐up)	N/A	N/A	No	N/A	SMD (‐0.388); SE (0.121)
Howell 2005	Mean number of night‐time awakenings in past 6 weeks	Post intervention (3‐month follow‐up)	4.25	0.05	No	N/A	SMD (0.253); SE (0.478)
McGhan 2003	Waking up in past 2 weeks twice or more	Post intervention (9‐month follow‐up)	9	0.05	Yes – transformed from odds ratio to SMD	OR (1.237); SE (lnOR) (0.412)	SMD (0.117); SE (0.227)
Use of reliever therapies
Gerald 2009	Rescue medication use over twice per week *November measure used	Post intervention (15‐month follow‐up)	N/A	N/A	N/A	OR (0.228); SE (lnOR) (0.582)	N/A
McGhan 2003	Number of students with appropriate use of reliever medication	Post intervention (9‐month follow‐up)	9	0.05	N/A	OR (3.48); SE (lnOR) (0.565)	N/A
McGhan 2010	Used short‐acting bronchodilators in past 2 weeks	Post intervention (12‐month follow‐up)	8.3	0.05	N/A	OR (0.878); SE (lnOR) (0.356)	N/A
Splett 2006	Students with access to reliever medication visiting health office (over 6 months following start of intervention) *Note low levels of children with reliever medication	Over 6 months following start of intervention	97.6	0.05	N/A	OR (1.28); SE (lnOR) (0.282)	N/A
Use of corticosteroids and/or use of add‐on therapies
Bruzzese 2011	Use of controller medication	Post intervention (12‐month follow‐up)	N/A	N/A	No	N/A	OR (1.451); SE (lnOR) (0.240)
Horner 2015	Inhaled corticosteroid adherence	Post intervention (5‐month follow‐up)	8.9	0.05	No	N/A	SMD (‐0.605); SE (0.173)
Howell 2005	Inhaled corticosteroid adherence as prescribed (during past week)	Post intervention (3‐month follow‐up)	4.25	0.05	No	N/A	SMD (0.953); SE (0.546)
McGhan 2003	Currently using inhaled steroids	Post intervention (9‐month follow‐up)	9	0.05	No	N/A	OR (1.112); SE (lnOR) (0.418)
McGhan 2010	Currently using inhaled steroids	Post intervention (12‐month follow‐up)	8.3	0.05	No	N/A	OR (0.962); SE (lnOR) (0.376)
Splett 2006	Students with access to controller medication visiting health office (over 6 months following start of intervention) *Note low levels of children with controller medication	Over 6 months following start of intervention	97.6	0.05	N/A	OR (1.703); SE (lnOR) (0.806)	SMD (0.293); SE (0.445)
Lung function
Gerald 2009	Poor peak flow measures (red/amber readings)	Post‐intervention (15‐month follow‐up)	N/A	N/A	No	OR (0.94); SE (lnOR) (0.334)
Horner 2015	Airway inflammation (exhaled nitric oxide, collected using the single‐use RTube collection device, was the biomarker of airway inflammation)	Post intervention (12‐month follow‐up)	8.9	0.05	No	N/A	SMD (‐0.011); SE (0.169)
Shah 2001	Forced expiratory volume in 1 second: forced vital capacity before bronchodilator	Post intervention (3‐month follow‐up)	Deemed that analysis methods accounted for clustering	Deemed that analysis methods accounted for clustering	No	N/A	SMD (0.074); SE (0.127)
Patterson 2005	Forced expiratory volume in 1 second (% predicted change)	Post intervention (2‐month follow‐up)	Deemed that analysis methods accounted for clustering	Deemed that analysis methods accounted for clustering	No	N/A	SMD (‐0.05); SE (0.177)
Velsor‐Friedrich 2005	Peak flow increases as a percentage of pretest peak flow (change)	Post intervention (12‐month follow‐up)	13	0.05	No	N/A	SMD (‐5.905); SE (0.839)
Quality of life						Mean difference (QoL only)	Standardised mean difference (QoL only)
Al‐Sheyab 2012	Arabic version of the Pediatric Asthma Quality of Life Questionnaire (PAQLQ) *because of uncertainty about SD values, derived from t/P value of difference between means	Post intervention (3‐month follow‐up)	Deemed that analysis methods accounted for clustering	Deemed that analysis methods accounted for clustering	No	MD 1.35 (CI 0.96 to 1.74)	SMD (0.299); SE (0.129)
Cicutto 2005	Juniper Pediatric Asthma Quality of Life Questionnaire overall quality of life	Post intervention (2‐month follow‐up)	9.85	0.05	No	MD 0.50 (CI 0.00 to 1.00)	SMD (0.356); SE (0.151)
Cicutto 2013	Juniper Pediatric Asthma Quality of Life Questionnaire overall quality of life	Post intervention (12‐month follow‐up)	7.7	0.05	No	MD 0.40 (CI 0.21 to 0.59)	SMD (0.308); SE (0.064)
Henry 2004	Juniper Pediatric Asthma Quality of Life Questionnaire overall quality of life	Post intervention (6‐month follow‐up)	15.2	0.05	No	MD 0.16 (CI ‐0.22 to 0.54)	SMD (0.128); SE (0.114)
Horner 2008	Juniper Pediatric Asthma Quality of Life Questionnaire overall quality of life	Post intervention (7‐month follow‐up)	10.2	0.05	No	MD 0.05 (CI ‐0.21 to 0.31)	SMD (0.083); SE (0.196)
Howell 2005	Juniper Pediatric Asthma Quality of Life Questionnaire overall quality of life	Post intervention (3‐month follow‐up)	6	0.05	No	MD 0.03 (CI ‐1.71 to 1.77)	SMD (0.020); SE (0.484)
Kintner 2009	Quality of life is defined through the Participation in Life Activities Scale	Immediately post intervention	Deemed that analysis methods accounted for clustering	Deemed that analysis methods accounted for clustering	No	N/A	SMD (0.583); SE (0.263)
Patterson 2005	Change in Juniper Pediatric Asthma Quality of Life Questionnaire overall quality of life	Change in quality of life between baseline and 4 months post intervention	Deemed that analysis methods accounted for clustering	Deemed that analysis methods accounted for clustering	No	MD 0.07 (CI ‐0.26 to 0.40)	N/A
Shah 2001	Juniper Pediatric Asthma Quality of Life Questionnaire overall quality of life; percentage of students with clinically significant improvement	Post intervention (3‐month follow‐up)	Deemed that analysis methods accounted for clustering	Deemed that analysis methods accounted for clustering	No	MD 0.09 (CI ‐0.23 to 0.41)	N/A
Withdrawal
Al‐Sheyab 2012	Withdrew between baseline and outcome collection	Post intervention (3‐month follow‐up)	65.25	0.05	No	N/A	OR (0.511); SE (lnOR) (1.074)
Bartholomew 2006	Lost to follow‐up at post‐test measure	Post intervention (duration unclear)	11.2	0.05	No	N/A	OR (0.237); SE (lnOR) (0.145)
Bruzzese 2008	Withdrew between baseline and outcome collection	Immediate post intervention	N/A	N/A	No	N/A	OR (0.307); SE (lnOR) (1.683)
Bruzzese 2011	Withdrew between baseline and outcome collection	Post intervention (12‐month follow‐up)	N/A	N/A	No	N/A	OR (1.313); SE (lnOR) (0.279)
Cicutto 2005	Withdrew between baseline and outcome collection	Post intervention (6‐month follow‐up)	9.85	0.05	No	N/A	OR (1.788); SE (lnOR) (0.629)
Gerald 2009	Withdrew between baseline and outcome collection	Post intervention (6‐month follow‐up)	N/A	N/A	No	N/A	OR (1.788); SE (lnOR) (0.613)
Horner 2008	Withdrew between baseline and outcome collection	Post intervention (7‐month follow‐up)	10.2	0.05	No	N/A	OR (1.333); SE (lnOR) (0.531)
Horner 2015	Failed to complete final data collection	Post intervention (12‐month follow‐up)	8.9	0.05	No	N/A	OR (0.75); SE (lnOR) (0.486)
Kintner 2009	Withdrew during intervention and between end of intervention and follow‐up	Post intervention (12‐month follow‐up)	13.2	0.05	No	N/A	OR (30.176); SE (lnOR) (1.860)
Levy 2006	Failure to complete outcome evaluation	Post intervention (12‐month follow‐up)	17.36	0.05	No	N/A	OR (0.357); SE (lnOR) (0.3881)
McGhan 2003	Withdrew between baseline and outcome collection	Post intervention (9‐month follow‐up)	9	0.05	No	N/A	OR (1.147); SE (lnOR) (0.5381)
McGhan 2010	Withdrew between baseline and interim outcome collection	Post intervention (6‐month follow‐up)	8.3	0.05	No	N/A	OR (1.007); SE (lnOR) (0.387)
Patterson 2005	Withdrew during intervention	Post intervention – immediately following intervention	7.95	0.05	No	N/A	OR (5.675); SE (lnOR) (1.087)
Shah 2001	Withdrew between baseline and outcome collection	Post intervention (3‐month follow‐up)	45.3	0.05	No	N/A	OR (1.343); SE (lnOR) (0.475)

CI: confidence interval.

ED: emergency department.

lnOR: log odds ratio.

MD: mean difference.

N/A: not applicable.

OR: odds ratio.

PAQLQ: Pediatric Asthma Quality of Life Questionnaire.

QoL: quality of life.

SD: standard deviation.

SE: standard error.

SMD: standardised mean difference.

Use of reliever therapies such as beta₂‐agonists

Four outcome evaluation studies assessed use of reliever therapies (Gerald 2009; McGhan 2003; McGhan 2010; Splett 2006). We combined in meta‐analyses two studies that reported on the use of rescue medication and short‐acting bronchodilators (SABAs), respectively (Gerald 2009; McGhan 2010). The former captured information on instances when rescue medication was used more than twice a week, and the latter measured any instance in which rescue medication was used; these studies sought to measure long‐term intervention effects at 12 months ‐ as in McGhan 2010 ‐ and at 15 months ‐ as in Gerald 2009. The remaining two studies measured appropriate use of reliever medication and access to reliever medication, respectively (McGhan 2003; Splett 2006). Because of conceptual differences in the way in which researchers measured use of reliever therapies, we chose not to meta‐analyse this information. We have presented information provided by all four studies in Table 13.

Corticosteroid dosage and/or use of add‐on therapies

Six studies measured corticosteroid usage and dosage (Bruzzese 2011; Horner 2015; Howell 2005; McGhan 2003; McGhan 2010; Splett 2006). One study measured whether children had access to controller medication while visiting the school health office (Splett 2006). Two studies measured whether children were adhering to guidance provided around the correct use of corticosteroid (Horner 2015; Howell 2005), and three studies measured any reported usage of corticosteroid or controller medication (Bruzzese 2011; McGhan 2003; McGhan 2010). We meta‐analysed data from these five studies separately, as adherence was deemed to conceptually differ from reports of usage. Horner 2015 and Howell 2005 included information from children at five months and three months, respectively, in meta‐analyses of corticosteroid adherence. All three studies in the second meta‐analysis on reported instances of corticosteroid or controller medication usage collected information at nine months or 12 months post intervention. We have presented data from all six studies in Table 13.

Health‐related quality of life (HRQoL)

Twelve outcome evaluation studies measured quality of life (Al‐Sheyab 2012; Cicutto 2005; Cicutto 2013; Clark 2010; Henry 2004; Horner 2008; Howell 2005; Kintner 2009; McCann 2006; McGhan 2010; Patterson 2005; Shah 2001). McCann 2006,McGhan 2010, and Clark 2010 did not present data in an extractable format (i.e. described data narratively, did not disaggregate data, or did not include the necessary information to extract an effect size); Patterson 2005 measured change in quality of life; and Shah 2001 measured clinically significant improvements (see Table 13). Among the nine studies that calculated an effect size, eight were based on the Juniper Pediatric Asthma Quality of Life Questionnaire overall quality of life (see Juniper 1996); Al‐Sheyab 2012 used an Arabic version of this questionnaire. Kintner 2009 measured quality of life by reviewing responses to the Participation in Life Activities Scale.

We constructed two sets of meta‐analyses for a model measuring changes in quality of life. One of these used SMD to calculate effect sizes; this allowed us to incorporate data from Kintner 2009. We meta‐analysed change scores to obtain an MD from the data reported in Patterson 2005 and Shah 2001. Therefore data from six studies were common to both models. Several studies measured quality of life within four months of the intervention (Al‐Sheyab 2012; Cicutto 2005; Howell 2005; Kintner 2009; Patterson 2005; Shah 2001), two studies collected data at six to seven months after the intervention (Henry 2004; Horner 2008), and one study collected data 12 months after the intervention (Cicutto 2013).

Withdrawal from the study

Researchers frequently presented withdrawal data, although not always in a format that allowed extraction of data to form an effect size. This often occurred because studies reported overall numbers lost during the study without disaggregating by treatment arm (Cicutto 2013; Velsor‐Friedrich 2005), or because studies reported no losses (Persaud 1996). Fourteen studies provided enough data to allow calculation of an effect size (OR) (Al‐Sheyab 2012; Bartholomew 2006; Bruzzese 2008; Bruzzese 2011; Cicutto 2005; Gerald 2009; Horner 2008; Horner 2015; Kintner 2009; Levy 2006; McGhan 2003; McGhan 2010; Patterson 2005; Shah 2001). Few studies reported on active withdrawal processes occurring during the intervention; instead investigators reported on failure to collect children's data at follow‐up (collected from children and parents). Researchers collected data at different points between intervention and follow‐up, including at four months or less (Al‐Sheyab 2012; Bruzzese 2008; Patterson 2005; Shah 2001), at six to seven months (Cicutto 2005; Gerald 2009; Horner 2008; McGhan 2010), and at nine to 12 months (Bruzzese 2011; Horner 2015; Kintner 2009; Levy 2006; McGhan 2003). Duration was unclear in one study (Bartholomew 2006).

Excluded studies

From the title and abstract screening, we excluded 28,318 records because they were clearly outside the remit of the review of process evaluations. Following full‐text screening, we excluded another 1029 records, for reasons detailed in the PRISMA diagram (Figure 2).

Based on title and abstract screening, we excluded 274 records as they were outside the remit of the review of outcome evaluation studies. Following full‐text screening, we excluded 67 additional records, for reasons detailed in the PRISMA diagram (Figure 3).

Risk of bias in included studies

We have displayed results of the risk of bias assessment for process and outcome evaluation studies in the risk of bias table and graph. We have presented the agreed judgement of two review authors (DK, KH) regarding the risk of bias for each included study as percentages for each bias item in the risk of bias graph (Figure 4; Figure 5).

Figure 4

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 5

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Risk of bias ‐ process evaluation studies

For process evaluation studies, we assessed risk of bias using a combination of two tools. The first tool was developed at the EPPI‐Centre (Harden 2004) to assess the methodological rigour of 'views' studies; the second tool, which was developed by the EPPI‐Centre to assess the quality of process evaluation data (O'Mara‐Eves 2013),

We assessed reporting quality across five indicators.

Transparent and clearly stated aims (0 high risk of bias, 27 low risk of bias, 6 unclear risk).
Explicit theories underpinning the intervention (10 high risk of bias, 14 low risk of bias, 9 unclear risk).
Transparent and clearly stated methods and tools (4 high risk of bias, 17 low risk of bias, 12 unclear risk).
Selective reporting (10 high risk of bias, 8 low risk of bias, 15 unclear risk).
Harmful effects (8 high risk of bias, 5 low risk of bias, 20 unclear risk).

We assessed population and selection factors using four indicators.

Population and sample described well (8 high risk of bias, 8 low risk of bias, 17 unclear risk).
Continuous evaluation (3 high risk of bias, 8 low risk of bias, 22 unclear risk).
Evaluation participation equity and sampling (9 high risk of bias, 7 low risk of bias, 17 unclear risk).
Design and methods overall approach (6 high risk of bias, 10 low risk of bias, 16 unclear risk).

We assessed reliability and transferability of findings using two indicators.

Reliability of findings and recommendations (11 high risk of bias, 8 low risk of bias, 14 unclear risk).
Transferability of findings (13 high risk of bias, 5 low risk of bias, 15 unclear risk).

Overall, process evaluation studies consisted of 10 high‐risk studies, five low‐risk studies, and 18 studies at unclear risk.

Risk of bias ‐ outcome evaluation (RCT) studies

Allocation

We judged 14 outcome evaluation studies to be at low risk of bias for random sequence generation (Al‐Sheyab 2012; Bruzzese 2011; Cicutto 2005; Cicutto 2013; Clark 2004; Clark 2010; Gerald 2009; Horner 2008; Kintner 2009; McGhan 2010; Patterson 2005; Shah 2001; Splett 2006; Srof 2012). We judged three to be at high risk (Henry 2004; Mosnaim 2011; Pulcini 2007). We judged the remainder to be at unclear risk. We judged six of these studies to be at low risk of allocation concealment bias (Bruzzese 2011; Cicutto 2005; Cicutto 2013; Gerald 2009; Shah 2001; Splett 2006). We judged nine studies to be at high risk of allocation concealment bias (Clark 2010; Horner 2008; Howell 2005; Kintner 2009; Levy 2006; McGhan 2010; Mosnaim 2011; Pulcini 2007; Velsor‐Friedrich 2005).

Blinding

We judged three outcome evaluation studies to be at low risk of bias for blinding of participants and personnel (Cicutto 2013; Horner 2015; Levy 2006). We judged two studies to be at high risk of bias for this component (Horner 2008; Kintner 2009). We judged seven outcome evaluation studies to be at low risk for blinding of outcome assessment (Bruzzese 2011; Cicutto 2005; Cicutto 2013; Horner 2015; Kintner 2009; Levy 2006; Persaud 1996). For two outcome evaluation studies, we determined that risk of bias for blinding of outcome assessment was high (Clark 2010; Srof 2012).

Results of synthesis ‐ part 1: qualitative comparative analysis of determinant conditions for successful intervention implementation

Descriptive results from process evaluation studies on implementation success

Across the 27 included studies, review authors identified eight studies as having high implementation scores for our combined outcome (attrition, adherence, dosage) and classified these studies as mainly or fully included in a set of studies marked as successfully implemented (Al‐Sheyab 2012a; Berg 2004; Bruzzese 2008; Bruzzese 2011; Henry 2004; Joseph 2010; Kintner 2012; Terpstra 2012). In contrast, we identified eight studies as having low implementation success scores and as mainly or entirely outside the successfully implemented set of studies (Brasler 2006; Bruzzese 2004; Gerald 2006; Howell 2005; Kouba 2012; Langenfeld 2010; Magzamen 2008; Spencer 2000). Other studies were more ambiguous regarding their implementation success and had high levels of missing data or conflicting results across indicators.

For many studies reporting lower implementation success, we viewed the difficulty of incorporating an intervention into the busy school curriculum and into children's busy schedules as undermining the intervention (Brasler 2006; Bruzzese 2004; Gerald 2006; Howell 2005; Kouba 2012). Additional factors included difficulties in terms of high staff turnover (Gerald 2006); high child turnover and/or chaotic families (Brasler 2006; Howell 2005); and low motivation among children, particularly in the absence of incentives (Magzamen 2008). Similarly, researchers provided a diverse set of explanations for successful implementation, including high levels of school‐level commitment (Henry 2004; Kintner 2012); high levels of child and teacher motivation (Al‐Sheyab 2012a; Berg 2004); and development of group cohesion (Bruzzese 2008), as well as specific intervention design features, including tailoring of messages to children, as in Bruzzese 2011 and Joseph 2010, and additional communications with parents, as in Terpstra 2012.

In the QCA analyses below, we examine factors that could further explain successful implementation by examining which characteristics are shared among studies that were successfully implemented, and whether these differ from studies that were not successfully implemented.

Summary of results from qualitative comparative analysis

We first explored different domains of implementation separately, before bringing this evidence together in a final model (Table 14). We used this strategy mainly because of the problem of limited diversity, by which observed studies did not support too many possible combinations of intervention characteristics. We found no configurations of characteristics that consistently triggered successful implementation with respect to recruitment and retention, as well as pedagogical factors, although these may be important in other ways for children's outcomes.

Table 14. Summary of interventions, conditions entered, and model results

Domain (model)	Conditions entered	Sufficient configurations identified that trigger successful implementation
1. Setting and participant features	School health centre; high school; parents direct intervention recipients; teachers direct intervention recipients; school nurses/others direct intervention recipients	Yes
2. Recruitment and retention processes	Additional marketing materials; provision of incentives; provision of catch‐up sessions; provision of reminders	No
3. Curriculum, pedagogy, and intervention emphasis	Focus on establishing alliances with care providers; focus on asthma symptom recognition and management; tailored content; emphasis on personal responsibility; interactive pedagogical style; diverse pedagogical style	No
4. Modifiable intervention processes	Theory driven; run in class time; run in students' free time; school nurse key role in delivery or teaching; personalised or individual 1‐to‐1 instruction	Yes
5. Stakeholder engagement	School asthma policy; child satisfaction; teachers engaged/relationships developed; parents engaged/relationships developed; school nurses engaged/relationships developed	Yes
6. Consolidated model	Theory driven; run in students' free time; child satisfaction; parents engaged/relationships developed; high school	Yes

In our consolidated model, we prioritised conditions that were included in configurations with high consistency and coverage scores. To facilitate interpretation in the consolidated model, we focused on conditions with a consistent direction. Working from the raw data (Table 15), we created a truth table (Table 16), which showed the extent to which sets of studies with particular configurations of conditions overlapped with a set of studies included in our successful intervention set. Boolean minimisation helped to simplify the solution (Table 17), and we inserted assumptions about logical remainders (configurations with no observed cases) to further simplify the solution (Table 18). After doing this, we observed that four pathways (or configurations of conditions) triggered the outcome, thereby forming our 'solution' (summarised in Table 19).

Table 15. Data table for QCA model 6 ‐ consolidated model

	Successful intervention	High school	Child satisfaction	Theory driven	Intervention takes place during students' own free time	Good relationships/engagement with parents
Joseph 2010	0.52	1	0	1	0.33	0
Kouba 2012	0.33	1	0	1	1	0
Dore‐Stites 2007	0.67	0	1	1	0.33	0.75
Joseph 2013	1.00	1	0	1	0.75	1
Mujuru 2011	0.67	0	0	0	0	0.25
Henry 2004	0.83	1	0	0	0	0
Pike 2011	0.67	0	0	0	0	0
Spencer 2000	0.33	0	0	0	0.33	1
Engelke 2013	0.50	0.5	0	0	0.33	1
Splett 2006	0.50	0.5	0	0	0.33	0
Kintner 2012	0.83	1	1	1	1	0.25
Berg 2004	0.83	1	1	1	0.33	0
Howell 2005	0.33	0	0.633333	1	0.33	0.75
Gerald 2006	0.33	0	0	0	0.33	0
Langenfeld 2010	0.33	0	0	0	0.33	0
Al‐Sheyab 2012	0.83	1	0.633333	1	0.33	0
Levy 2006	0.52	0	0	0	0.33	0
Terpstra 2012	1.00	0.66	0	1	1	0.25
Horner 2015	0.67	0	0	1	1	0
Bruzzese 2008	0.94	0.66	1	1	0.33	1
Lee 2011	0.50	0	0	1	0	0
Bruzzese 2004	0.33	1	0.633333	1	0.75	0
Cicutto 2013	0.67	0	0	1	1	0
Brasler 2006	0.00	0.66	0.633333	0	0.75	0
Crane 2014	0.50	0	0	1	1	0
Bruzzese 2011	0.88	1	0	1	0.33	0
Magzamen 2008	0.19	0.75	0	0	1	0

QCA: qualitative comparative analysis.

Table 16. Truth table for QCA model 6 ‐ consolidated model

High school	Child satisfaction	Theory driven	Intervention takes place during students' own free time	Good relationships/ engagement with parents	Outcome code (based on consistency score)	Number of studies with membership in causal combination > 0.5	Consistency score with subset relationship (n = 27 in each assessment)	Proportional reduction in inconsistency	Cases
1	1	1	0	0	1	2	1	1	Al‐Sheyab 2012; Berg 2004
1	0	1	1	1	1	1	1	1	Joseph 2013
1	1	1	0	1	1	1	1	1	Bruzzese 2008
1	0	1	0	0	1	2	0.924	0.841	Bruzzese 2011; Joseph 2010
1	1	1	1	0	1	2	0.853	0.752	Bruzzese 2004; Kintner 2012
0	1	1	0	1	1	2	0.815	0.668	Dore‐Stites 2007; Howell 2005
1	0	1	1	0	0	2	0.768	0.595	Kouba 2012; Terpstra 2012
0	0	0	0	1	0	1	0.763	0	Engelke 2013; Spencer 2000
1	0	0	0	0	0	1	0.762	0.615	Henry 2004
0	0	1	1	0	0	3	0.675	0.463	Cicutto 2013; Crane 2014; Horner 2015
0	0	0	0	0	0	5	0.67	0.322	Gerald 2006; Langenfeld 2010; Levy 2006; Mujuru 2011; Pike 2011; Splett 2006
0	0	1	0	0	0	1	0.6	0	Lee 2011
1	0	0	1	0	0	1	0.358	0	Magzamen 2008
1	1	0	1	0	0	1	0	0	Brasler 2006

QCA: qualitative comparative analysis.

Table 17. Complex solution for QCA model 6 ‐ consolidated model

		Consistency score with subset relationship (n = 27 in each assessment)	Proportional reduction in inconsistency	Raw coverage	Unique coverage	Cases
1	CHILDSATTHEORYDRIVENruninstudenttime*GOODRELPAR	0.846	0.756	0.106	0.106	Bruzzese 2008; Dore‐Stites 2007; Howell 2005
2	HIGHSCHOOLCHILDSATTHEORYDRIVEN*goodrelpar	0.845	0.786	0.162	0.063	Al‐Sheyab 2012; Berg 2004; Bruzzese 2004; Kintner 2012
3	HIGHSCHOOLTHEORYDRIVENruninstudenttime*goodrelpar	0.949	0.914	0.177	0.078	Al‐Sheyab 2012; Berg 2004; Bruzzese 2011; Joseph 2010
4	HIGHSCHOOLchildsatTHEORYDRIVENRUNINSTUDENTTIMEGOODRELPAR	1	1	0.064	0.064	Joseph 2013
	M1	0.875	0.823	0.41

QCA: qualitative comparative analysis.

[Notation: Upper case = condition is present; Lower case = condition is absent; * = logical and; + logical or; Key: HIGHSCHOOL = High School (lower case not in high school); THEORYDRIVEN = Authors explicitly named theory or presented conceptual model for intervention; RUNINSTUDENTTIME = Substantial component run in students' own time (e.g. lunchtime); GOODRELPAR = Good level of reported in engagement and/or developing relationships with parents; CHILDSAT = Children reported as satisfied; SUCCESSFULIMPLEMENTATION = Implementation of intervention successful]

Table 18. Intermediate solution for QCA model 6 ‐ consolidated model

		Consistency score with subset relationship (n = 27 in each assessment)	Proportional reduction in inconsistency	Raw coverage	Unique coverage	Cases
1	HIGHSCHOOLCHILDSATTHEORYDRIVEN	0.839	0.791	0.21	0.053	Al‐Sheyab 2012; Berg 2004; Bruzzese 2004; Bruzzese 2008; Kintner 2012
2	HIGHSCHOOLTHEORYDRIVENGOODRELPAR	1	1	0.138	0.064	Bruzzese 2008; Joseph 2013
3	HIGHSCHOOLTHEORYDRIVENruninstudenttime	0.961	0.942	0.235	0.078	Al‐Sheyab 2012; Berg 2004; Bruzzese 2008; Bruzzese 2011; Joseph 2013
4	CHILDSATTHEORYDRIVENruninstudenttime*GOODRELGPAR	0.846	0.756	0.106	0.064	Bruzzese 2008; Dore‐Stites 2007; Howell 2005
	M1	0.862	0.81	0.432

QCA: qualitative comparative analysis.

[Notation: Upper case = condition is present; Lower case = condition is absent; * = logical and; + logical or; Key: HIGHSCHOOL = High School (lower case not in high school); THEORYDRIVEN = Authors explicitly named theory or presented conceptual model for intervention; RUNINSTUDENTTIME = Substantial component run in students' own time (e.g. lunchtime); GOODRELPAR = Good level of reported in engagement and/or developing relationships with parents; CHILDSATB = Children reported as satisfied; SUCCESSFULIMPLEMENTATION = Implementation of intervention successful]

Table 19. Summary of results from consolidated model

Consolidated model	Theory driven	Run in children's free time	Child satisfaction	Parents engaged/relationships developed	High school	Successful intervention
Pathway 1	Present	‐	Present	‐	Present	Yes
Pathway 2	Present	‐	‐	Present	Present	Yes
Pathway 3	Present	Absent	‐	‐	Present	Yes
Pathway 4	Present	Absent	Present	Present	‐	Yes

Absent: absence of condition is essential in triggering success.

Present: presence of condition is essential in triggering success.

‐ (symbol): presence or absence of condition is not essential in triggering success.

This solution emphasises the importance of a theory‐driven intervention across all settings for successful implementation. Three of these pathways are specific to high schools. Here, the evidence suggests that in addition to the importance of a theory‐based intervention, good levels of engagement with parents, high levels of child satisfaction, or running the intervention outside the child's own time can lead to a successfully implemented intervention. A pathway that is not specific to high schools reinforces these findings by showing that being theory‐based, fostering high levels of child satisfaction, reporting good levels of parental engagement, and running an intervention outside the child's own time are sufficient conditions for triggering a positive outcome.

As a whole solution, these pathways had a consistency score of 0.862, suggesting that they were sufficient in triggering the outcome. Interventions that are designed with these sets of characteristics are therefore highly likely to be successfully implemented. We also checked whether any of the configurations described also predicted negation of the outcome, but we found no such evidence. Our coverage score of 0.432, which is modest, suggests that other pathways can also trigger successful implementation, which may be explained by factors not explored in these models. We were not able to incorporate risk of bias judgements directly into the QCA solution.

Based on results of QCAs, we intended to include the following conditions in meta‐analyses, either in the form of subgroup analyses or as covariates in meta‐regression. We planned to examine these as binary or ordinal variables in meta‐analyses; they reflect the single conditions thought to most commonly trigger a successful outcome.

Type of school: high school; primary/elementary school; junior/middle school; other.
Theory driven: does the study name a theoretical framework that underpins the intervention design or delivery style?
Parental engagement: did parents engage or participate in the ways they were expected to?
Child satisfaction: did at least 75% of children report satisfaction with the intervention, or did study authors report high levels of satisfaction?
Timing of the intervention: does the intervention interfere with the child's own time (during lunch or after school)?

Due to data constraints, we were not able to explore child satisfaction in meta‐analyses, as very few studies captured this information, and we operationalised parental engagement as 'parental involvement' ‐ whether or not parents were actively included in the intervention ‐ for similar reasons. We entered the factors beginning "Theory driven", "Parental engagement", and "Timing of the intervention" into subgroup analyses as configurations of conditions in an attempt to replicate the results of the QCA (above). We further explored the link between implementation and outcomes in the next section.

Results of synthesis ‐ part 2: meta‐analyses of effectiveness

Primary outcome: asthma symptoms or exacerbations leading to hospitalisation

We extracted effect sizes from seven studies (Atherly 2009; Bruzzese 2011; Clark 2005; Gerald 2006; Horner 2008; Horner 2015; Levy 2006), and we analysed the data from six. Evidence showed that school‐based asthma self‐management interventions were effective in reducing numbers of hospitalisations among children (standardised mean difference (SMD) ‐0.19, 95% confidence interval (CI) ‐0.35 to ‐0.04; participants = 1873; Figure 6 Analysis 1.1). Effect sizes from all six studies were in the same direction, and I² and Q statistic values provided no evidence of statistical heterogeneity. Gerald 2006 presented data on the median number of hospitalisations, which were not compatible with other extracted data, although it is worth noting that the median level of hospitalisation appeared higher for the intervention group than for the control group post intervention.

Figure 6

Forest plot of comparison: 1 School‐based asthma interventions vs usual care: outcome: 1.1. Exacerbations leading to hospitalisation.

Given that we found no indication of heterogeneity in these models and the likelihood that these analyses would be underpowered, we did not conduct further subgroup analyses. We considered sensitivity analyses, although the small number of studies included in the models precluded a full analysis. All but one of the studies ‐ Bruzzese 2011 ‐ reported on cluster randomised trials, and half of the studies originally reported on binary outcomes (Atherly 2009; Clark 2010; Horner 2008), although sensitivity analyses on these factors revealed no significant differences in effect size. Egger's test for publication bias suggested no evidence of publication bias (the P value for the bias coefficient stood at 0.626), although the small number of studies meant that the test and observations of the funnel plot (not displayed) were ultimately underpowered.

The small number of included studies precluded a detailed investigation of the way in which risk of bias influenced the effect size for this outcome. However, two of the largest studies, which contributed three‐fifths of weighting to the pooled effect size, had low or unclear risk of bias across all domains (Bruzzese 2011; Horner 2015), and in the case of Horner 2015, low risk of bias was seen for each domain, apart from blinding of participants and personnel (unclear risk of bias).

Evidence therefore suggests that school‐based asthma self‐management interventions do reduce the frequency of asthma symptoms and exacerbations requiring hospitalisation among children, with a high level of consistency in the direction and magnitude of effect.

Primary outcome: asthma symptoms or exacerbations leading to emergency department visits

We meta‐analysed effect sizes from 13 studies and found clear evidence that school‐based asthma self‐management interventions were effective in reducing the frequency of ED visits (odds ratio (OR) 0.70, 95% CI 0.53 to 0.92; participants = 3883). Gerald 2006 presented data on the median number of hospitalisations, which were not compatible with other extracted data (full details in Table 13), although the median level of ED visits was observed to be slightly lower for the intervention group than for the control group post intervention.

Heterogeneity in the effects of studies was evident, in terms of both magnitude and direction of effect, with three studies having negligible effect sizes (close to one ‐ Atherly 2009; Clark 2005; Horner 2015) and two studies suggesting a negative intervention effect (McGhan 2003; McGhan 2010); this resulted in an I² value of 26%. The number of studies and the level of heterogeneity allowed us to explore potential study characteristics that could help to explain the observed variation.

Subgroup analyses: exacerbations leading to emergency department visits

Subgroup analyses suggested that the heterogeneity shown in Figure 7 ‐ Analysis 1.2 ‐ was not explained by school type (Figure 8), age (Analysis 3.1), or socio‐economic status of children and intervention deliverers involved in the intervention (Analysis 4.1; Analysis 5.1).

Figure 7

Forest plot of comparison: 1. Effect of school‐based asthma interventions vs usual care, outcome: 1.2. Exacerbations leading to emergency department (ED) visits.

Figure 8

Forest plot of comparison: 2. Effect of school‐based asthma interventions vs usual care subgrouped by school type, outcome: 2.1. Exacerbations leading to emergency department (ED) visits.

We employed subgroup analyses to examine whether any of the intervention conditions that consistently predicted successful implementation in earlier QCAs, namely, explicit use of theory (Analysis 6.1), inclusion of parents (Analysis 7.1), or timing of the intervention (Analysis 8.1), also helped to explain any of the observed heterogeneity in effect sizes. However, we found no evidence that these factors helped to explain heterogeneity.

We also constructed a variable that attempted to replicate some of the implicants (combinations of intervention characteristics) identified in QCAs that trigger successful implementation; however, results appeared to contradict the findings of earlier analyses. A subgroup of studies that replicated one of the configurations theorised to trigger successful intervention implementation (five studies that were theory driven, did not take place in children's own time, and did not involve school nurses) had inconclusive effect sizes as a group (OR 0.85, 95% CI 0.47 to 1.52); in contrast, a subgroup of studies that did not replicate a configuration were found to trigger successful intervention implementation in the QCAs (OR 0.67, 95% CI 0.47 to 0.94). We also created a variable based on a count of intervention characteristics found to trigger successful implementation in our earlier QCAs, and we tested these in subgroup analyses. We constructed a variable reflecting a count of three of the conditions generally found to trigger successful implementation (theory driven, not run in children's own time, and parental engagement (assessed by active involvement of parents)), whereby studies could include zero to three of these 'ingredients'. All studies included in the meta‐analyses had incorporated at least one of these conditions, and subgroup analyses suggested that the number of components was inversely related to effect size, with studies with one component (three studies; OR 0.56, 95% CI 0.33 to 0.97) or two components (seven studies; OR 0.67, 95% CI 0.49 to 0.94) having lower effect sizes than the three studies that included all three components (OR 1.48, 95% CI 0.65 to 3.40); however, the test for differences between subgroups did not suggest that these differences were significant, and moderate heterogeneity remained within one of the subgroups. Among the latter group of studies, two of the three studies evaluated the effectiveness of the RAP (Roaring Adventures of Puff) intervention (McGhan 2003; McGhan 2010). One of these studies provided evidence of a baseline imbalance that could influence the outcome (McGhan 2003), whereby the proportion of intervention group children who had been admitted to an ED was almost ten percentage points higher in the intervention group (23.7%) than in the control group (14%). The second study provided evidence that the mean number of ED visits was higher post intervention in the control group (McGhan 2010), although study authors did not present full data allowing for extraction of the mean number of visits, and the measure used reflected the odds of reporting ED visits.

Sensitivity analyses: exacerbations leading to emergency department visits

We conducted sensitivity analyses to explore the impact of decisions to transform or combine the data. We detected no differences between effect sizes that were originally measured through binary effect sizes (ORs) and those that were originally measured through continuous measures (SMDs). We detected no differences in whether studies assessed intervention effects at 12 months, four to seven months, or within three months (intervals reflecting the spread of studies). All but two studies ‐ Bruzzese 2011 and Persaud 1996 ‐ had randomised children at the school level (cluster RCTs); little evidence suggested that this distinction explained heterogeneity in effect sizes.

In assessing the impact of study quality on effect sizes, we undertook supplementary analyses using meta‐regression in STATA, and, due to the limited number of studies, we combined categories of high and unclear risk when assessing the impact of study quality. We classified none of the studies included in the meta‐analysis for ED visits as having high risk of bias for random sequence generation, although we deemed that eight studies were at unclear risk. Results of sensitivity analyses provided moderate evidence that studies had high or unclear risk of selection bias with respect to breaches in allocation concealment with significantly different effect sizes (OR 0.86, 95% CI 0.64 to 1.16), compared to the three studies that we deemed to have low risk of bias (OR 0.51, 95% CI 0.33 to 0.78). Finally, evidence showed that studies with low risk of bias with respect to collection of outcome data and blinding of collectors were significantly more effective (OR 0.58, 95% CI 0.41 to 0.81) than the seven studies with unclear or high risk of bias (OR 1.04, 95% CI 0.69 to 1.58). Differences in the risk of bias classification for other domains did not significantly explain heterogeneity in effect sizes between studies. We conducted sensitivity analyses to explore the impact of a random‐effects specification on pooled effect size, noting only moderate differences in point estimates between fixed‐effect (OR 0.68, 95% CI 0.55 to 0.85) and random‐effects models (OR 0.70, 95% CI 0.53 to 0.92); however, the level of heterogeneity (I² = 26%) suggested that studies were not measuring a single common effect size, thereby undermining the fixed‐effect assumption (and model results).

Our investigations into the potential impact of publication bias revealed that neither the funnel plot nor Egger's test was indicative of publication bias (the bias coefficient provided weak evidence that smaller studies differed systematically from studies with larger sample sizes).

Evidence therefore suggests that school‐based asthma self‐management interventions do reduce the frequency of asthma symptoms and exacerbations requiring emergency care among children, although variation in the magnitude and direction of effect was not explained coherently by planned subgroup analyses.

Primary outcome: absences from school

Ten studies contributed to our meta‐analyses of effects of interventions on school absences, although there was uncertainty as to whether school‐based self‐management interventions had an impact on reducing absences from school (SMD ‐0.07, 95% CI ‐0.22 to 0.08; participants = 4609; Analysis 1.3; Figure 9). These studies showed substantial heterogeneity between effect size estimates, with I² estimated at 70%. Effect sizes from half of the studies included in the meta‐analysis indicated that the intervention had a negative impact in slightly or significantly increasing the number of school absences in the intervention group relative to the control group (Gerald 2006; Gerald 2009; Howell 2005; McGhan 2010; Splett 2006).

Figure 9

Forest plot of comparison: 1. Effect of school‐based asthma interventions vs usual care, outcome: 1.3. Absence from school.

Subgroup analyses: absences from school

We undertook subgroup analyses to explore study‐level characteristics that could explain this between‐study heterogeneity, although it is worth noting that these analyses were likely to be underpowered and to represent indicative factors that could explain observed differences in the direction and magnitude of effect sizes across studies. The only study included in the meta‐analyses that focused on high schools (and consequently older children) was highly effective in reducing school absences (SMD ‐0.38, 95% CI ‐0.62 to ‐0.15) (Bruzzese 2011); this study appeared to drive much of the heterogeneity explained by subgroup analyses examining school type and child age (Figure 10; Analysis 2.2; Analysis 3.2). Studies that included 25% to 50% children from lower socio‐economic backgrounds were significantly more effective in reducing levels of school absence (SMD ‐0.23, 95% CI ‐0.36 to ‐0.09; studies = 2) than studies with greater numbers of children from deprived backgrounds (over 50%) for whom the effect was negligible (SMD 0.01, 95% CI ‐0.09 to 0.11; 2 studies) and studies in which less than 25% of children were from deprived backgrounds or in which this was unclear, where the pooled effect size indicated negligible effect (SMD ‐0.02, 95% CI ‐0.29 to 0.24; 6 studies).

Figure 10

Forest plot of comparison: 2. Effect of school‐based asthma interventions vs usual care subgrouped by school type, outcome: 2.2. Absence from school.

Studies that involved existing school staff (teachers or school nurses) in delivery of the intervention were significantly less effective (SMD 0.08, 95% CI ‐0.08 to 0.24; 3 studies) than studies in which the intervention was mainly delivered and facilitated by stakeholders who were external to the school (SMD ‐0.17, 95% CI ‐0.32 to ‐0.02; 7 studies; Analysis 5.2;Figure 11). Findings of the earlier QCA show that involvement of internal stakeholders within the school in delivery of the intervention did not always lead to successful intervention implementation, but they also show that involving school staff in intervention delivery may be one of a configuration of conditions that trigger successful implementation, none of which are sufficient alone. Similar processes may occur around their role in reducing the level of school absence.

Figure 11

Forest plot of comparison: 10. Effect of school‐based asthma interventions vs usual care subgrouped by configuration of conditions (iI), outcome: 10.3. Absence from school.

We conducted subgroup analyses involving the conditions and configurations found to be sufficient in earlier QCAs to trigger successful implementation. But these findings did not significantly explain the heterogeneity in effect sizes, with two exceptions. Analysis 8.2 provided evidence that interventions that took place during the child's own time had significantly greater impacts in reducing school absence (SMD ‐0.23, 95% CI ‐0.36 to ‐0.11; 2 studies) than those that took place at another point in the school day (SMD ‐0.01, 95% CI ‐0.18 to 0.16; 8 studies), although a substantial level of heterogeneity remained among this latter group of studies (I² = 62%). We noted strong evidence around the role of theory (Analysis 6.2), whereby studies that reported drawing upon a defined theoretical framework had a significantly more impactful pooled effect size (SMD ‐0.20, 95% CI ‐0.36 to ‐0.04; 6 studies) than studies that did not (SMD 0.08, 95% CI ‐0.05 to 0.20; 4 studies). Although moderate levels of heterogeneity remained (I² = 41% for studies that explicitly drew upon theory and I² = 28% for those that did not), and even though interpretation of these results is not straightforward (see discussion), this result indicates that theory‐driven studies may achieve better outcomes with respect to this domain.

Sensitivity analyses: absences from school

We conducted sensitivity analyses to explore whether the following factors, reflecting study design or analytical decisions made during the review process, helped to explain heterogeneity in effect size: (I) transformations were made to the original effect size (conversions between ORs and SMDs; Chinn 2000); (ii) cluster RCT or not; (iii) the data collection period; and (iv) the study's risk of bias . We found no evidence to suggest that transformations in effect sizes explained heterogeneity, and no evidence indicated that the unit of randomisation (school vs child) explained variation in effect size. The three studies that collected absence data within three months post intervention (or for which the collection date was unclear) did exhibit a weaker effect in reducing school absences (Gerald 2006; Howell 2005; Persaud 1996), with Gerald 2006 and Howell 2005 showing a negative intervention impact, although this was not significantly different from studies that assessed absences over the 12 months post intervention. Little evidence suggests that risk of bias influenced the effect size obtained; however, studies that had taken steps to blind assessment of outcomes and to avoid detection bias had a greater impact in reducing school absences (SMD ‐0.27, 95% CI ‐0.38 to ‐0.17; 3 studies) than studies that did not take these steps (SMD ‐0.07, 95% CI ‐0.02 to 0.16; 7 studies).

We investigated the potential impact of publication bias by examining a funnel plot and the results of Egger's test. These tests did not provide strong evidence that data were impacted by publication bias (the bias coefficient provided weak evidence that smaller studies differed systematically from studies with larger sample sizes). We examined differences between the fixed‐effect model and the random‐effects model reported above. The fixed‐effect model showed that the pooled point estimate remained similar, but with a less conservative confidence interval (SMD ‐0.05, 95% CI ‐0.11 to 0.02). However the level of heterogeneity was substantial (I² = 70%), which suggested that these studies were not measuring a single common effect size and thereby undermined the fixed‐effect assumption (and model results).

Evidence from the overall pool of studies therefore suggests that school‐based asthma self‐management interventions did not have an impact in reducing absence from school, although variation in direction and magnitude was substantial. Planned subgroup analyses assisted in identifying particular groups of studies and did, or did not, have a beneficial effect.

Primary outcome: days of restricted activity

Three studies contributed data to our meta‐analysis of the impact of school‐based asthma self‐management interventions in reducing the number of days of restricted activity that children experienced (Bruzzese 2011; Cicutto 2005; Cicutto 2013). These studies provided evidence that the intervention mode could reduce the number of days of restricted activity experienced (SMD ‐0.30, 95% CI ‐0.41 to ‐0.18; 1852 participants; 3 studies; Analysis 1.4), albeit based on a limited number of studies, two of which evaluated the same intervention design (the Roaring Adventures of Puff) (Cicutto 2005; Cicutto 2013). All three studies provided relatively consistent evidence around the direction and magnitude of effect (I² = 0%). Reporting on the results of subgroup analyses is not meaningful in the presence of low heterogeneity and small numbers of studies, and many sensitivity analyses could not be conducted for the same reason, although it is worth noting that we rated none of the included studies as having high risk of bias for any domain assessed for the outcome evaluation risk of bias tool (Figure 12).

Figure 12

Forest plot of comparison: 1. Effect of school‐based asthma interventions vs usual care, outcome: 1.4. Days of restricted activity.

Secondary outcome: unplanned visits to a medical provider

From a meta‐analysis of five studies (Analysis 1.5), evidence shows that school‐based asthma self‐management interventions did reduce the number of unplanned or unscheduled visits to a medical provider (OR 0.74, 95% CI 0.60 to 0.90; 3490 participants; 5 studies). Despite inconsistency in the magnitude (and direction) of effect in the case of McGhan 2003, which indicated a small negative intervention impact, the meta‐analysis provided little evidence of statistical heterogeneity (I² = 0%). As was the case above, the small number of studies and the absence of heterogeneity did not support meaningful investigation of subgroup analyses, nor the opportunity to undertake a full assessment of some of the assumptions made in pooling the data (see Table 13 for further details on the derivation of effect sizes). Similarly, we were not able to assess the potential impact of publication bias. Two studies contributed almost 75% towards the pooled effect size (Bruzzese 2011; Cicutto 2013), and we rated neither study as having high risk of bias in any domain (Figure 13).

Figure 13

Forest plot of comparison: 1. Effect of school‐based asthma interventions vs usual care, outcome: 1.5. Unplanned visit to hospital or GP due to asthma symptoms.

Secondary outcome: experience of daytime and night‐time symptoms

As described in the section on Included studies, trialists adopted different strategies in measuring the impact of interventions on children's daytime and night‐time symptoms. We constructed separate models of meta‐analysis for studies reporting on daytime symptoms (Analysis 1.6) and night‐time symptoms (Analysis 1.7), although some variation remained in the way in which symptom data were collected (Table 13).

Uncertainty surrounded the question of whether school‐based self‐management interventions reduced the level of daytime symptoms that children experienced (SMD ‐0.15, 95% CI ‐0.32 to 0.02; I² = 0%; 1065 participants; 5 studies), with the confidence interval just crossing the line of no effect (zero). However, study reports show consistency in the direction of effects (Figure 14). Even greater uncertainty surrounded whether self‐management interventions in schools reduced the level of night‐time symptoms reported by children in random effects meta‐analysis (SMD ‐0.18, 95% CI ‐0.52 to 0.15; I² = 40%; 459 participants; 4 studies), with two studies providing weak evidence that night‐time symptoms actually increased among children receiving school‐based asthma self‐management interventions. We performed sensitivity analyses using a fixed‐effect model, with the pooled effect size across the four studies indicating that night‐time symptoms decreased (SMD ‐0.26, 95% CI ‐0.46 to ‐0.06; 4 studies), although given the inconsistency in the direction of effect, the underlying assumptions of the fixed‐effect model cannot be substantiated, and the random‐effects model may provide a more realistic estimate of intervention effects on night‐time symptoms.

Figure 14

Forest plot of comparison: 1. Effect of school‐based asthma interventions vs usual care, outcome: 1.6. Experience of daytime and night‐time symptoms ‐ daytime symptoms.

Reporting on the results of subgroup analyses was not meaningful with the few included studies; other sensitivity analyses could not be conducted for the same reason. One study that measured change in daytime symptoms showed a weak effect of the intervention in lowering the level of daytime symptoms (see Table 13) (Clark 2010).

Secondary outcome: lung function

We extracted outcomes measuring trial impacts on lung function from five studies, although we did not combine these data in meta‐analyses due to conceptual (and statistical) heterogeneity. We have presented these outcomes in full in Table 13.

Secondary outcome: use of reliever therapies

Four studies reported on the use of reliever therapies among children who had received self‐management interventions in school (Table 13), and we included effect sizes from two studies with clinical and conceptual equivalence in a random‐effects meta‐analysis (Figure 15; Analysis 1.8). The pooled result provided uncertain evidence on the impact of the intervention on children's use of reliever therapies (OR 0.52, 95% CI 0.15 to 1.81; 437 participants; 2 studies). The level of heterogeneity between studies was substantial (I² = 68%), although both were somewhat consistent in the direction of effect, indicating lower odds of (frequent) reliever therapy usage. One study had low or unclear risk of bias across all domains considered (Gerald 2009), and we judged McGhan 2010 to have high risk of bias in terms of attrition bias and selective reporting.

Figure 15

Forest plot of comparison: 1. Effect of school‐based asthma interventions vs usual care, outcome: 1.8. Use of reliever therapies, e.g. beta₂‐agonists.

Secondary outcome: corticosteroid dosage and use of add‐on therapies

We found unclear evidence on the impact of interventions on children's use of corticosteroids and add‐on therapies (OR 1.25, 95% CI 0.88 to 1.77; 614 participants; 3 studies; Figure 16; Analysis 1.9). We noted no evidence of statistical heterogeneity between these study impacts on corticosteroid usage (I² = 0%), and as reporting on the results of subgroup analyses is not meaningful with low heterogeneity and few studies, we could not conduct other sensitivity analyses for the same reason. We deemed one study included in the model to have low risk of bias for all domains except blinding of participants and personnel, for which we deemed the risk to be unclear (Bruzzese 2011); we deemed the other two studies to have high risk of bias in one and two domains (McGhan 2003; McGhan 2010), respectively, with both deemed to have high risk of attrition bias from incomplete and unexplained dropouts at outcome data collection.

Figure 16

Forest plot of comparison: 1. Effect of school‐based asthma interventions vs usual care, outcome: 1.9. Corticosteroid dosage and/or use of add‐on therapies (usage of).

We included two studies reporting appropriate usage of corticosteroids and add‐on therapies. Although the direction of findings differed substantially between studies, resulting in considerably high levels of heterogeneity (I² = 87%), we did not estimate a pooled effect size (Analysis 1.10).

Secondary outcome: health‐related quality of life

Nine studies provided data on the effectiveness of school‐based self‐management interventions in improving children's quality of life. Because of conceptual differences in the way in which the outcome was measured, one meta‐analysis of seven studies explored intervention impacts on quality of life measures assessed through standardised mean differences using mainly the Paediatric Asthma Quality of Life Questionnaire (PAQLQ) (Figure 17; Analysis 1.11), and provided evidence of effectiveness (SMD 0.27, 95% CI 0.18 to 0.36; 2587 participants; 7 studies). This model provided no evidence of statistical heterogeneity in effectiveness (I² = 0%), with all studies providing estimates of positive improvements, although these were not all statistically significant in all studies. The low level of heterogeneity and the few included studies meant that conducting subgroup analyses was not appropriate. We deemed that five of the seven studies included in the meta‐analysis were at high risk of bias in at least one domain (Al‐Sheyab 2012; Henry 2004; Horner 2008; Howell 2005; Kintner 2009), although the two studies with low or unclear risk of bias in all domains contributed over 60% of the weighted effect size (Cicutto 2005; Cicutto 2013). Explorations of the funnel plot and Egger's test were underpowered, and publication bias could not be adequately tested.

Figure 17

Forest plot of comparison: 1. Effect of school‐based asthma interventions vs usual care, outcome: 1.11. Health‐related quality of life (SMD).

A second meta‐analysis involving eight studies also provided evidence that children in intervention groups had higher HRQoL than children in control groups (MD 0.35, 95% CI 0.06 to 0.64; 2950 participants; 8 studies) based on PAQLQ results at follow‐up (Analysis 1.12). The mean difference, while again indicating that the impact did not cross the threshold of no effect, fell below 0.5 ‐ the threshold considered to indicate a clinically significant change in HRQoL on this scale. Heterogeneity among studies was considerably high (I² = 81%). One study in particular had relatively high levels of baseline imbalance, and a sensitivity analysis removing this value resulted in a lower point estimate but much lower levels of heterogeneity (MD 0.21, 95% CI 0.07 to 0.36; I² = 24%) (Al‐Sheyab 2012). We included this same study in Analysis 1.11, although we used different data to obtain an effect size (P value and sample size). We did not further explore heterogeneity because included studies were few and, similarly, explorations of the funnel plot and Egger's test were underpowered; therefore, we could not adequately assess publication bias. We deemed that four of the studies included in Analysis 1.12 were at high risk of bias in at least one domain. A further sensitivity analysis involving constructing a fixed‐effect model yielded a similar point estimate (MD 0.32, 95% CI 0.21 to 0.43; I² = 81%; 8 studies), although the considerably high level of heterogeneity indicates that this is not a suitable analytical framework.

Despite the additional study included in Analysis 1.12, we consider the results from Analysis 1.11 to be more reliable because of the considerably high heterogeneity observed in the model for MD and the insufficient number of studies to fully explore drivers of this heterogeneity.

Therefore, evidence suggests that school‐based asthma self‐management interventions do improve children's quality of life, although this finding may not reach a point of clinically significant improvement. Although all studies provided an indication of a positive beneficial effect, variation in the size of the effect was substantial.

Secondary outcome: withdrawal from the study

Meta‐analysis provided no evidence that participation in the intervention was linked to withdrawal from the study (OR 1.14, 95% CI 0.92 to 1.43; 3442 participants; 13 studies; Figure 18; Analysis 1.13). We detected no substantial statistical heterogeneity (I² = 0%), although some qualitative differences were apparent between studies that reported very low levels of withdrawal among those receiving treatment relative to those in control groups (Bruzzese 2008), and relative to those with very high levels of withdrawal (Kintner 2009; Patterson 2005); in neither case would the level of withdrawal be described as problematic (not exceeding 25% of participants), and the stark relative effect was driven by very small sample sizes in some studies (Bruzzese 2008; Kintner 2009).

Figure 18

Forest plot of comparison: 1. Effect of school‐based asthma interventions vs usual care, outcome: 1.13. Withdrawal from the study.

Despite the low level of heterogeneity, we have presented subgroup analyses because of the link between this outcome and the QCAs presented earlier. When we replicated one set of configurations in the subgroup analysis to mirror QCA findings (Analysis 9.3), we found weak/uncertain evidence to suggest that studies that used theory, while avoiding running the intervention in children's own time and having no substantial school staff involvement, were less likely to have children drop out before outcomes were assessed (OR 0.88, 95% CI 0.55 to 1.40; 4 studies) when compared with studies with other configurations of conditions (OR 1.23, 95% CI 0.95 to 1.58; 8 studies). We also found no evidence that withdrawal from the intervention was associated with school type (Figure 19).

Figure 19

Forest plot of comparison: 2. Effect of school‐based asthma interventions vs usual care subgrouped by school type, outcome: 2.3. Withdrawal from the study.

Subgroup analyses seeking to reveal patterns of heterogeneity in the odds of withdrawal did not show that timing of assessment, unit of randomisation (cluster vs individually randomised trials), or risk of bias explained patterns of withdrawal. This included risk of attrition bias assessments, although the meta‐analysis explored differential patterns of attrition and did not account for instances in which both intervention and control groups had high levels of attrition (as was the case when risk of attrition bias was assessed). We found no evidence to show that publication bias was an issue in terms of withdrawal data. Because of the low level of statistical heterogeneity, fixed‐effect and random‐effects specifications for the meta‐analyses were equivalent, with one study accounting for 46% of the weighting; we classified this study as having high risk of bias in three domains and unclear risk of bias in the remaining four domains (Bartholomew 2006).

Results of synthesis ‐ part 3: adjunct meta‐analyses exploring the link between implementation and effectiveness of school‐based asthma self‐management interventions

We conducted adjunct meta‐analyses to explore whether interventions that were deemed successful in terms of implementation were also deemed successful in terms of their effectiveness (see Figure 2), using a subset of studies contained within the process evaluations. For inclusion in these analyses, we considered studies that included a control group; however studies could have employed randomisation or quasi‐experimental methods, and control group children could have received an alternative intervention that might have included an asthma component.

Because of conceptual and methodological differences in study design, these studies provide indicative evidence only pertaining to the impact of self‐management interventions on children's asthma outcomes, but they help us to establish links between implementation factors and asthma outcomes. Researchers defined successful implementation the same way it was defined in our QCA, and this represented a combined indicator around attrition, adherence, and dosage. We considered two outcomes ‐ ED visits and school absences ‐ when we found sufficient studies to form a meta‐analysis. Both models included effect sizes from seven studies, with five studies in each appearing in earlier meta‐analyses (with studies considered as process and outcome evaluation studies (Bruzzese 2011; Cicutto 2013; Horner 2015; Howell 2005; Levy 2006)), and two studies in each meta‐analysis included as process evaluation studies only (Joseph 2010; Joseph 2013).

Meta‐analysis of ED visits shows that the included interventions were successful in reducing the number of ED visits (Figure 20; Analysis 11.1), but with a high I² value (52%) signalling substantial levels of heterogeneity. Subgroup analyses, based on implementation scores, indicated that studies classified as successfully implemented had a greater impact in reducing ED visits (SMD ‐0.26, 95% CI ‐0.48 to ‐0.04; 4 studies) than studies that were not as successful (SMD ‐0.09, 95% CI ‐0.28 to 0.10; 3 studies), although this difference was not statistically significant (P value for subgroup differences = 0.26). Meta‐analysis of the impact of self‐management interventions provided uncertain evidence that these interventions were successful in reducing children's absences from school (SMD ‐0.12, 95% CI ‐0.28 to 0.04; 7 studies). However, subgroup analyses based on the combined implementation score indicate that studies that were successfully implemented had significantly higher effect sizes (SMD ‐0.28, 95% CI ‐0.39 to ‐0.18; 3 studies) than those that were not successfully implemented (SMD 0.04, 95% CI ‐0.09 to 0.18; Figure 21).

Figure 20

Forest plot of comparison: 11. Adjunct analyses ‐ impact of Implementation on selected outcomes, outcome: 11.1. Exacerbations leading to emergency department (ED) visits.

Figure 21

Forest plot of comparison: 11. Adjunct analyses ‐ impact of Implementation on selected outcomes, outcome: 11.2. Absence from school.

In both models, had the focus been restricted to well‐implemented studies only, the conclusions would have changed, and these studies would have provided evidence that school‐based asthma self‐management interventions were effective in reducing these outcomes. Although restricted to selected outcomes and a subset of studies, these models help to illuminate the links between successful implementation and intervention effectiveness, and provide justification for meta‐analyses based on earlier QCAs to test emerging hypotheses.

Part 4: update of the logic model

Figure 22 presents an updated logic model. This is a graphical depiction of synthesised evidence showing that school‐based asthma interventions have a positive impact in reducing healthcare usage, improving quality of life (albeit not at a clinically meaningful level), and reducing days of restricted activity (shaded green). These were termed 'intermediate outcomes' in our original model (Figure 1), although some of the pathways through which these improvements are achieved remain poorly understood, particularly around proximal outcomes including lung function and daytime/night‐time symptoms (shaded blue and grey). We found evidence of a link between successful implementation (through results presented in part 3) and improved outcomes, although Figure 22 shows that other factors around the intervention design may directly lead to improvement in 'intermediate' outcomes. Of these, being theory driven is likely to be the most important element leading to successful implementation, and later, successfully improving children's outcomes, although the logic model shows that other conditions are likely to be important in certain circumstances.

Figure 22

Use of QCA alongside meta‐analysis has helped to disentangle the ways that school‐based asthma interventions 'work' to a certain extent. The logic model helps to show the strength of evidence for many parts of the causal chain but also shows gaps in evidence on which future reviewers may focus their efforts (boxes shaded grey in Figure 22), including (I) establishing which proximal outcomes are important elements of the causal chain between intervention and intermediate outcomes; (ii) improving understanding of the role of contextual and participant characteristics; and (iii) examining distal characteristics and stability of improved outcomes.

Discussion

Summary of main results

Summary and further description of qualitative comparative analysis (QCA) results

One of the most consistently positive conditions that appeared in configurations triggering a successfully run intervention was a named theoretical framework described as underpinning the intervention (Table 20). However, a diverse set of theoretical standpoints were represented (see Description of studies), and we are unable to attribute a successful intervention to a single conceptual or theoretical framework. Merely the use of named or explicitly expressed theory, in conjunction with other conditions, led to better implementation. These configurations also included interventions not run in children's own time, good levels of engagement from parents and satisfaction from children reported, and some configurations specific to high schools. It is not clear whether the theories used to underpin an intervention were equally suitable, and we were not able to ascertain how the theoretical framework was used to shape or inform different stages of the intervention.

Table 20. Summary of QCA results based on intermediate solutions

Model 1. Setting and participant features	School health centre	High school	Parents direct intervention recipients	Teachers direct intervention recipients	School nurses/others direct intervention recipients		Successful intervention
Pathway 1	Present	Present	Present	Absent	‐		Yes
Pathway 2	Absent	Present	Absent	‐	‐		Yes
Pathway 3	Absent	‐	Absent	Absent	Absent		Yes
Pathway 4	Present	Present	Present	‐	Present		Yes
Model 2. Recruitment and retention processes	Additional marketing materials	Provision of incentives	Provision of catch‐up sessions	Provision of reminders			Successful intervention
	‐	‐	‐	‐	‐		No solution found
Model 3. Curriculum, pedagogy, and intervention emphasis	Focus on establishing alliances with care providers	Focus on asthma symptom recognition and management	Tailored content	Emphasis on personal responsibility	Interactive pedagogical style	Diverse pedagogical style	Successful intervention
	‐	‐	‐	‐	‐		No solution found
Model 4. Modifiable intervention processes	Theory driven	Run in class time	Run in students’ free time	School nurse key role in delivery or teaching	Personalised or individual 1‐to‐1 instruction		Successful intervention
Pathway 1	Present	‐	Absent	Absent	‐		Yes
Pathway 2	Present	‐	‐	Present	Absent		Yes
Model 5. Stakeholder engagement	School asthma policy	Child satisfaction	Teachers engaged/ relationships developed	Parents engaged/ relationships developed	School nurses engaged/ relationships developed		Successful intervention
Pathway 1	Absent	‐	‐	Present	Absent		Yes
Pathway 2	‐	Present	‐	‐	Absent		Yes
Model 6. Consolidated model	Theory driven	Run in students’ free time	Child satisfaction	Parents engaged/ relationships developed	High school		Successful intervention
Pathway 1	Present	‐	Present	‐	Present		Yes
Pathway 2	Present	‐	‐	Present	Present		Yes
Pathway 3	Present	Absent	‐	‐	Present		Yes
Pathway 4	Present	Absent	Present	Present	‐		Yes

Absent: absence of condition is essential in triggering success.

Present: presence of condition is essential in triggering success.

QCA: qualitative comparative analysis.

‐ (symbol): presence or absence of condition is not essential in triggering success.

We found that good levels of engagement from parents and positive experiences among children, in combination with other conditions, were sufficient conditions for a successful intervention. Positive parental engagement reflected high levels of co‐operation in providing information to trialists, as noted by study authors (Dore‐Stites 2007; Joseph 2013), or more active forms of engagement, including co‐operation with home or school visits (Engelke 2013; Howell 2005), attendance at seminars (Bruzzese 2008), or telephone appointments received from the trialists (Engelke 2013). In contrast, a different set of studies reported difficulties in engaging parents to provide consent (when consent was actively sought) or to assist with data collection (Berg 2004; Gerald 2006; Terpstra 2012); difficulties in participation (Brasler 2006; Cicutto 2013; Kintner 2012; Kouba 2012; Levy 2006); or problems with adherence or behaviour change (Mujuru 2011). Children's satisfaction was found to be a sufficient condition for successful implementation (in combination with other conditions) and was collected in eight included studies, with four studies providing evidence that most children were satisfied through qualitative statements based on children's other stakeholders' perceptions (Al‐Sheyab 2012a; Brasler 2006; Bruzzese 2004; Howell 2005), and four studies providing evidence based on quantitative data (Berg 2004; Bruzzese 2008; Dore‐Stites 2007; Kintner 2012). None of the studies included in the QCAs reported low levels of child satisfaction, although one study (not included in the QCAs), which provided a low‐intensity intervention, did report low levels, with low levels of satisfaction (64% and 67%) for some indicators (Jackson 2006).

With respect to school nurse involvement, the presence and involvement of school nurses in interventions appear to be instrumental in triggering a successful intervention under certain conditions when children are not engaged in personalised or tailored interventions. Finally, the timing of interventions was important in triggering successful interventions, with interventions that did not interrupt children's own time triggering successful implementation in two different configurations.

No single condition appeared in isolation as a trigger for successful implementation. This highlights the complexity of triggering a successful intervention, as well as the utility of the QCA approach in capturing complex causal recipes. This finding is further supported by modest levels of coverage of any pathway.

Summary of outcome evaluation results

Results from meta‐analyses show that school‐based self‐management interventions led to small average improvements in several important outcomes, including hospitalisations (six studies), emergency department (ED) visits (13 studies), and health‐related quality of life (seven studies). A smaller number of studies contributed to meta‐analyses suggesting positive results for unplanned medical visits (five studies) and days of restricted activity (three studies). Effects on school absences, symptoms, and use of medication were also small, although our certainty for these outcomes was low or very low and confidence intervals included small or no effect. The effect on withdrawals suggested similar levels of attrition between intervention and control conditions.

The original logic model and the updated logic model show that evidence for effectiveness of the intervention was stronger than for urgent care contact and quality of life than for symptoms (Figure 1; Figure 22, respectively). We did not measure distal outcomes (e.g. academic achievement). This is likely a partial reflection of heterogeneity in measurement approach in terms of lung function and daytime and night‐time symptoms.

Researchers observed the most prominent intervention impacts for outcomes involving healthcare usage. Although conceptually relatively homogeneous, they were measured in several different ways, prompting us to undertake several transformations to facilitate meta‐analysis. The magnitude of effect sizes for hospitalisations, ED visits, and other instances of unplanned healthcare usage was similarly small across all three outcomes when considered in absolute terms. However, this indicates that intervention effect can also reach across children's healthcare pathways to include both primary and secondary episodes of care. In contrast, it was anticipated that a greater effect would be evident for school absences than was apparent in our review. However, heterogeneity was substantial in meta‐analyses of school absence (I² = 70%), and additional subgroup analyses suggest that the way in which the intervention was implemented may have had a substantial impact on this outcome.

Effects of the intervention were relatively consistent across outcomes, with the exception of school absences and ED visits. Most planned investigations of heterogeneity were generally uninformative or inconclusive in explaining variation in our results. Indications suggest that both school type and age of the child may help to explain some between‐study heterogeneity in models for school absence, with the intervention exerting a greater impact on older children in high schools, although this result was primarily driven by a single study (Bruzzese 2011). Two studies suggested that interventions with moderate to high numbers of children from lower socio‐economic backgrounds (between 25% and 50% of children) resulted in fewer school absences for intervention children (Cicutto 2013; McGhan 2003), although the relationship between the proportion of children from a lower socio‐economic background and effect size was not linear. We found generally mixed evidence around the impact of including parents. Based on subgroup analyses, interventions that did include parents appeared to confer no additional benefit compared with those that did not. Similarly, meta‐analyses provided contrasting evidence as to whether involvement of school nurses had a positive impact on children's outcomes.

Contribution of a mixed methods approach

The mixed methods approach adopted here allowed us to (I) understand design and implementation processes associated with more successful implementation of school‐based self‐management interventions; (ii) develop judicious and theory‐driven hypotheses for testing in subsequent meta‐analyses with covariates that reflected configurations of study conditions as well as single conditions; and (ii) explore the links between successful implementation and intervention outcomes.

Adjunct analyses showed links between intervention implementation and more impactful interventions, although the strength of these relationships differed for Analysis 11.1 and Analysis 11.2. Analysis of ED visits did not rule out differential effects between subgroups. We classified implementation of the intervention as successful in four studies. Study authors reported lower levels of ED visits with the intervention, and this finding was consistent with results for subgroups of studies that classified interventions as not successful. However, the result was inconclusive for studies that did not implement the intervention successfully. In the case of school absence, evidence shows greater impact of studies that were well implemented versus those that were not successfully implemented. This held when we restricted our focus to direct comparison of interventions (five studies) versus usual care.

Meta‐analyses based on the findings of earlier QCAs, which assessed the impact of school‐based asthma self‐management interventions in lowering levels of school absence, also show that individual conditions that were frequently part of configurations that triggered successful intervention implementation explained some of the between‐study heterogeneity. Notably, studies that were theory driven had greater impact on reducing school absences than those that were not, with the confidence interval for the subgroup of studies that explicitly used theory clearly within the boundaries of an effective intervention.

Further meta‐analyses suggesting that interventions that did not involve existing school staff in a substantial delivery or facilitating role were those that achieved greater levels of impact in lowering school absence. This corresponds with QCA findings that involvement of school staff could be counterproductive in certain configurations. Well‐implemented interventions that are supported by theory and can be implemented independently of existing school staff appear to be sufficient for lowering levels of school absence in these analyses.

Translating evidence into practice

The financial implications of asthma treatment and care for healthcare systems are significant; costs up to £1 billion per year are reported in the UK. A formal economic evaluation would be needed to determine how the reduction in healthcare use observed in this review impacts the financial burden on healthcare systems incurred by managing asthma. Although a similar reduction in school absence has not been established in this review, subgroup analyses developed on the basis of earlier QCAs identified study‐level characteristics associated with substantial reductions in absence, most notably interventions explicitly using theory.

In terms of the design of interventions, the importance of theory was emphasised in QCA results and was given further limited support by some of the subgroup analyses conducted as part of the meta‐analyses. However, it is not clear if the use of theory in interventions is a marker of the quality of the interventions and the experience of researchers, or is more integral to intervention success and provides an anchor for trialists to return to and actively draw upon. Based on QCA results, when trialists take steps to measure levels of child satisfaction (including levels of enjoyment and fulfilment from activities), this is reflected in delivery of a successful intervention. The presence and involvement of school nurses appear to be instrumental in successful implementation of the intervention under certain conditions, particularly when children are not engaged in personalised or tailored interventions.

Overall completeness and applicability of evidence

Most of the included studies were conducted in the USA, specifically in inner city areas with large numbers of children from ethnic minority backgrounds and/or lower‐income households; very few of the included studies came from the UK or Europe. Although we anticipated that broader contextual factors around health policy and access to health care are likely to shape the design and implementation of the intervention (see logic model in Figure 22), we have not synthesised the impact of these contextual factors.

The US focus of studies may have differing implications for the transferability of interventions. The nature of healthcare delivery and the large number of people without adequate healthcare coverage could mean that the intervention has a greater impact in US settings, particularly among lower‐income populations with substantial levels of underdiagnosis and low levels of access to appropriate medication plans. Several interventions (e.g. those of Bruzzese 2011 and Gerald 2009) were developed precisely on the basis of this rationale, focusing on low‐income groups or ethnic minority groups with inadequate access to health care, and selected schools as the delivery site because of the universality of education (as opposed to health care) in these settings. The implications for transferability could mean that weaker effect sizes are achieved in settings with better healthcare coverage, higher rates of diagnosis, and greater equality in access to medication (e.g. in settings such as the UK, where health care is universally free at the point of delivery). In contrast, many of the findings around intervention implementation are likely to be universal across several settings because of the relative universality of the way in which children attend schools, for example, better implementation when the intervention takes place outside children's own time.

Many outcomes with stronger evidence of an intervention effect were those commonly experienced by children with relatively severe asthma. For example, in Atherly 2009, when an intervention was implemented in high schools among children with mild to severe asthma, around 3% of children had been hospitalised for asthma at baseline, and less than 10% had visited an ED. Values suggesting that unplanned secondary healthcare utilisation is relatively rare among children with asthma are also observed in prevalence studies, for example, in Harris 2017, which examined asthma patterns in London high schools.

Many of the studies included in the QCA and in the meta‐analyses were conducted as cluster randomised controlled trials (RCTs); however few of these studies described the impact of this clustering effect either quantitatively or qualitatively. It is likely that conducting school‐level randomisation is an important consideration in terms of the feasibility of the study and serves as a step toward prevention of contamination of treatment impact, although the opportunity to explore implementation and impact of school‐level designs is not taken up by many trialists. This means that we are unable to comment on the generalisability of study findings with regards to school cultures.

High schools were better represented among studies included as process evaluations than among those included as outcome evaluations. Whether this is a reflection of the challenge of implementing RCT designs in high schools compared to primary schools was not directly addressed by the studies included in this review, although distinct configurations of conditions that triggered successful interventions were identified in QCAs for high schools and/or older children. Meta‐analyses revealed little 'qualitative' impact of conducting interventions in high schools rather than in other types of schools for most outcomes except school absences, although this assertion is based on inclusion of few high schools in subgroup analyses and low heterogeneity for many healthcare usage outcomes.

Many studies did not report on the outcomes specified in the protocol for this review and encountered further issues with the incompatibility of some reported effect sizes. In fact, any of the meta‐analyses performed (the largest including 13 studies) provided only a partial account of the total number of studies included. Some models, especially subgroup analyses, may have been effectively underpowered. Future systematic reviewers exploring public health interventions may wish to explicitly include a narrative synthesis of all studies in terms of study design, which may examine both the nature of the intervention, the types of outcomes collected, and the impact of interventions on these outcomes, including graphical representations (Thomson 2013), for a more complete account and understanding of the impact and feasibility of the model.

Finally, because we excluded studies that delivered similar interventions in different settings, we do not know the added value of running an intervention in a school compared with running an intervention in a hospital or community setting. What is clear, however, is that schools provide access to large numbers of children with asthma, including those who do not regularly attend appointments with their medical provider; therefore the school environment can be considered an important third space for delivery of interventions that can improve both children's outcomes and healthcare usage. This review has shown that school‐based self‐management interventions are effective in improving several outcomes for children with asthma, and that those who design future interventions should consider a number of configurations, including instructor, theory, and time of day, in their design. The outcomes of this review will directly inform the development of a school‐based self‐management intervention for children with asthma in London secondary schools.

Certainty of the evidence

The 'Summary of findings' table highlights our reasons for downgrading the certainty of evidence for the main outcomes of interest in this review, with process evaluations considered separately below. We noted issues in the execution of all studies, although the impact of risk of bias differed across outcomes. We deemed that several studies had high or unclear risk of bias, although these results did not appear to inflate the effect size relative to that provided by low‐risk studies, and in most cases they did not influence the direction of effect. Studies that we deemed to have unclear or high risk of bias may nevertheless have contributed to decisions to downgrade the certainty of evidence through other factors, including directness of outcome measurements. For example, school absences were measured in a variety of ways, and not all approaches were specific to asthma‐related school absences.

We deemed the certainty of evidence to be moderate for four outcomes delineated in the 'Summary of findings' table: hospitalisation, unplanned medical visits, quality of life, and symptoms. Each of these outcomes showed positive intervention effects (or effects that were very close to being classed as positive effects in the case of daytime symptoms). For two of the outcomes reported in the 'Summary of findings' table, we deemed that the certainty of evidence was low (school absences and ED visits), and we found evidence certainty to be very low for a further outcome on medication usage. Again, indirectness and unexplained heterogeneity were the main drivers for downgrading of evidence.

Additional considerations not necessarily captured in the 'Summary of findings' table should be considered when quality of the evidence is examined. First, we decided to include in our analyses some cluster RCTs with relatively low numbers of clusters. Although these studies tended to be comparatively small by their nature and therefore did not contribute greatly to pooled effect sizes, there remains the possibility that the intervention effects are slightly exaggerated compared to those of individually randomised trials or large cluster RCTs (see also the section on bias below). Nevertheless, this risk should be balanced against the potential bias introduced by overlooking information from such (smaller) trials. Similarly, effect sizes were harmonised for most outcomes, with the most substantial transformations involving conversion between standardised mean differences (SMDs) and odds ratios (ORs) to develop a common metric; although this appeared to have minimal impact, and different effect sizes tended to be consistent in direction/impact regardless of original measurement (see Analysis 12.1 through to Analysis 12.27), this is further evidence of indirectness in outcome measures, which is an indicator of lower‐certainty evidence.

In contrast, we judged the quality of the process evaluation literature to be almost uniformly poor, with many studies having high or unclear risk of bias across several domains. This is likely due to various factors but most plausibly is a reflection of previous lack of guidance around the conduct of process evaluations, as well as difficulty in identifying process evaluations in the literature; there remains a methodological gap in terms of tools to report on and help in identifying process evaluations (as opposed to guidance on conducting process evaluations (Moore 2015)). This review included process evaluation studies that were integrated with outcome evaluation studies, that were presented as separate sections, or that could be considered stand‐alone evaluations. The tool used to measure risk of bias in process evaluation studies was an amalgamation of two tools used in reviews of process evaluation studies and resulted in a comprehensive assessment (O'Mara‐Eves 2013; Shepherd 2010). We deemed only four studies to have low risk of bias in most domains (Bruzzese 2011; Kintner 2012; Kouba 2012; Mujuru 2011). Of these, only Kintner 2012 could be considered a stand‐alone evaluation, with Mujuru 2011 including defined sections evaluating processes, and Bruzzese 2011 and Kouba 2012 presenting process evaluation data that were more integrated. We classified the latter two studies as process evaluations due to their exploration of process‐related questions using recognised tools and exploration of context and potential mechanisms. The main weakness of the process evaluation studies included is that they lacked breadth and had considered only a single process of importance in‐depth. The impact of these poor quality studies on the QCA is difficult to ascertain, although absence of richer and broader process data may have been a factor as to why we were able to explain only a relatively modest amount of successful implementation via QCA models. A commonly occurring risk of bias among the included process evaluation studies is that the tools and methods of collecting and analysing data were not always deemed to be reliable or credible.

Potential biases in the review process

Current evidence around the introduction of potential bias through restrictions on publication language is mixed, with some recent studies finding no systematic bias in effect size estimates when languages other than English were excluded (Morrison 2012), although many remain concerned that the results of ineffective trials will be submitted to local (non‐English language) journals, leading to the potential for language restrictions and systematic bias (Guyatt 2011). We assessed a potential impact of this restriction by conducting explorations of the impact of publication bias. Imposing a language restriction may also have influenced results of the synthesis of process evaluation data, and may impede the generalisability of results to individuals of non‐English speaking cultures, although we were not able to explore the impact of this decision in this review.

We encountered the following limitations in the review process.

Potential measurement error: we noted variation in the way in which many outcomes were measured, for example, lung function and school absences. Although no 'gold standard' is available for measuring school absences, lack of continuity across studies may reduce the validity of findings. Further, data for both school attendances and healthcare use may be subject to substantial measurement error, for example, we cannot say for certain that all school absences and healthcare visits that were recorded were specifically due to asthma, or were authorised by either the school or the medical centre. Similarly, measurement error may be a factor with some of the covariates used in subgroup analyses, for example, socio‐economic status (SES) can be measured in different ways ‐ through stated household income or evidence of free school meals ‐ although it was not possible to further explore these differences in measurement in the present review.
Effect size transformations: this review sought to include comprehensive trial data within meta‐analytical models, while maintaining construct validity across effect sizes. This often necessitated transforming the data to ensure statistical compatibility, following recommendations within the Cochrane Handbook for Systematic Reviews of Interventions, and undertaking Chinn's transformation (Chinn 2000). Although we have attempted to ensure transparency in fully presenting disaggregated effect sizes alongside those that have been consolidated, and despite sensitivity analyses conducted to ensure the validity of findings, there is potential for these analyses themselves to be confounded, and underlying assumptions around the transformation of effect sizes may not hold with further interrogation. For example, to facilitate transformations, we combined data on SMDs and ORs, although the skewness usually associated with data such as hospitalisations, for example, may not have been fully accounted for in the transformation. This is an important limitation, but it needs to be balanced against research wastage and information lost by excluding studies that use different approaches in measuring outcomes. Encountering such diverse data reinforces our recommendation below for development of a core outcome set.
Potentially underpowered analyses and treatment of heterogeneity: we included few studies in many of the meta‐analysis models, and for random‐effects models, the models themselves may have been underpowered (Jackson 2017). In addition, when heterogeneity was encountered, the low number of studies meant either that subgroup analyses were unsuitable, or that the subgroups themselves included few studies. We deemed that planned meta‐regression analyses were not suitable for any of the outcomes. Furthermore, unlike many other systematic reviews, we did not present all planned subgroup analyses when we encountered a low number of studies (under 10) and/or a low level of heterogeneity; in this respect, several deviations from the protocol occurred. However, we have greater confidence in the results of subgroup analyses because of our judicious use of these methods.
Identification of process evaluation studies: identification of process evaluations was a challenge in this review. Although guidance is available to assist trialists in conducting a process evaluation (Moore 2015), this did not necessarily aid in the identification of process evaluation studies from a systematic review perspective. All process evaluation studies included an examination of a given process (or processes) and implementation outcome(s) of interest, as well as their relationship to context (in this case, the immediate context of the schools). However, this group spanned a range of studies ‐ from those that were self‐described process evaluations, to those with defined process evaluation sections, to those that included process evaluation data embedded within other evaluation data. Although we developed an inclusive strategy around identification of process evaluation studies, there remains the possibility that some trialists may not have considered their own study as fulfilling the remit of a process evaluation. In addition, although guidance for process evaluations states that they can adopt a range of methods for data collection (Moore 2015), unlike other recent reviews (Dickson 2016), many of the studies that we included did not draw upon robust qualitative methods of data collection, which in turn may have limited our understanding of some of the issues surrounding implementation. Consequently, we deemed there remained greater scope within several of the included studies to explore the way in which the school context, and particularly the broader health service context, influenced delivery of the intervention, and we graded much of the process evaluation information as having high risk of bias because of this weakness. This review highlights the need for greater support for review authors in identifying process evaluation studies. In the current review, our original logic model was instrumental in helping to identify the processes and process metrics of interest and informed the selection of studies (Figure 1); in the absence of clearer guidance in this area, the use of logic models may represent an important step in helping review authors to draw criteria around which processes should be considered in a process evaluation.
Harmful effects: some studies reported negative intervention impacts among children, such as increased levels of ED visits. Such negative effects may reflect the content of self‐management information delivered to children, which may, for example, have recommended greater contact with healthcare providers when experiencing exacerbations (although such detail was not reported in studies), in which case an increase in ED visits could be viewed as a positive. A narrative approach to synthesis of outcome evaluations data could lead to a more nuanced understanding.
Alternative explanations: many other factors might also have influenced review results. For example, although these are school‐based asthma self‐management interventions, few, if any, of the studies considered seasonality of asthma exacerbations and their relationship within the school year. Another Cochrane Review considered the issue of seasonality and showed that seasonal omalizumab treatment between four and six weeks before children return to school might reduce the number of asthma exacerbations seen in autumn (Pike 2018); however the effect of this on outcomes such as asthma control remains unclear.
Low number of clusters: some of the cluster RCTs included in this review randomised only a small number of schools. Although it is universally agreed that randomising one cluster per arm would entirely conflate the randomisation/intervention and clustering effect, there is less agreement on the minimum number of clusters needed for a study to qualify as a cluster RCT (one source recommends four clusters per arm). Studies involving a low number of clusters are generally indicative of a small trial and often contribute only sparse data to any one model. Sensitivity analyses for studies with a low number of clusters per arm were conducted (two or three clusters per arm: Al‐Sheyab 2012; Howell 2005; Kintner 2012; Shah 2001; Velsor‐Friedrich 2005). Results were generally inconclusive and inclusion/exclusion of these studies in models did not qualitatively change results of meta‐analyses, with the exception of Al‐Sheyab 2012 in one quality of life model. These studies may be particularly prone to baseline imbalances, as well as to issues involving introduction of bias, and their inclusion does represent a potential source of bias.

Agreements and disagreements with other studies or reviews

This review is one of the first of its kind to employ a mixed study and mixed methods approach to understanding how school‐based asthma self‐management interventions work, and whether they are effective. It is also the first to undertake quantitative synthesis of studies seeking to develop children's asthma self‐management skills in the school environment. Direct comparisons are challenging, but a number of similar reviews have focused on different settings, different study designs, or use of different synthesis methods, which allows us to understand results in the context of other evidence.

Pinnock 2015 is one of few reviews that have explored how asthma self‐management interventions should be implemented. Review authors focused on a range of settings and age groups and addressed a targeted question around whether interventions primarily targeted at patients, professionals, or the organisation, or explicitly targeting all three levels simultaneously, were differentially effective in changing outcomes, or in changing process measures. They found that complex interventions that explicitly address patient education, professional training, and organisational commitment were associated with improvements in process measures, markers of asthma control, and reduced use of unscheduled health care. Their conclusions that 'individually, the separate components (professional, patient, organisation) of comprehensive self‐management support do not appear to be sufficient consistently to improve outcomes in asthma' (p14) are congruent with our own findings from QCA synthesis, which emphasised that no single condition was necessary and sufficient to trigger successful implementation outside a configuration of conditions.

An earlier Cochrane Review explored the effectiveness of self‐management education interventions for children aged two to 18 with asthma across a range of settings between 1980 and 2002 (Wolf 2002). Review findings were similar to the findings of this review, with data suggestive of moderate reductions in ED visits and in days of restricted activity. This earlier review also found evidence that self‐management education led to a small reduction in school absences, and review authors were able to ascertain a small impact on lung function. It is unclear to what extent the discrepancy in settings, age groups, or inclusion criteria for studies on date would drive the discrepancy in school absence, or another factor. In contrast to the promising results observed for night‐time symptoms in the previous review (Wolf 2002), our review did not find evidence that the intervention made a positive impact, although this was consistent with the findings of a later review that narratively summarised study results (Coffman 2009).

Subsequent reviews include Al Aloola 2014, which focused on primary schools and used a narrative approach to synthesise data. Review authors concluded that most studies were suggestive of positive effects, but as was the case in the present review, they were critical of the measurement of outcomes, which varied greatly among included studies. They also highlighted lack of detail in the descriptions provided for intervention content and processes, which is consistent with the outcome evaluations included here. Ahmad 2011 also took a narrative approach to synthesising outcome data from studies that involved school nurses, but nevertheless concluded that results indicated that a decrease in school absences could be expected, but that results for reductions in ED visits and hospital admissions were less certain, in contrast to the results provided here. Coffman 2009 also undertook a narrative descriptive synthesis of the effectiveness of a school‐based approach, although review authors concluded that there was heterogeneity in the direction and/or magnitude of effect on quality of life, symptom days, night‐time symptoms, and school absences, which largely corroborates the findings of the present review. Finally, a more recent review included school‐based self‐management interventions provided across a diffuse set of studies with regards to design (Carvalho 2016). These review authors also took a narrative approach when synthesising study results, and again showed an overall trend suggestive of heterogeneity in magnitude and direction of effect across a range of outcomes.

This systematic review makes a contribution to the literature by providing the first meta‐analyses of asthma self‐management interventions focused in schools, and it provides evidence of the effectiveness of this approach in reducing healthcare usage. Methodologically, this is also one of the first Cochrane Reviews to employ a mixed methods approach in synthesising evidence. This mixed methods approach helped to show that although intervention as a whole did not appear to be effective in reducing school absences, interventions that were drawing upon theory were effective in improving school absences.

Figure 1

Logic model of school‐based asthma interventions.

Figure 2

Process evaluation study flow diagram.

Figure 3

Outcome evaluation study flow diagram.

Figure 4

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

Figure 5

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Figure 6

Forest plot of comparison: 1 School‐based asthma interventions vs usual care: outcome: 1.1. Exacerbations leading to hospitalisation.

Figure 7

Forest plot of comparison: 1. Effect of school‐based asthma interventions vs usual care, outcome: 1.2. Exacerbations leading to emergency department (ED) visits.

Figure 8

Forest plot of comparison: 2. Effect of school‐based asthma interventions vs usual care subgrouped by school type, outcome: 2.1. Exacerbations leading to emergency department (ED) visits.

Figure 9

Forest plot of comparison: 1. Effect of school‐based asthma interventions vs usual care, outcome: 1.3. Absence from school.

Figure 10

Forest plot of comparison: 2. Effect of school‐based asthma interventions vs usual care subgrouped by school type, outcome: 2.2. Absence from school.

Figure 11

Forest plot of comparison: 10. Effect of school‐based asthma interventions vs usual care subgrouped by configuration of conditions (iI), outcome: 10.3. Absence from school.

Figure 12

Forest plot of comparison: 1. Effect of school‐based asthma interventions vs usual care, outcome: 1.4. Days of restricted activity.

Figure 13

Forest plot of comparison: 1. Effect of school‐based asthma interventions vs usual care, outcome: 1.5. Unplanned visit to hospital or GP due to asthma symptoms.

Figure 14

Forest plot of comparison: 1. Effect of school‐based asthma interventions vs usual care, outcome: 1.6. Experience of daytime and night‐time symptoms ‐ daytime symptoms.

Figure 15

Forest plot of comparison: 1. Effect of school‐based asthma interventions vs usual care, outcome: 1.8. Use of reliever therapies, e.g. beta₂‐agonists.

Figure 16

Forest plot of comparison: 1. Effect of school‐based asthma interventions vs usual care, outcome: 1.9. Corticosteroid dosage and/or use of add‐on therapies (usage of).

Figure 17

Forest plot of comparison: 1. Effect of school‐based asthma interventions vs usual care, outcome: 1.11. Health‐related quality of life (SMD).

Figure 18

Forest plot of comparison: 1. Effect of school‐based asthma interventions vs usual care, outcome: 1.13. Withdrawal from the study.

Figure 19

Forest plot of comparison: 2. Effect of school‐based asthma interventions vs usual care subgrouped by school type, outcome: 2.3. Withdrawal from the study.

Figure 20

Forest plot of comparison: 11. Adjunct analyses ‐ impact of Implementation on selected outcomes, outcome: 11.1. Exacerbations leading to emergency department (ED) visits.

Figure 21

Forest plot of comparison: 11. Adjunct analyses ‐ impact of Implementation on selected outcomes, outcome: 11.2. Absence from school.

Figure 22

Analysis 1.1

Comparison 1 Effects of school‐based asthma interventions vs usual care, Outcome 1 Exacerbations leading to hospitalisation.

Analysis 1.2

Comparison 1 Effects of school‐based asthma interventions vs usual care, Outcome 2 Exacerbations leading to emergency department (ED) visits.

Analysis 1.3

Comparison 1 Effects of school‐based asthma interventions vs usual care, Outcome 3 Absence from school.

Analysis 1.4

Comparison 1 Effects of school‐based asthma interventions vs usual care, Outcome 4 Days of restricted activity.

Analysis 1.5

Comparison 1 Effects of school‐based asthma interventions vs usual care, Outcome 5 Unplanned visit to hospital or GP due to asthma symptoms.

Analysis 1.6

Comparison 1 Effects of school‐based asthma interventions vs usual care, Outcome 6 Experience of daytime and night‐time symptoms ‐ daytime symptoms.

Analysis 1.7

Comparison 1 Effects of school‐based asthma interventions vs usual care, Outcome 7 Experience of daytime and night‐time symptoms ‐ night‐time symptoms.

Analysis 1.8

Comparison 1 Effects of school‐based asthma interventions vs usual care, Outcome 8 Use of reliever therapies, e.g. beta₂‐agonists.

Analysis 1.9

Comparison 1 Effects of school‐based asthma interventions vs usual care, Outcome 9 Corticosteroid dosage and/or use of add‐on therapies (usage of).

Analysis 1.10

Comparison 1 Effects of school‐based asthma interventions vs usual care, Outcome 10 Corticosteroid dosage and/or use of add‐on therapies (appropriate usage of).

Analysis 1.11

Comparison 1 Effects of school‐based asthma interventions vs usual care, Outcome 11 Health‐related quality of life (SMD).

Analysis 1.12

Comparison 1 Effects of school‐based asthma interventions vs usual care, Outcome 12 Health‐related quality of life (MD).

Analysis 1.13

Comparison 1 Effects of school‐based asthma interventions vs usual care, Outcome 13 Withdrawal from the study.

Analysis 2.1

Comparison 2 Effects of school‐based asthma interventions vs usual care subgrouped by school type, Outcome 1 Exacerbations leading to emergency department (ED) visits.

Analysis 2.2

Comparison 2 Effects of school‐based asthma interventions vs usual care subgrouped by school type, Outcome 2 Absence from school.

Analysis 2.3

Comparison 2 Effects of school‐based asthma interventions vs usual care subgrouped by school type, Outcome 3 Withdrawal from the study.

Analysis 3.1

Comparison 3 Effects of school‐based asthma interventions vs usual care subgrouped by age of children, Outcome 1 Exacerbations leading to emergency department (ED) visits.

Analysis 3.2

Comparison 3 Effects of school‐based asthma interventions vs usual care subgrouped by age of children, Outcome 2 Absence from school.

Analysis 3.3

Comparison 3 Effects of school‐based asthma interventions vs usual care subgrouped by age of children, Outcome 3 Withdrawal from the study.

Analysis 4.1

Comparison 4 Effects of school‐based asthma interventions vs usual care subgrouped by child socio‐economic status (SES), Outcome 1 Exacerbations leading to emergency department (ED) visits.

Analysis 4.2

Comparison 4 Effects of school‐based asthma interventions vs usual care subgrouped by child socio‐economic status (SES), Outcome 2 Absence from school.

Analysis 4.3

Comparison 4 Effects of school‐based asthma interventions vs usual care subgrouped by child socio‐economic status (SES), Outcome 3 Withdrawal from the study.

Analysis 5.1

Comparison 5 Effects of school‐based asthma interventions vs usual care subgrouped by involvement of school staff in direct delivery of self‐management skills to children, Outcome 1 Exacerbations leading to emergency department (ED) visits.

Analysis 5.2

Analysis 5.3

Analysis 6.1

Comparison 6 Effects of school‐based asthma interventions vs usual care subgrouped by explicit use of theory, Outcome 1 Exacerbations leading to emergency department (ED) visits.

Analysis 6.2

Comparison 6 Effects of school‐based asthma interventions vs usual care subgrouped by explicit use of theory, Outcome 2 Absence from school.

Analysis 6.3

Comparison 6 Effects of school‐based asthma interventions vs usual care subgrouped by explicit use of theory, Outcome 3 Withdrawal from the study.

Analysis 7.1

Comparison 7 Effects of school‐based asthma interventions vs usual care subgrouped by whether design included active inclusion or participation of parents, Outcome 1 Exacerbations leading to emergency department (ED) visits.

Analysis 7.2

Comparison 7 Effects of school‐based asthma interventions vs usual care subgrouped by whether design included active inclusion or participation of parents, Outcome 2 Absence from school.

Analysis 7.3

Comparison 7 Effects of school‐based asthma interventions vs usual care subgrouped by whether design included active inclusion or participation of parents, Outcome 3 Withdrawal from the study.

Analysis 8.1

Comparison 8 Effects of school‐based asthma interventions vs usual care subgrouped by timing of intervention, Outcome 1 Exacerbations leading to emergency department (ED) visits.

Analysis 8.2

Comparison 8 Effects of school‐based asthma interventions vs usual care subgrouped by timing of intervention, Outcome 2 Absence from school.

Analysis 8.3

Comparison 8 Effects of school‐based asthma interventions vs usual care subgrouped by timing of intervention, Outcome 3 Withdrawal from the study.

Analysis 9.1

Comparison 9 Effects of school‐based asthma interventions vs usual care subgrouped by configuration of conditions, Outcome 1 Exacerbations leading to emergency department (ED) visits.

Analysis 9.2

Comparison 9 Effects of school‐based asthma interventions vs usual care subgrouped by configuration of conditions, Outcome 2 Absence from school.

Analysis 9.3

Comparison 9 Effects of school‐based asthma interventions vs usual care subgrouped by configuration of conditions, Outcome 3 Withdrawal from the study.

Analysis 10.1

Comparison 10 Effects of school‐based asthma interventions vs usual care subgrouped by number of consistent conditions (use of theory, parental involvement, not in own time), Outcome 1 Exacerbations leading to emergency department (ED) visits.

Analysis 10.2

Analysis 10.3

Analysis 11.1

Comparison 11 Adjunct analyses ‐ impact of Implementation on selected outcomes, Outcome 1 Exacerbations leading to emergency department (ED) visits.

Analysis 11.2

Comparison 11 Adjunct analyses ‐ impact of Implementation on selected outcomes, Outcome 2 Absence from school.

Analysis 12.1

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 1 Exacerbations leading to hospitalisation ‐ standardised mean difference.

Analysis 12.2

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 2 Exacerbations leading to hospitalisation ‐ odds ratio.

Analysis 12.3

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 3 Exacerbations leading to hospitalisation ‐ harmonised effect sizes.

Analysis 12.4

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 4 Exacerbations leading to emergency department (ED) visits ‐ standardised mean difference.

Analysis 12.5

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 5 Exacerbations leading to emergency department (ED) visits ‐ odds ratio.

Analysis 12.6

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 6 Exacerbations leading to emergency department (ED) visits ‐ harmonised effect sizes.

Analysis 12.7

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 7 Absence from school ‐ standardised mean difference.

Analysis 12.8

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 8 Absence from school ‐ odds ratio.

Analysis 12.9

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 9 Absence from school ‐ harmonised effect sizes.

Analysis 12.10

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 10 Days of restricted activity ‐ standardised mean difference.

Analysis 12.11

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 11 Days of restricted activity ‐ odds ratio.

Analysis 12.12

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 12 Days of restricted activity ‐ harmonised effect sizes.

Analysis 12.13

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 13 Experience of daytime and night‐time symptoms ‐ daytime symptoms ‐ standardised mean difference.

Analysis 12.14

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 14 Experience of daytime and night‐time symptoms ‐ daytime symptoms ‐ odds ratio.

Analysis 12.15

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 15 Experience of daytime and night‐time symptoms ‐ daytime symptoms ‐ harmonised effect sizes.

Analysis 12.16

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 16 Experience of daytime and night‐time symptoms ‐ night‐time symptoms ‐ standardised mean difference.

Analysis 12.17

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 17 Experience of daytime and night‐time symptoms ‐ night‐time symptoms ‐ odds ratio.

Analysis 12.18

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 18 Experience of daytime and night‐time symptoms ‐ night‐time symptoms ‐ harmonised effect sizes.

Analysis 12.19

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 19 Use of reliever therapies, e.g. beta₂‐agonists ‐ odds ratio.

Analysis 12.20

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 20 Corticosteroid dosage and/or use of add‐on therapies (usage of).

Analysis 12.21

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 21 Corticosteroid dosage and/or use of add‐on therapies (appropriate usage of).

Analysis 12.22

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 22 Health‐related quality of life ‐ standardised mean difference.

Analysis 12.23

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 23 Health‐related quality of life (MD).

Analysis 12.24

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 24 Unplanned visit to hospital or GP due to asthma symptoms ‐ standardised mean difference.

Analysis 12.25

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 25 Unplanned visit to hospital or GP due to asthma symptoms ‐ odds ratio.

Analysis 12.26

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 26 Unplanned visit to hospital or GP due to asthma symptoms ‐ harmonised effect sizes.

Analysis 12.27

Comparison 12 Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes, Outcome 27 Withdrawal from the study.

Summary of findings for the main comparison. Effects of school‐based asthma interventions compared to usual care for asthma among children and adolescents

Effects of school‐based asthma interventions compared to usual care for asthma among children and adolescents
Patient or population: asthma among children and adolescents Setting: primary/elementary schools through to high/senior schools Intervention: effects of school‐based asthma interventions Comparison: usual care
Outcomes	*Anticipated absolute effects (95% CI)**		Relative effect (95% CI)	No. of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with usual care	Risk with effect of school‐based asthma interventions
Exacerbations leading to hospitalisation (hospitalisations) assessed with RCT Follow‐up: range 1 week to 12 months		Mean level of hospitalisation at post‐treatment in the intervention group was 0.19 standard deviations lower than in the control group (0.35 to 0.04 lower)	‐	1873 (6 RCTs)	⊕⊕⊕⊝ MODERATE^a	Meta‐analysis based on SMD including data transformed from OR (data on median level from Gerald 2006 not included)
Asthma symptoms leading to emergency hospital visits (ED visits) Follow‐up: range 1 week to 12 months	Less than 10% experience ED visit annually		OR 0.70 (0.53 to 0.92)	3883 (13 RCTs)	⊕⊕⊝⊝ LOW^b	Data from Gerald 2006 on median visits not combined Assumed risk based on rates over 12 months < 10% based on Horner 2008, McGhan 2010, Velsor‐Friedrich 2005 ≥ 10% based on Cicutto 2013, McGhan 2003
	75 per 1000	54 per 1000 (41 to 69)
	Over 10% experience ED visit annually
	281 per 1000	215 per 1000 (172 to 264)
Unplanned visit to hospital or GP due to asthma symptoms (unplanned medical visits) Follow‐up: range 1 week to 12 months	Unplanned visits over 6 to 9 months		OR 0.74 (0.60 to 0.90)	3283 (5 RCTs)	⊕⊕⊕⊝ MODERATE^c	Unplanned visits over 6 to 9 months based on McGhan 2003, Splett 2006; unplanned visits over 12 months based on Cicutto 2013, McGhan 2010
	264 per 1000	210 per 1000 (177 to 244)
	Unplanned visits over 12 months
	318 per 1000	257 per 1000 (219 to 296)
Absence from school Follow‐up: range 1 week to 15 months	Mean absence from school was 4.3 school days missed annually	MD 0.399 school days missed annually lower (1.254 lower to 0.456 higher)	‐	4609 (10 RCTs)	⊕⊕⊝⊝ LOW^d	Meta‐analysis based on SMD including data transformed from OR; transformation to mean difference undertaken based on data from Cicutto 2005
Experience of daytime and night‐time symptoms ‐ daytime symptoms (daytime symptoms) Follow‐up: range 2 months to 12 months	Mean experience of daytime and night‐time symptoms ‐ daytime symptoms was 3.3 days experienced in past 2 weeks	MD 0.377 days experienced in past 2 weeks lower (0.828 lower to 0.05 higher)	‐	1065 (5 RCTs)	⊕⊕⊕⊝ MODERATE^e	CI for this pooled estimate crossed the line of no effect by a small margin. Original meta‐analysis based on SMDs, including transformations from ORs. SMD to MD based on Bruzzese 2011
Use of reliever therapies, e.g. beta₂‐agonists (reliever therapies) Follow‐up: range 1 week to 15 months	Study population		OR 0.52 (0.15 to 1.81)	437 (2 RCTs)	⊕⊝⊝⊝ VERY LOW^f
	228 per 1000	133 per 1000 (42 to 349)
Health‐related quality of life Follow‐up: range 1 week to 12 months	Mean health‐related quality of life was 4.96 Paediatric Asthma Quality of Life Questionnaire points	MD 0.36 Paediatric Asthma Quality of Life Questionnaire points higher (0.06 higher to 0.64 higher)	‐	2587 (7 RCTs)	⊕⊕⊕⊝ MODERATE^g	Two studies provided information on change in QoL. Both showed positive intervention effects. Risk with usual care based on follow‐up scores
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; ED: emergency department; GP: general practitioner; MD: mean difference; OR: odds ratio; QoL: quality of life; RCT: randomised controlled trial; RR: risk ratio; SMD: standardised mean difference.
GRADE Working Group grades of evidence. High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.
^aStudies with high or unclear risk of bias contribute the least to the overall effect size. Hospitalisations may be due to reasons other than asthma (‐1 for indirectness). ^bFour studies had high risk of bias around allocation concealment; four also had high risk of bias around attrition; many other studies had unclear risks of bias. However, these risks did not appear to inflate the effect size nor systematically influence the effect. A high degree of inconsistency was evident, as measured by heterogeneity statistics in the meta‐analysis, which was partially explained by subgroup analyses. A large degree of variation was evident in measurement of the outcome, prompting concerns about indirectness; similarly, wide confidence intervals were detected (0.53 to 0.95). Study results led to concerns that not all ED visits may be due to asthma (‐1 for inconsistency; ‐1 for indirectness). ^cNo guarantee that unplanned medical visits were due to asthma (‐1 for indirectness). ^dSchool absences could be due to causes other than asthma; heterogeneity statistics suggested a large degree of statistical inconsistency (‐1 for indirectness; ‐1 for inconsistency). ^eHigh risk of bias detected in at least one domain for two out of five studies, which accounted for around a third of the pooled effect size. This included high risk of bias suspected for attrition bias in one study (‐1 for risk of bias). ^fRisk of bias deemed high for attrition and reporting bias for one of the two studies included in the meta‐analysis; very wide confidence interval; although both studies were consistent in the direction of effect, they showed large differences in the magnitude of effect (‐1 for risk of bias; ‐1 for inconsistency; ‐1 for imprecision). ^gImprecision was deemed to be serious based on the nature of the outcome; five of the seven studies were deemed to have high risk of bias in at least one domain. This included three studies deemed to have high risk of bias for allocation concealment. However, these did not appear to differentially influence the effect size (‐1 for imprecision).

Summary of findings for the main comparison. Effects of school‐based asthma interventions compared to usual care for asthma among children and adolescents

Table 1. Detailed coding framework for conditions and outcomes

	Field	Instructions for extractors	Coding values and method
	Setting and participants
1	Number of children	Recorded total number of children involved in intervention	Transformational assignment implemented to condition, reflecting whether it was a ‘large intervention’. Interventions with 15 or fewer children = 0; interventions with 90 children = 0.5; interventions with 300 or more children = 1. Other values fell between 0 and 1
2	Multiple settings	Evidence if delivered at more than 1 school	Direct assignment: yes (mentioned) = 1; no evidence = 0
3	Single sex school	Evidence if delivered at a single sex school	Direct assignment: yes (mentioned) = 1; no evidence = 0
4	Type of school	High school; primary/elementary; junior/middle; other Variable transformed to reflect whether the intervention took place at a high school	Direct assignment: high school = 1; middle/junior = 0.66; elementary/primary = 0.33; missing = 0.5; mixture of high schools and middle schools = 0.75
5	Ethnicity of children	Whether minority ethnic children were targeted/represented. Actual proportions recorded where possible	Transformational assignment Interventions with 25% or fewer children from ethnic minority = 0; interventions with 33.3% of children from ethnic minority = 0.5; interventions with 50% or more children from ethnic minority = 1 When value is missing (and no qualitative statement supports assumption of targeting), assume that this is ‘probably not’ – i.e. probably not targeted – input value of 0.25
6	Socio‐economic status of children	Whether children from lower socio‐economic groups were targeted/represented Actual proportions recorded where possible Indicators included parents with low levels of education; low household income; receipt of free school meals	Transformational assignment Interventions with 25% or fewer children from low socio‐economic groups = 0; interventions with 33.3% of children from low socio‐economic groups = 0.5; interventions with 50% or more children from low socio‐economic groups = 1 Where value is missing (and no qualitative statement supports assumption of targeting), assume that this is ‘probably not’ – i.e. probably not targeted – input value of 0.25
7	Child age	Age groups/classes targeted: ages 5 to 10	Direct assignment: yes (mentioned) = 1; no evidence = 0
8		Age groups/classes targeted: ages 11 to 14	Direct assignment: yes (mentioned) = 1; no evidence = 0
9		Age groups/classes targeted: ages 15 to 18	Direct assignment: yes (mentioned) = 1; no evidence = 0
10	Direct recipients	Children directed recipients	Direct assignment: yes (mentioned) = 1; no evidence = 0
11		Teachers directed recipients	Direct assignment: yes (mentioned) = 1; no evidence = 0
12		Parents directed recipients	Direct assignment: yes (mentioned) = 1; no evidence = 0
13		School nurses directed recipients	Direct assignment: yes (mentioned) = 1; no evidence = 0
	Programme design
14	Theory driven	Did the study name a theoretical framework that underpins the intervention design or delivery style?	Direct assignment: yes (mentioned) = 1; no evidence = 0
15	Intensity of the programme	Coded initially as follows: high intensity = 6+ sessions (group and individual); medium intensity = 3 to 5 sessions; low intensity/no evidence of med/high = 1 to 2 sessions; unclear. Variable transformed to reflect whether the intervention was of high intensity	Direct assignment: high intensity = 1, medium intensity = 0.66; low intensity = 0.33. When no evidence on intensity of intervention was included (1 study = (Richmond 2011)), this was coded as 0.33 (no evidence of high intensity) – interpreted as no evidence of high intensity; for Splett (Splett 2006), such is the degree of personalisation/tailoring that 0.5 was selected as the intensity – each individual session was personalised and lengthy
16	Personalisation/tailoring	Did the programme include individual sessions or use personalisation in any way to alter curriculum to individual students’ needs?	Direct assignment: yes, all sessions implemented were personalised/tailored = 1; some sessions were personalised/tailored = 0.66; personalisation/tailored sessions account for only a minor component = 0.5; no evidence, only generic group sessions implemented = 0 Note that this was personalised by or individual sessions were held with an instructor (included guided online sessions); self‐study components including homework were not included here
17	Timing of the intervention	Did the intervention interfere with the child’s own time (during lunch or after school)?	Direct assignment: yes, all sessions did = 1; yes, but not all sessions = 0.75; missing data = 0.5; described as not interfering with child’s own time = 0
18		Did the intervention interfere with the child’s lessons/other education?	Direct assignment: yes, all sessions did = 1; yes, but not all sessions = 0.75; missing data = 0.5; described as not interfering with child’s lessons/other education = 0
19	Information about control condition	Described whether trialists were also providing a control for the main intervention (intended to capture complexity of running an intervention and a control)	Direct assignment: yes, an equivalent control = 1; yes, but not an equivalent = 0.66; no control described = 0
20	Instructor or facilitator	Teacher	Direct assignment: yes, main instructor = 1; secondary instructor or facilitator = 0.66; not mentioned as an instructor/facilitator = 0
21		Peer	Direct assignment: yes, main instructor = 1; secondary instructor or facilitator = 0.66; not mentioned as an instructor/facilitator = 0
22		School nurse	Direct assignment: yes, main instructor = 1; secondary instructor or facilitator = 0.66; not mentioned as an instructor/facilitator = 0
23		Self‐directed/child‐directed	Direct assignment: yes, main instructor = 1; secondary instructor or facilitator = 0.66; not mentioned as an instructor/facilitator = 0
24		Parent	Direct assignment: yes, main instructor = 1; secondary instructor or facilitator = 0.66; not mentioned as an instructor/facilitator = 0
25		Other	Direct assignment: yes, main instructor = 1; secondary instructor or facilitator = 0.66; not mentioned as an instructor/facilitator = 0
	Programme content
26	Curriculum	Lung physiology/asthma biology	Direct assignment: yes (mentioned) = 1; no evidence = 0
27		Asthma acceptance/asthma into identity	Direct assignment: yes (mentioned) = 1; no evidence = 0
28		Symptom monitoring and correct medication use	Direct assignment: yes (mentioned) = 1; no evidence = 0
30		Avoiding triggers	Direct assignment: yes (mentioned) = 1; no evidence = 0
31		General health including exercise	Direct assignment: yes (mentioned) = 1; no evidence = 0
32		Strengthening alliances including asthma action plans with primary care providers	Direct assignment: yes (mentioned) = 1; no evidence = 0
33		Specific focus on smoking	Direct assignment: yes (mentioned) = 1; no evidence = 0
34		Personalised/tailored (individualised)	Direct assignment: yes (mentioned) = 1; no evidence = 0
35		School performance	Direct assignment: yes (mentioned) = 1; no evidence = 0
36		Emergencies	Direct assignment: yes (mentioned) = 1; no evidence = 0
37		Unknown	Direct assignment: yes (mentioned) = 1; no evidence = 0
38		Specific focus on breathing/relaxation techniques	Direct assignment: yes (mentioned) = 1; no evidence = 0
39	Learning styles	Problem‐solving component	Direct assignment: yes (mentioned) = 1; no evidence = 0
40		Self‐directed (including homework) component	Direct assignment: yes (mentioned) = 1; no evidence = 0
41		Peer delivery component	Direct assignment: yes (mentioned) = 1; no evidence = 0
42		Interactive (non‐didactic) components	Direct assignment: yes (mentioned) = 1; no evidence = 0
43		Didactic components	Direct assignment: yes (mentioned) = 1; no evidence = 0
44		Other style/unclear	Direct assignment: yes (mentioned) = 1; no evidence = 0
45	Programme ethos/aims	Emphasis on social benefit	Direct assignment: yes (mentioned) = 1; no evidence = 0
46		Emphasis on improving well‐being	Direct assignment: yes (mentioned) = 1; no evidence = 0
47		Emphasis on having fun	Direct assignment: yes (mentioned) = 1; no evidence = 0
48		Emphasis on fostering independence/personal responsibility	Direct assignment: yes (mentioned) = 1; no evidence = 0
49		Emphasis on developing children's knowledge	Direct assignment: yes (mentioned) = 1; no evidence = 0
50		Emphasis on collaboration	Direct assignment: yes (mentioned) = 1; no evidence = 0
51		Emphasis on tailoring for specific group needs	Direct assignment: yes (mentioned) = 1; no evidence = 0
52		Emphasis on breathing technique	Direct assignment: yes (mentioned) = 1; no evidence = 0
53		Unclear	Direct assignment: yes (mentioned) = 1; no evidence = 0
54	Additional components – school asthma policy	Additional support provided for developing school policy	Direct assignment: yes (mentioned) = 1; no evidence = 0
55		School asthma policy developed organically	Direct assignment: yes (mentioned) = 1; no evidence = 0
	Additional processes undertaken – planned and unplanned
56	Recruitment methods ‐ school	Ad hoc/convenience sample of schools	Direct assignment: yes (mentioned) = 1; no evidence = 0
57		Census of school district (all schools invited and potentially eligible)	Direct assignment: yes (mentioned) = 1; no evidence = 0
58		Unspecified methods of school recruitment	Direct assignment: yes (mentioned) = 1; no evidence = 0
59	Additional processes to improve/attenuate attrition/enrolment	Marketing materials sent to parents	Direct assignment: yes (mentioned) = 1; no evidence = 0
60		Low motivation of students acknowledged and addressed	Direct assignment: yes (mentioned) = 1; no evidence = 0 Note that 1 study received a value of 0.75, as low motivation was acknowledged but was not explicitly described as being addressed (Magzamen 2008)
61		Incentives used (child or parent)	Direct assignment: yes (mentioned) = 1; no evidence = 0 Incentives for teachers and no evidence for children/teachers coded as 0.5
62		Make‐up/catch‐up sessions provided	Direct assignment: yes (mentioned) = 1; no evidence = 0
63		Reminders sent to parents/children	Direct assignment: yes (mentioned) = 1; no evidence = 0
64	Relationships/engagement	Did teachers engage or participate in the way they were expected to?	Direct assignment: yes, good reported throughout = 1; yes, some weaker evidence of good relationships or evidence that relationships improved during the course of the intervention = 0.75; missing, not applicable, or undetermined = 0.5; no, some weaker evidence of poorer relationships or evidence that relationships deteriorated during the course of the intervention = 0.25; evidence of poor relationships throughout = 0
65		Did parents engage or participate in the way they were expected to?	Direct assignment: yes, good reported throughout = 1; yes, some weaker evidence of good relationships or evidence that relationships improved during the course of the intervention = 0.75; missing, not applicable, or undetermined = 0.5; no, some weaker evidence of poorer relationships or evidence that relationships deteriorated during the course of the intervention = 0.25; evidence of poor relationships throughout = 0 One study described good levels of engagement, but review authors assigned value of 0.25 as a third of parents did not engage as expected (Kintner 2012); similar rationale for Mujuru 2011
66		Did school nurses engage or participate in the way they were expected to?	Direct assignment: yes, good reported throughout = 1; yes, some weaker evidence of good relationships or evidence that relationships improved during the course of the intervention = 0.75; missing, not applicable, or undetermined = 0.5; no, some weaker evidence of poorer relationships or evidence that relationships deteriorated during the course of the intervention = 0.25; evidence of poor relationships throughout = 0
67		Did other relevant stakeholders engage or participate in the way they were expected to?	Direct assignment: yes, good reported throughout = 1; yes, some weaker evidence of good relationships or evidence that relationships improved during the course of the intervention = 0.75; missing, not applicable, or undetermined = 0.5; no, some weaker evidence of poorer relationships or evidence that relationships deteriorated during the course of the intervention = 0.25; evidence of poor relationships throughout = 0
	Process outcomes
68	Child satisfaction	Put in level of satisfaction (%) or record qualitative statement on child satisfaction with the intervention experience. Indicators of satisfaction include children reporting that they enjoyed the intervention; whether the children would recommend the intervention to others; whether children found the intervention helpful. Knowledge development should not be included here	Elements of direct and transformational assignment included here [First] Direct assignment: where there is a qualitative statement indicating positive agreement, assign value of 0.66; where a qualitative statement indicating negative agreement, assign value of 0.33; where no child satisfaction data were collected or data were missing, assign value of 0.5 [Second; including of direct above] Transformational assignment implemented to condition reflecting whether children were satisfied. Interventions with 25% or fewer children satisfied = 0; interventions with 50% of children satisfied = 0.5; missing data coded as 0.5; interventions with 75% or more children satisfied See text for further justification on use of the 75% threshold
69	Child attrition (overall level)	Put in level of completion (%) or record qualitative statement on child completion rate	Elements of direct and transformational assignment here. Note thresholds were higher than for satisfaction, as fewer data were missing [First] Direct assignment: where there is a qualitative statement indicating high level of completion, assign value of 0.83; where a qualitative statement indicating problematic completion, assign value of 0.66. Where data are missing, assign value of 0.75 [Second; including of direct above] Transformational assignment implemented to condition reflecting level of completion. Interventions with 66% or fewer children completing the intervention = 0; interventions with 75% of children completing the intervention = 0.5; interventions with 83% or more children completing the intervention = 1. Missing data coded as 0.5 See text for further justification on the use of thresholds
70	Child dosage level	Did the children receive the intended dosage of the intervention? Put in level of dosage (%) or record qualitative statement on child dosage.	Elements of direct and transformational assignment here. Note thresholds are higher than for satisfaction, as fewer data are missing [First] Direct assignment: where there is a qualitative statement indicating high level of dosage, assign value of 0.83; where a qualitative statement indicating problematic dosage, assign value of 0.66. Where data are missing, assign value of 0.75 [Second; including of direct above] Transformational assignment implemented to condition reflecting level of dosage. Interventions with 66% or fewer children receiving the full dosage = 0; interventions with 75% of children receiving the full dosage = 0.5; interventions with 83% or more of children receiving the full dosage = 1. Missing data coded as 0.5 See text for further justification on the use of thresholds
71	Child adherence	Did the children adhere to the intervention instructions, e.g. students being compliant with paperwork; completing homework; going to visit PCPs as instructed, etc. Put in level of adherence (%) or record qualitative statement on child dosage	Elements of direct and transformational assignment here. Note thresholds are higher than for satisfaction as fewer data are missing [First] Direct assignment: where there is a qualitative statement indicating high level of adherence, assign value of 0.83; where a qualitative statement indicating problematic adherence, assign value of 0.66. Where data are missing, assign value of 0.75 [Second; including of direct above] Transformational assignment implemented to condition reflecting level of adherence. Interventions with 66% or fewer children adherent = 0; interventions with 75% of children adherent = 0.5; interventions with 83% or more children adherent = 1. Missing data coded as 0.5 See text for further justification on the use of thresholds
72	Consolidated process variable	Summation of attrition, adherence, and dosage scores as a marker of implementation success	Transformational assignment Score of 0 = 0 implementation not successful; score of 1.5 = mid point between successful and unsuccessful implementation; score of 3 = full implementation success

Table 1. Detailed coding framework for conditions and outcomes

Table 2. Original and reduced conditions for curriculum content, delivery style, and programme emphasis

Curriculum – original conditions	Curriculum – reduced conditions^a
I. Lung physiology ii. Asthma acceptance iii. Symptom monitoring and treatment iv. Trigger avoidance v. General health vi. Forming alliances vii. Smoking viii. Tailored/personalised ix. School performance x. Emergencies xi. Unknown content	I. Symptom monitoring and alliances ii. Lung physiology and general health iii. Symptom monitoring and trigger avoidance iv. Other various foci v. Unknown
Pedagogical delivery style – original conditions	Pedagogical delivery style – reduced conditions^b
I. Problem‐solving ii. Self‐direct iii. Peer delivery iv. Interactive v. Didactic vi. No information/other focus	I. Interactive focused style ii. Diverse style iii. Unknown style
Intervention emphasis – original conditions	Intervention emphasis – reduced conditions^c
I. Emphasis on social benefit ii. Emphasis on well‐being iii. Emphasis on having fun iv. Emphasis on personal responsibility v. Emphasis on children’s knowledge vi. Emphasis on collaboration vii. Emphasis on tailoring/personalisation viii. Emphasis unclear	I. Emphasis on tailoring/personalisation ii. Emphasis on personal responsibility iii. Diffuse emphasis/other
^aPseudo‐F index = 5.66. ^bPseudo‐F index = 8.36. ^cPseudo‐F index = 6.50.

Table 2. Original and reduced conditions for curriculum content, delivery style, and programme emphasis

Table 3. Data table for QCA model 1 ‐ setting and participants

	Successful intervention	School‐based health centre	High school	Parents directly involved	Teachers received training	School nurses or other stakeholders received training
Joseph 2010	0.52	0.55	1	0	0	0
Kouba 2012	0.33	0.33	1	1	0	0
Dore‐Stites 2007	0.67	0.66	0	1	0	0
Joseph 2013	1.00	0.55	1	0	0	0
Mujuru 2011	0.67	0.66	0	0	1	0
Henry 2004	0.83	0.33	1	0	1	0
Pike 2011	0.67	0.33	0	0	1	0
Spencer 2000	0.33	0.66	0	1	0	0
Engelke 2013	0.50	0.66	0.5	1	1	1
Splett 2006	0.50	1.00	0.5	0	1	1
Kintner 2012	0.83	0.66	1	1	0	1
Berg 2004	0.83	0.66	1	0	0	0
Howell 2005	0.33	0.75	0	1	0	0
Gerald 2006	0.33	0.55	0	0	0	0
Langenfeld 2010	0.33	0.66	0	0	1	0
Al‐Sheyab 2012	0.83	0.33	1	0	0	0
Levy 2006	0.52	0.33	0	0	1	0
Terpstra 2012	1.00	0.66	0.66	1	0	0
Horner 2015	0.67	0.66	0	0	0	0
Bruzzese 2008	0.94	0.66	0.66	1	0	0
Lee 2011	0.50	0.66	0	0	0	0
Bruzzese 2004	0.33	0.55	1	0	0	1
Cicutto 2013	0.67	0.33	0	0	0	1
Brasler 2006	0.00	0.66	0.66	1	0	0
Crane 2014	0.50	0.33	0	0	0	0
Bruzzese 2011	0.88	0.55	1	0	0	1
Magzamen 2008	0.19	0.55	0.75	0	1	0
QCA: qualitative comparative analysis.

Table 3. Data table for QCA model 1 ‐ setting and participants

Table 4. Data table for QCA model 2 ‐ recruitment and retention processes

	Successful intervention	Provision of additional marketing materials	Provision of incentives	Make‐up sessions provided	Reminders provided for attendance at activity
Joseph 2010	0.52	1	1	0	0
Kouba 2012	0.33	1	0	1	0
Dore‐Stites 2007	0.67	1	1	0	0
Joseph 2013	1.00	1	1	0	0
Mujuru 2011	0.67	0	0	0	1
Henry 2004	0.83	0	0	0	0
Pike 2011	0.67	0	0.5	0	0
Spencer 2000	0.33	1	0	0	0
Engelke 2013	0.50	0	0	0	0
Splett 2006	0.50	0	0	0	0
Kintner 2012	0.83	1	1	1	0
Berg 2004	0.83	0	1	0	0
Howell 2005	0.33	0	1	1	1
Gerald 2006	0.33	0	0	0	0
Langenfeld 2010	0.33	0	1	0	0
Al‐Sheyab 2012	0.83	0	0	0	0
Levy 2006	0.52	0	0	0	0
Terpstra 2012	1.00	1	1	1	1
Horner 2015	0.67	0	0	0	0
Bruzzese 2008	0.94	0	0	1	0
Lee 2011	0.50	0	0.75	0	0
Bruzzese 2004	0.33	0	1	0	1
Cicutto 2013	0.67	0	0	1	0
Brasler 2006	0.00	1	1	1	1
Crane 2014	0.50	0	0	0	0
Bruzzese 2011	0.88	0	0	1	0
Magzamen 2008	0.19	1	1	0	1
QCA: qualitative comparative analysis.

Table 4. Data table for QCA model 2 ‐ recruitment and retention processes

Table 5. Data table for QCA model 4 ‐ modifiable design features

	Successful intervention	Theory driven	Personalised or individual sessions	Intervention takes place during lesson time	Intervention takes place during students’ own free time	School nurse involved in delivery of the intervention
Joseph 2010	0.52	1	1	1	0.33	0
Kouba 2012	0.33	1	1	0	1	0
Dore‐Stites 2007	0.67	1	0	0.33	0.33	0.66
Joseph 2013	1.00	1	1	0.75	0.75	0
Mujuru 2011	0.67	0	0	1	0	0
Henry 2004	0.83	0	0	1	0	0
Pike 2011	0.67	0	0	1	0	0
Spencer 2000	0.33	0	1	0.33	0.33	0.66
Engelke 2013	0.50	0	0.66	0.33	0.33	1
Splett 2006	0.50	0	1	0.33	0.33	1
Kintner 2012	0.83	1	0	1	1	0.66
Berg 2004	0.83	1	0.66	0.33	0.33	0.66
Howell 2005	0.33	1	1	0.33	0.33	0.66
Gerald 2006	0.33	0	0	1	0.33	0
Langenfeld 2010	0.33	0	1	0.33	0.33	1
Al‐Sheyab 2012	0.83	1	0	0.33	0.33	0
Levy 2006	0.52	0	0.66	0.33	0.33	1
Terpstra 2012	1.00	1	0	0	1	0.66
Horner 2015	0.67	1	0	0	1	0
Bruzzese 2008	0.94	1	0	0.33	0.33	0.66
Lee 2011	0.50	1	0	1	0	0.66
Bruzzese 2004	0.33	1	1	0.75	0.75	0
Cicutto 2013	0.67	1	0	0	1	0
Brasler 2006	0.00	0	0	0.75	0.75	0.66
Crane 2014	0.50	1	0	0	1	0.66
Bruzzese 2011	0.88	1	1	0.33	0.33	0
Magzamen 2008	0.19	0	0	0	1	1
QCA: qualitative comparative analysis.

Table 5. Data table for QCA model 4 ‐ modifiable design features

Table 6. Data table for QCA model 5 ‐ stakeholder involvement and engagement

	Successful intervention	School asthma policy	Good relationships/engagement with parents	Good relationships/engagement with school nurses	Child Satisfaction	School asthma policy
Joseph 2010	0.52	0	0	0	0	0
Kouba 2012	0.33	0	0	0	0	0
Dore‐Stites 2007	0.67	0	0.75	1	1	0
Joseph 2013	1.00	0	1	0	0	0
Mujuru 2011	0.67	0	0.25	0	0	0
Henry 2004	0.83	1	0	0	0	1
Pike 2011	0.67	0	0	0	0	0
Spencer 2000	0.33	0	1	1	0	0
Engelke 2013	0.50	1	1	0	0	1
Splett 2006	0.50	1	0	1	0	1
Kintner 2012	0.83	0	0.25	0	1	0
Berg 2004	0.83	0	0	0	1	0
Howell 2005	0.33	0	0.75	0.75	0.633333	0
Gerald 2006	0.33	0	0	0	0	0
Langenfeld 2010	0.33	1	0	1	0	1
Al‐Sheyab 2012	0.83	0	0	0	0.633333	0
Levy 2006	0.52	1	0	0	0	1
Terpstra 2012	1.00	0	0.25	0	0	0
Horner 2015	0.67	0	0	0	0	0
Bruzzese 2008	0.94	0	1	0	1	0
Lee 2011	0.50	0	0	0	0	0
Bruzzese 2004	0.33	0	0	0	0.633333	0
Cicutto 2013	0.67	1	0	0	0	1
Brasler 2006	0.00	1	0	1	0.633333	1
Crane 2014	0.50	0	0	1	0	0
Bruzzese 2011	0.88	0	0	0	0	0
Magzamen 2008	0.19	0	0	0	0	0
QCA: qualitative comparative analysis.

Table 6. Data table for QCA model 5 ‐ stakeholder involvement and engagement

Table 7. Included process evaluation studies: methodological characteristics and processes described

Study	Type of study	Approach	Process evaluation elements
Al‐Sheyab 2012a	Feasibility study	Qualitative	Thematic analyses of student perceptions
Berg 2004	Outcome and process evaluation	Qualitative and quantitative	Thematic analyses of student perceptions
Bignall 2015	Feasibility study	Qualitative and quantitative	Thematic analyses of student perceptions
Brasler 2006	Feasibility/case study of implementation	Quantitative data and trialist reports	Implementation challenges and facilitators identified
Bruzzese 2004	Feasibility study	Qualitative and quantitative	Section evaluating intervention reach, dosage, and student satisfaction
Bruzzese 2011	Outcome evaluation with section on process evaluation	Quantitative	Section evaluating intervention reach (dosage)
Bruzzese 2008	Feasibility study	Qualitative and quantitative	Stand‐alone section on process evaluation results assessing implementation and student perceptions
Carpenter 2016	Outcome and process evaluation	Qualitative and quantitative	Thematic analyses of student perceptions
Cicutto 2013	Outcome and process evaluation	(Mainly) Quantitative	In addition to information on other processes of interest, provided a description of wider school support through policy changes (process of interest included in the logic model)
Crane 2014	Feasibility study	Quantitative	Study was included as it represented an implementation study (through focus on the impact of changing dosage schedule)
Dore‐Stites 2007	Feasibility study	Quantitative	In addition to information on other processes of interest, provided information on student satisfaction
Engelke 2013	Feasibility study	Quantitative	Detailed process/implementation information was provided
Gerald 2006	Outcome and process evaluation	(Mainly) Quantitative	In addition to information on other processes of interest, provided a description of implementation challenges
Henry 2004	Outcome and process evaluation	(Mainly) Quantitative	In addition to information on other processes of interest, provided a description of wider school support through policy changes (process of interest in the logic model) and assessment of sustainability
Horner 2015	Outcome evaluation with process evaluation information	Quantitative	Included detailed information on attrition and cost‐effectiveness
Howell 2005	Outcome and process evaluation	Quantitative	In addition to information on other processes of interest, provided information on student satisfaction
Jackson 2006	Outcome evaluation with process evaluation information	Quantitative	In addition to information on other processes of interest, provided information on student satisfaction
Joseph 2010	Outcome and process evaluation	Quantitative	In addition to information on other processes of interest, provided detailed information on non‐adherence
Joseph 2013	Outcome and process evaluation	Quantitative	Included detailed studies of non‐adherence and relationship with student characteristics
Kintner 2012	Feasibility study	Quantitative	In addition to information on other processes of interest, provided information on student satisfaction
Kouba 2012	Outcome evaluation with process evaluation information	Quantitative	In addition to information on other processes of interest, provided detailed information on dosage (and dose‐response)
Langenfeld 2010	Implementation study	Quantitative	In addition to information on other processes of interest, provided detailed information on dosage (and dose‐response)
Lee 2011	Implementation study	Qualitative	In addition to information on other processes of interest, provided detailed information on instructor experiences
Levy 2006	Outcome evaluation with process evaluation information	Quantitative	In addition to information on other processes of interest, provided information on parental adherence to intervention protocol
Magzamen 2008	Outcome evaluation with process evaluation information	Quantitative	In addition to information on other processes of interest, provided information on attrition
McCann 2006	Outcome evaluation with process evaluation information	Quantitative	In addition to information on other processes of interest, provided information on teacher adherence/school level commitment
Mickel 2016	Outcome and process evaluation	Qualitative and quantitative	Thematic analyses of student perceptions
Mujuru 2011	Outcome and process evaluation	(Mainly) Quantitative	In addition to information on other processes of interest, provided a description of parental satisfaction
Pike 2011	Outcome and process evaluation	(Mainly) Quantitative	In addition to information on other processes of interest, provided information on teacher adherence/school level commitment
Richmond 2011	Outcome and process evaluation	(Mainly) Quantitative	Included detailed information on adherence and awareness
Spencer 2000	Outcome and process evaluation	Quantitative	In addition to information on other processes of interest, provided information on instructor satisfaction and school level commitment
Splett 2006	Outcome and process evaluation	Quantitative	In addition to information on other processes of interest, provided information on adherence and school level commitment
Terpstra 2012	Outcome and process evaluation	Quantitative	In addition to information on other processes of interest, represented an implementation study by including a focus on the impact of parental involvement/increasing parental awareness

Table 7. Included process evaluation studies: methodological characteristics and processes described

Table 8. Process evaluation studies ‐ summary of intervention characteristics

	Named theoretical framework	Aim	Intervention type	Control	Intensity	Included in QCA
Al‐Sheyab 2012a	Developmental stages (not named)	To assess feasibility in the Jordanian context of a peer‐led, school‐based asthma education programme	Triple A. Children received education through interactive teaching and learning activities	N/A	14 hours over 6 days	Setting and participants; further modifiable design features; stakeholder involvement and engagement
Berg 2004	Social learning theory	To evaluate effects of the Power Breathing programme and individual coaching sessions on asthma knowledge and functional health status	Power Breathing. Children received education in a group session on asthma management	N/A	2 weeks	Stakeholder involvement and engagement
Bignall 2015	None	To test the feasibility and preliminary efficacy of a school‐based RCT on breathing retraining for asthma outcomes and anxiety symptoms	Single workshop for children. Children received information on relaxation/breathing techniques	30 minutes of standard asthma education	2 face‐to‐face visits 1 month apart	None
Brasler 2006	None	To provide adolescents with knowledge and skills to take control of their asthma; to enhance knowledge and skills of school staff, health professionals, and parents	Power Breathing. Children received basic asthma education and addressed social/lifestyle concerns	N/A	3× 90‐minute or 6× 45‐minute sessions	None
Bruzzese 2004	Self‐regulation theory	To help students weave asthma and management strategies into their self‐identity	ASMA. Students were taught how to manage their asthma to prevent symptoms and reduced quality of life. Continued medical education was also offered to medical providers	Usual care	3 workshops 2 or 3 weeks apart for 8 weeks	Stakeholder involvement and engagement
Bruzzese 2011	Social cognitive theory	To test the efficacy of ASMA	ASMA; academic detailing. Students attended workshops to empower them to manage their asthma. Parents received training on how to support their child's need to manage his or her asthma	Usual care	8‐week programme/3× 45‐minute sessions and individual coaching sessions once a week for 5 weeks	Further modifiable design features
Bruzzese 2008	Social cognitive theory; cognitive‐behavioural therapy	To test the feasibility and short‐term outcomes of asthma: it’s a family affair!	OAS and ASMA; caregiver education. Intervention students received education about asthma, based on existing materials, from coping with asthma at home and at school; OAS and ASMA	Usual care	6× 75‐minute group sessions once a week for 6 weeks; caregiver 5× 90‐minute sessions once a week	Setting and participants; further modifiable design features; stakeholder involvement and engagement
Carpenter 2016	None	To test whether a tailored inhaler technique video intervention could be feasibly implemented by school nurses; to improve the inhaler technique of children with asthma	Multiple sessions for children. Children watched a tailored video and demonstrated their inhaler technique before and after	N/A	6 weeks or less	None
Cicutto 2013	Social cognitive theory	To prepare and support children with asthma to be successful managers of their asthma, thereby reducing school absenteeism, interrupted activity, and health service use	Roaring Adventures of Puff. Workshops included goal‐setting and self‐monitoring, trigger identification, control and avoidance, basic pathophysiology, medication use, symptom recognition, and the asthma action plan, using interactive techniques	Usual care	Unclear	Setting and participants
Crane 2014	Educational theory of Jean Piaget	To pilot a shorter, condensed OAS education programme as an alternative, yet still effective, delivery approach compared to the lengthier original programme	OAS. Children received education from OAS	Non‐equivalent intervention	10 weeks	Setting and participants; further modifiable design features
Dore‐Stites 2007	None	Unclear	OAS; Quest for the Code. Children received a computer game, home activities, and caregiver information	N/A	20 minutes a week for 8 to 9 weeks	Further modifiable design features
Engelke 2013	Case management theory	To identify the process of case management used by school nurses, and when they provide case management to students with asthma. The second aim was to identify the impact of case management on parent perception of how well the child manages illness; parent perception of how well the child keeps up with school work; quality of life and academic achievement of children	Case management; nurse meetings; multiple sessions for children; multiple sessions for staff. Children received education and counselling, and parent/family education was delivered, as well as education and healthcare co‐ordination for teachers/staff	N/A	Unclear	None
Gerald 2006	None	To evaluate a comprehensive school‐based asthma management programme in an inner city, largely African American school system	OAS. The intervention included 3 educational programmes and medical management for children, as well as education for school staff	Usual care	Unclear	None
Henry 2004	Unclear	To determine whether an asthma education programme in schools would have a direct impact on student knowledge and attitudes toward asthma and quality of life of students with asthma; an indirect impact on teacher knowledge and attitudes on asthma and on school policies about asthma; and a sustainable programme after resources were withdrawn	Asthma education. A package about asthma was taught within the PD/H/PE (Personal Development, Health and Physical Education) strand of the school curriculum	Usual care	Unclear	Setting and participants
Horner 2015	Bruhn’s theoretical model of asthma self‐management	To test effects of 2 modes of delivering an asthma educational intervention on health outcomes and asthma management	7‐topic curriculum. The intervention was designed for children in rural areas and included asthma information	In‐school asthma classes	16× 15‐minute sessions for 5 weeks	None
Howell 2005	Learning theory and behaviour modification	To examine whether it was feasible to implement an interactive computer game at school health centres. Second, to examine whether exposure to the game was effective in increasing asthma knowledge, reducing asthma symptoms, and reducing unnecessary healthcare use compared with no exposure to the game	Quest for the Code. Computer game	Usual care	4× 30‐minute sessions	None
Jackson 2006	None	To evaluate knowledge and attitude outcomes of an educational asthma programme for third grade children with and without asthma	Single sessions for children. Children completed an educational programme. Teachers were also encouraged to attend	N/A	3 classes per session for 11 sessions	None
Joseph 2010	None	To develop and evaluate a multi‐media, web‐based asthma management programme	Puff City. A web‐based programme was delivered to children to focus on adherence, inhaler availability, and smoking cessation/reduction	Generic asthma websites	Unclear	Further modifiable design features; stakeholder involvement and engagement
Joseph 2013	Behavioural theory	To evaluate a school‐based RCT to evaluate Puff City	Adapted version of the Puff City computer programme	Generic asthma education	4× 15‐minute sessions	None
Kintner 2012	Lifespan development perspective	To evaluate the feasibility of the SHARP programme for students, their family, school personnel, and community partners	SHARP; Community Coalition component	N/A	Once a week for 10 weeks plus a 3‐hour community component	Setting and participants; further modifiable design features; stakeholder involvement and engagement
Kouba 2012	Orem’s self‐care deficit theory	To determine the effectiveness of the ICAN programme for nutrition knowledge and dietary behaviours	Single workshop for staff; multiple sessions for children; Quest for the Code; Fight Asthma Now; additional nurse meetings; combined education	N/A	8 weeks	None
Langenfeld 2010	None	Unclear	OAS; case management; stand‐alone respiratory therapy. Children received the OAS curriculum and case management asthma strategies developed with teachers	N/A	6× 40‐minute sessions for 1 school year	None
Lee 2011	The functional context approach	To evaluate the effectiveness and feasibility of using undergraduate nursing students as facilitators to deliver an asthma management programme	OAS. Children received the OAS curriculum	N/A	Unclear	Further modifiable design features
Levy 2006	None	To evaluate the effectiveness of a school‐based nurse case management approach to asthma in students with poor control	OAS; monitoring of students; health status. Students received OAS education and weekly monitoring of their health status	Usual care	1 school term	None
Magzamen 2008	None	To evaluate the implementation of Kickin’ Asthma	Multiple sessions for children; Kickin’ Asthma. Educational sessions, similar to the OAS curriculum. Customised letters were also sent home to describe health needs and goals for each child	N/A	3 months	None
McCann 2006	None	To assess whether a school‐based intervention would produce clinical and psychological benefits for children with asthma	Education; role‐play. The intervention focused on describing the respiratory condition through a role‐play	Respiratory education	45‐minute session	None
Mickel 2016	None	To provide Iggy education to more than 75% of children with asthma; To increase asthma knowledge; increase families’ awareness of asthma; and cultivate collaboration between school nurses and asthma providers	Iggy and the Inhalers intervention. Children received an asthma education video, poster, comic book, sticker, and trading card programme	N/A	Unclear	None
Mujuru 2011	None	To demonstrate the feasibility of a school‐based asthma education programme for students and to evaluate parents’ perspectives on the intervention	OAS. Children received the OAS curriculum	N/A	40‐minute session once a week for 2 months	None
Pike 2011	None	To assess student asthma knowledge gain, teacher acceptance, and grade appropriateness after an intervention	Multiple sessions for children; integrated into the curriculum. Teachers taught lessons with information about asthma	Usual care	7 lesson plans	Setting and participants
Richmond 2011	None	To increase the number of current provider‐written asthma action plans submitted to the school nurse at the beginning of the school year	Breathe Your Best. Students were encouraged to receive an asthma action plan from their doctor and to collect their prescriptions	N/A	Unclear	None
Spencer 2000	None	To evaluate the OAS programme for children	OAS. Children received the OAS curriculum	N/A	6× 40‐minute sessions	None
Splett 2006	None	To evaluate the effectiveness and sustainability of the Healthy Learners Asthma Initiative	Children received training on asthma self‐management. Licensed nurses and healthcare assistants received coaching and reinforcement from asthma resource nurses	Usual care	Varied according to asthma severity and need	None
Terpstra 2012	Social cognitive theory	To test a version of an intervention with a caregiver newsletter vs no newsletter	Multiple sessions for children; materials for parents. Children received skills training on how to use a peak flow meter. Parents received a newsletter about an important theme from the research	Intervention or intervention with a newsletter	6‐week training	Setting and participants; further modifiable design features
ASMA: Asthma Self‐Management for Adolescents. ICAN: I Can Control Asthma and Nutrition Now. N/A: not applicable. OAS: Open Airways for Schools. RCT: randomised controlled trial. SHARP: Staying Healthy–Asthma Responsible & Prepared. Triple A: Adolescent Asthma Action.

Table 8. Process evaluation studies ‐ summary of intervention characteristics

Table 9. Process evaluation studies ‐ summary of study design, setting, and population

	Study design	Number of children	Country	Type of School	Recipients	Age of children (years)	Representation of children from BME backgrounds	Representation of children from low SES backgrounds
Al‐Sheyab 2012a	Case study	31	Jordan	High	Children	11 to 18	Unclear	Unclear
Berg 2004	Quasi‐experimental	13	USA	High	Children	15 to 18	46.2% African American	Unclear
Bignall 2015	Parallel‐group RCT	33	USA	High	Children	11 to 18	100% Black or African American	Unclear
Brasler 2006	Case study	342	USA	Junior/middle	Children; teachers; parents	11 to 14	Unclear	Unclear
Bruzzese 2004	Parallel‐group RCT	45	USA	High	Children	11 to 18	Unclear	Unclear
Bruzzese 2011	Parallel‐group RCT	345	USA	High	Children	11 to 18	45.5% Hispanic; 37.7% African American; 11.6% mixed; 5.2% other	75% free school meals
Bruzzese 2008	Parallel‐group RCT	24	USA	Junior/middle	Children; parents	11 to 14	41% Hispanic; 17% White; 8% African American; 34% other	8% unemployed; 21% part‐time employment; 71% full‐time employment
Carpenter 2016	Quasi‐experimental	25	USA	All school types	Children; nurses	Unclear	72% White; 12% Hispanic; 8% African American; 8% Black	Unclear
Cicutto 2013	Cluster RCT	1316	Canada	Primary/elementary	Children; school board; head teacher; teachers; peers	5 to 10	Unclear	25% to 50% deprived
Crane 2014	Quasi‐experimental	45	USA	Primary/elementary	Children	5 to 10	Unclear	Unclear
Dore‐Stites 2007	Quasi‐experimental	32	USA	Primary/elementary	Children; parents	5 to 10	39% African American; 28.6% Caucasian; 14.3% Hispanic; 18% biracial	34.6% < $20,000; 53.8% $21,000 to $40,000
Engelke 2013	Quasi‐experimental	143	USA	All school types	Children; teachers; parents; nurses	Unclear	40.6% Caucasian; 37.8% African American; 7% Latino; 14% other	63.6% on Medicaid
Gerald 2006	Cluster RCT	736	USA	Primary/elementary	Children; teachers	5 to 10	97% African American	Unclear
Henry 2004	Cluster RCT	4161	Australia	High	Children; teachers	11 to 14	Predominantly Caucasian	Unclear
Horner 2015	Cluster RCT	292	USA	Primary/elementary	Children	5 to 10	21.2% African American; 25% Spanish speaking	30.7% low SES
Howell 2005	Cluster RCT	24	USA	Primary/elementary	Children; parents	5 to 10	75% African American	Unclear
Jackson 2006	Quasi‐experimental	943	USA	Primary/elementary	Children	5 to 10	Unclear	Unclear
Joseph 2010	Parallel‐group RCT	314	USA	High	Children	11 to 18	Unclear	52% eligible for free school meals
Joseph 2013	Parallel‐group RCT	422	USA	High	Children	11 to 18	98% African American	73% on Medicaid
Kintner 2012	Quasi‐experimental	28	USA	High	Children; peers; families; teachers	11 to 14	53.6% African American; 32.1% White; 3.6% American; 10.7% biracial	35.7% low SES; 42.9% low middle SES; 17.8% upper middle SES; 3.6% high SES
Kouba 2012	Quasi‐experimental	25	USA	High	Children	11 to 18	92% African American; 4% Hispanic; 4% mixed	25% to 50% deprived
Langenfeld 2010	Quasi‐experimental	286	USA	Primary/elementary	Children; teachers	5 to 10	63% African American; 23.9% Hispanic; 6.4% White; 2.6% Asian	High percentage on free school meals
Lee 2011	Quasi‐experimental	827	USA	Primary/elementary	Children	5 to 10	Unclear	Unclear
Levy 2006	Cluster RCT	243	USA	Primary/elementary	Children; teachers	5 to 10	97% African American	80% on Medicaid
Magzamen 2008	Quasi‐experimental	845	USA	High; junior/middle	Children	11 to 18	Unclear	Unclear
McCann 2006	Parallel‐group RCT	219	UK	Primary/elementary	Children; teachers	5 to 10	Unclear	< 25% deprived
Mickel 2016	Quasi‐experimental	173	USA	Primary/elementary	Children	5 to 10	63.6% African American; 13.3% Hispanic; 20.2% White	> 50% deprived
Mujuru 2011	Quasi‐experimental	18	USA	Primary/elementary	Children; parents	5 to 10	Unclear	39% Medicaid
Pike 2011	Quasi‐experimental	236	USA	Primary/elementary	Children; teachers	5 to 10	75% African American (approx.)	80% free school meals (approx.)
Richmond 2011	Narrative	Unclear	USA	Primary/elementary	Children	5 to 10	100% African American	80% free school meals
Spencer 2000	Quasi‐experimental	369	USA	Primary/elementary	Children; parents	5 to 14	Unclear	34% free school meals
Splett 2006	Cluster RCT	1561	USA	All school types	Children; school staff	Unclear	66% African American; 6% Hispanic; 5% American Indian; 3% Asian; 20% White	73% free school meals
Terpstra 2012	Quasi‐experimental	58	USA	Junior/middle	Children; parents	11 to 14	> 50% BME	> 50% deprived
BME: black and minority ethnicity. RCT: randomised controlled trial.

Table 9. Process evaluation studies ‐ summary of study design, setting, and population

Table 10. Outcome evaluation studies not included in the analyses

Study included as outcome	Reason data not included in quantitative analysis
Bruzzese 2004	Feasibility study uses randomised controlled trial (RCT) design with no quantitative data presented
Bruzzese 2010	Abstract only located and outcomes were not presented in an extractable format
Clark 2004	Published effect sizes that were extractable but of a different effect size from other studies
Clark 2010	No outcome measured in the study matched the review protocol
McCann 2006	Outcomes were not presented in an extractable format (disaggregated data for asthmatic children unavailable)
Monforte 2012	Abstract only located and outcomes were not presented in an extractable format
Mosnaim 2011	No outcome measured in the study matched the review protocol
Praena‐Crespo 2010	Abstract only located and outcomes were not presented in an extractable format
Pulcini 2007	No outcome measured in the study matched the review protocol
Srof 2012	Outcomes were not presented in an extractable format (data on overall quality of life were not presented in full; only subdomains of quality of life are available)

Table 10. Outcome evaluation studies not included in the analyses

Table 11. Outcome evaluation studies ‐ summary of study design, setting, and population

	Study design	Number of children	Country	Type of school	Recipients	Age of children (years)	Representation of children from BME backgrounds	Representation of children from low SES backgrounds
Al‐Sheyab 2012	Clustered parallel RCT	261	Jordan	4 public high schools	Children	11 to 15	Unclear	Unclear
Atherly 2009	Clustered parallel RCT	524	USA	Junior and high schools	Children	11 to 15	Unclear	Unclear
Bartholomew 2006	Clustered parallel RCT	948	USA	Elementary schools	Children; care providers; parents/carers	5 to 10	45% African American; 51% Hispanic; 3% Caucasian	Deprived individuals > 50%
Bruzzese 2004	RCT	45	USA	2 public high schools	Children	Unclear	Unclear	Unclear
Bruzzese 2008	Clustered parallel RCT	24	USA	1 middle school	Children; caregivers	11 to 15	41% Hispanic; 17% African American	71% parents full‐time employment
Bruzzese 2010	Clustered parallel RCT	Unclear	USA	25 public schools	Children; caregivers	Mean age, 12.8	Unclear	Unclear
Bruzzese 2011	Clustered parallel RCT	340	USA	5 high schools	Children	Mean age, 15	> 80% BME	Unclear
Cicutto 2005	Clustered parallel RCT	256	Canada	26 elementary schools	Children	5 to 10	Unclear	Average household income $53,000
Cicutto 2013	Clustered RCT	1316	Canada	170 primary/elementary schools	Children; families	5 to 10	Unclear	Deprived individuals 25% to 50%
Clark 2004	Clustered parallel RCT	835	USA	14 public high schools	Children; parents/carers; classmates; school personnel	5 to 10	98% African American	45% annual income < $15,000
Clark 2005	Clustered parallel RCT	639	China	21 elementary schools	Children	7 to 11	Unclear	Unclear
Clark 2010	Clustered parallel RCT	1292	USA	19 middle schools	Children	Mean age, 11.6	93% African American	48% annual income < $15,000
Gerald 2006	Parallel‐group RCT	736	USA	54 elementary schools	Children	Mean age, 11	97% Black	Unclear
Gerald 2009	Parallel‐group RCT	290	USA	Unclear	Children	5 to 10	91% Black	Unclear
Henry 2004	Clustered parallel RCT	Unclear	Australia	Secondary schools	Children	11 to 15	< 50% BME	Unclear
Horner 2008	Clustered parallel RCT	183	USA	18 elementary schools	Children	5 to 10	47% Hispanic; 30% White; 22% African American	Unclear
Horner 2015	Clustered parallel RCT	196	USA	3 elementary schools	Children	5 to 10	> 50% BME	Deprived individuals 25% to 50%
Howell 2005	Clustered parallel RCT	25	USA	4 elementary schools	Children; families	5 to 10	75% African American	Unclear
Kintner 2009	Clustered parallel RCT	59	USA	5 schools	Children	9 to 12	30% Black; 36% White; 18% biracial	Deprived individuals 25% to 50%
Levy 2006	Clustered parallel RCT	243	USA	14 elementary schools	Children	5 to 10	98% African American	85% TennCare
McCann 2006	Clustered parallel RCT	229	England	24 primary/junior schools	Children; parents	5 to 10	Unclear	Deprived individuals < 25%
McGhan 2003	Clustered parallel RCT	162	Canada	18 elementary schools	Children	5 to 10	< 50% BME	Deprived individuals 25% to 50%
McGhan 2010	Clustered parallel RCT	206	Canada	Elementary schools	Children; parents/carers; teachers	Mean age, 8.6	Unclear	Unclear
Monforte 2012	Clustered parallel RCT	Unclear	USA	8 elementary schools	Children	5 to 10	Unclear	Unclear
Mosnaim 2011	Clustered parallel RCT	344 youth; 192 teens	USA	Elementary schools	Children	Median age 10	> 50% BME	Deprived individuals > 50%
Patterson 2005	Clustered parallel RCT	175	Ireland	Primary schools	Children	7 to 11	Unclear	Deprived individuals 25% to 50%
Persaud 1996	Parallel‐group RCT	36	USA	10 schools	Children	Mean age, 10.2	69% African American	69% received Medicaid
Praena‐Crespo 2010	Clustered parallel RCT	279	Spain	16 high schools	Children	11 to 15	Unclear	Unclear
Pulcini 2007	Clustered parallel RCT	40	USA	Middle schools	Children	11 to 15	Unclear	Unclear
Shah 2001	Clustered parallel RCT	272	Australia	High schools	Children	11 to 15	Unclear	Unclear
Splett 2006	Clustered parallel RCT	1561	USA	K‐8 schools	Children	5 to 15	66% African American	73% free school meals
Srof 2012	Parallel group RCT	39	USA	High schools	Children	14 to 18	Unclear	Unclear
Velsor‐Friedrich 2005	Clustered parallel RCT	52	USA	4 elementary schools	Children	Mean age, 10.1	100% African American	Unclear
BME: black and minority ethnicity. RCT: randomised controlled trial.

Table 11. Outcome evaluation studies ‐ summary of study design, setting, and population

Table 12. Outcome evaluation studies ‐ summary of intervention characteristics

	Named theoretical framework	Aim	Intervention type	Control	Intensity	Outcomes Included in meta‐analysis
Al‐Sheyab 2012	Self‐efficacy	To test the impact of the Triple A programme on health‐related outcomes in high school students	Triple A. Bilingual health workers trained peer leaders from year 11 to deliver 3 Triple A lessons	Unclear	3× lessons	HRQoL
Atherly 2009	None	To describe an analysis and results of the cost‐effectiveness of the Power Breathing programme	Power Breathing. This intervention focussed on education about asthma, asthma control strategies, and psychosocial concerns	Unclear	3× 90‐minute lessons	Hospitalisations; ED visits; Experience of daytime and night‐time symptoms
Bartholomew 2006	Social cognitive theory	To describe the evaluation of a school‐based intervention to improve asthma self‐management, medical care, the school environment, symptoms, and the functional status of children	Multi‐component intervention involving direct delivery to children, care providers, and parents/guardians. Children received education through the Watch, Discover, Think and Act interactive computer programme	Unclear	Unclear	Withdrawal
Bruzzese 2004	None	Unclear	ASMA. Continued medical education was also offered to medical providers	Usual care	3× lessons	None
Bruzzese 2008	Social cognitive theory; cognitive‐behavioural theory	To describe asthma: it’s a family affair; to present feasibility and preliminary outcome data from a pilot RCT	Elements of OAS and ASMA were provided to students; caregivers also received education	Usual care	6× lessons	Experience of daytime and night‐time symptoms; Withdrawal
Bruzzese 2010	None	To test the efficacy of an RCT: it’s a family affair, a school‐based, family‐focussed intervention to improve asthma outcomes in pre‐adolescents	ASMA and academic detailing. Students received workshops to empower them to manage their asthma. Parents received training to support their child’s need to manage their asthma	Unclear	Children: 6× lessons; caregivers: 5× lessons	Withdrawal
Bruzzese 2011	Social cognitive theory	Unclear	ASMA. Students received group sessions and individual tailored coaching sessions, delivered by trained health educators	Wait‐list control	3× group sessions; individual coaching sessions	Hospitalisations; ED visits; School absence; Restricted activity days; Unplanned GP or hospital visits; Experience of daytime and night‐time symptoms; Use of corticosteroids; Withdrawal
Cicutto 2005	Social cognitive theory; self‐regulation theory	To evaluate an asthma education programme for children with asthma	Roaring Adventures of Puff. Children received group sessions on asthma and goal‐setting	Usual care	6× lessons	Hospitalisations; ED visits; School absence; Restricted activity days
Cicutto 2013	Social cognitive theory	To implement an elementary school‐based asthma self‐management education programme for children with asthma; to work with schools to create an asthma‐friendly supportive school environment; to evaluate the programme	Roaring Adventures of Puff. Children received group sessions on asthma and goal‐setting	Usual care	6× lessons	ED visits; School absence; Restricted activity days; Unplanned GP or hospital visit; HRQoL; Withdrawal
Clark 2004	None	To assess the impact of a comprehensive school‐based asthma programme	OAS; control strategies for schools	Wait‐list control	6× lessons and 2× classroom sessions	School absence
Clark 2005	Social cognitive theory	To assess effectiveness in children in China of an asthma education programme adapted from a model developed in the USA	OAS; intervention directed at children only	Unclear	5× lessons	Hospitalisations; ED visits
Clark 2010	None	To assess self‐management and self‐management plus peer involvement	OAS; peer component. In the first treatment arm, an adapted form of OAS was delivered to children. In the second treatment arm, a peer education component was added	Usual care	6× lessons	Experience of daytime and night‐time symptoms
Gerald 2006	None	Unclear	OAS. The intervention included educational programmes and medical management for children, as well as education for school staff	Usual care	6× lessons	Hospitalisations; ED visits; School absence
Gerald 2009	None	To determine the effectiveness of school‐based supervised asthma therapy in improving asthma control	Children received asthma education, including a discussion of trigger avoidance (not manualised)	Usual care	1× lesson; multiple supervisions	School absence; Lung function; Use of reliever therapies; Withdrawal
Henry 2004	None	To determine whether an asthma education programme in schools would have a direct impact on student knowledge and attitudes on asthma and an indirect impact on teacher knowledge and attitudes	Asthma education. A package about asthma was taught within the PD/H/PE strand of the school curriculum	Usual care	3× lessons	HRQoL
Horner 2008	Asthma health education model	To examine changes in rural children’s asthma self‐management after they received classes, but before they received the family education session	Asthma self‐management. The curriculum included a 7‐step asthma self‐management plan	Health promotion education	16× lessons	Hospitalisations; Withdrawal
Horner 2015	Bruhn’s theoretical model of asthma self‐management	To test effects of 2 modes of delivering an asthma educational intervention on health outcomes and asthma self‐management in school‐aged children living in rural areas	7‐topic curriculum. The intervention was designed for children in rural areas and included asthma information	Health promotion education	16× lessons	Hospitalisations; ED visits; Withdrawal
Howell 2005	Social learning theory	To examine the feasibility of an interactive computer game in school‐based health centres; to test whether exposure to the game was effective in improving knowledge and reducing symptoms and healthcare use	Quest for the Code computer game. The caregiver also participated in medication interviews and received a home visit	Usual care	30‐minute session	ED visits; Experience of daytime and night‐time symptoms; HRQoL; School absence; Corticosteroid dosage
Kintner 2009	Lifespan development perspective	To evaluate the preliminary efficacy of SHARP	SHARP. Students worked through the SHARP curriculum. Caregivers also received a 3‐hour information sharing programme	Usual care	10× lessons	HRQoL; Withdrawal
Levy 2006	None	To evaluate the effectiveness of a school‐based nurse case management approach to asthma in students with poor control	OAS; monitoring of students; health status. Students received OAS education and weekly monitoring of their health status	Usual care	Weekly group sessions and weekly individual sessions	Hospitalisations; ED visits; Withdrawal
McCann 2006	None	To assess whether schools are an appropriate context for an intervention designed to produce clinical and psychological benefits for children with asthma	Education; role‐play. The intervention focussed on describing the respiratory condition through a role‐play	Education about the respiratory system	1× workshop	None
McGhan 2003	Social cognitive theory	To determine whether an interactive childhood asthma education programme improved asthma management behaviours, health status, and quality of life in elementary school children	Roaring Adventures of Puff. Children received education on asthma in a group setting. Parents and teachers were invited to participate in a school‐based asthma awareness event	Usual care	6× lessons	ED visits; School absence; Unplanned GP or hospital visit; Experience of daytime and night‐time symptoms; Withdrawal
McGhan 2010	Social cognitive theory; self‐regulation theory	To assess the feasibility and impact of the Roaring Adventures of Puff programme	Roaring Adventures of Puff delivered to children. Parents and teachers participated in an asthma awareness event.	Usual care	6× lessons	ED visits; School absence; Unplanned GP or hospital visit; Experience of daytime and night‐time symptoms; Withdrawal
Monforte 2012	None	To evaluate the implementation of OAS	OAS. No further information was given	Unclear	Unclear	HRQoL
Mosnaim 2011	None	To assess the impact of the Fight Asthma Now educational programme among 2 populations of predominantly low‐income minority students	One‐to‐one training on spacer technique, peak flow meter use, and use of an asthma action plan. Teens also received education on tobacco avoidance and peer pressure	Usual care	4× sessions	None
Patterson 2005	PRECEDE model	To evaluate the effectiveness of a programme of asthma clubs in improving quality of life for primary school children with asthma	SCAMP. Children used a workbook during sessions to learn about asthma	Wait‐list control	8× sessions	Restricted activity days; Lung function; HRQoL; Withdrawal
Persaud 1996	None	To assess the effectiveness of an intervention on knowledge, locus of control, attitudes towards asthma, functional status, school attendance, and ED visits	Individualised education sessions. Children had a personal peak flow meter in the school health office. The school nurse also reviewed the student asthma diary and discussed this with them	Usual care	3× lessons and weekly education sessions	ED visits; School absence
Praena‐Crespo 2010	None	To verify whether an asthma education program in schools would have direct benefit for student knowledge and attitudes towards asthma and quality of life for students with asthma	Asthma programme. No further information was given (abstract only)	Unclear	3× lessons	None
Pulcini 2007	None	To determine the effectiveness of an intervention to increase the number of AAPs in schools	Peak flow education. Children were given a peak flow meter and were educated on the correct technique to measure lung function	Unclear	Daily for 2 weeks	None
Shah 2001	None	To determine the effects of a peer‐led programme for asthma education on quality of life and related morbidity in adolescents with asthma	Triple‐A: asthma education and empowerment. Students learnt how to educate their peers about asthma. Peers also led 3 health lessons for classes in school	Wait‐list control	3× sessions	Experience of daytime and night‐time symptoms; Lung function; HRQoL; Withdrawal
Splett 2006	None	To improve asthma management among school children and reduce asthma‐related school absences, hospitalisations, and ED visits	Children received training on managing their asthma. Licensed nurses and healthcare assistants received coaching and reinforcement from asthma resource nurses	Usual care	Unclear	School absence; Unplanned GP or hospital visit
Srof 2012	Health promotion model	To determine effects of coping skills on asthma self‐efficacy, social support, quality of life, and peak flow among adolescents	Asthma diary; 5× coping skills sessions. Students received coping skills training and completed diary entries	Usual care	Sessions over 5 weeks	None
Velsor‐Friedrich 2005	Self‐care deficit theory	To test a 2‐part intervention on selected psychosocial and health outcomes for children with asthma	OAS; nurse practitioner visits. Children received the OAS education curriculum and nurse practitioner visits to assess asthma health and further education	Usual care	6× group sessions; individual nurse sessions	ED visits; Experience of daytime and night‐time symptoms; Lung function
AAP: XXX. ASMA: Asthma Self‐Management for Adolescents. ED: emergency department. GP: general practitioner. HRQoL: health‐related quality of life. ICAN: I Can Control Asthma and Nutrition Now. OAS: Open Airways for Schools. PD/H/PE: personal development/health/physical education. PRECEDE: Predisposing, Reinforcing, and Enabling Causes in Educational Diagnosis and Evaluation. RCT: randomised controlled trial. SCAMP: School Care and Asthma Management Project. SHARP: Staying Healthy–Asthma Responsible & Prepared. Triple A: Adolescent Asthma Action.

Table 12. Outcome evaluation studies ‐ summary of intervention characteristics

Table 13. Details of data transformations and adjustments made for meta‐analyses

Study	Indicator	Collection/reporting point	Mean cluster size (if applicable)	Intracluster correlation coefficient applied (if applicable)	Data transformation	Original effect size and standard error (with adjustment for clustering if applicable)	Final or transformed effect size and standard error (with adjustment for clustering if applicable)
Hospitalisations
Atherly 2009	Instances of hospitalisation in previous 4 weeks	Post intervention (3‐month follow‐up)	45.8	0.05	Yes – transformed from odds ratio to SMD	OR (0.7736); SE (lnOR) (1.385)	SMD (‐0.141); SE (0.764)
Bruzzese 2011	Hospitalisations in the past 2 months	Post intervention (12‐month follow‐up)	N/A	N/A	No	N/A	SMD (‐0.219); SE (0.120)
Clark 2005	Hospitalisations	Post intervention (12‐month follow‐up)	Deemed that analysis methods accounted for clustering	Deemed that analysis methods accounted for clustering	Yes – transformed from odds ratio to SMD	OR (1.43); SE (estimated from P value (lnOR)) 0.39	SMD (‐0.197); SE (0.215)
Gerald 2006	Median hospitalisations (not combined)	N/A	N/A	N/A	N/A	N/A	N/A
Horner 2008	Any hospital stays in the past 12 months (based on parents reporting any stay)	Post intervention (7‐month follow‐up)	10.1 (reported by study authors)	0.05	Yes – transformed from odds ratio to SMD	OR (0.882); SE (lnOR) (0.791)	SMD (‐0.069); SE (0.436)
Horner 2015	Mean number of hospitalisations since the previous data collection (at 8 months)	Post intervention (12‐month follow‐up)	8.9 (approx.)	0.05	No	N/A	SMD (‐0.057); SE (0.169)
Levy 2006	Mean hospital days	Post test (at intervention end)	17.36	0.05	No	N/A	SMD (‐0.293); SE (0.174)
Emergency department visits
Atherly 2009	Instances of ED visits in previous 4 weeks	Post intervention (3‐month follow‐up)	45.8	0.05	No	N/A	OR (1.036); SE (lnOR) (0.916)
Bruzzese 2011	Mean ED visits in the past 2 months	Post intervention (12‐month follow‐up)	N/A	N/A	Yes ‐ transformed from SMD to OR	SMD (‐0.289); SE (0.120)	OR (0.592); SE (lnOR) (0.218)
Cicutto 2005	ED visits in the past year	Post intervention (12‐month follow‐up)	9.85	0.05	No	N/A	OR (0.697); SE (lnOR) (0.407)
Cicutto 2013	ED visits in the past year (reports of)	Post intervention (12‐month follow‐up)	7.7	0.05	No	N/A	OR (0.318); SE (lnOR) (0.317)
Clark 2005	ED visits	Post intervention (12‐month follow‐up)	Deemed that analysis methods accounted for clustering	Deemed that analysis methods accounted for clustering	No, but see notes	N/A	OR (1.002); SE (estimated from P value (lnOR)) 0.072 Note that the OR was reported as 1.00 in the paper with a P value of 0.98. So information could be used and an SE extracted, a small correction to an OR of 1.002 was applied
Gerald 2006	Median ED visits (not combined)	N/A	N/A	N/A	N/A	N/A	N/A
Horner 2008	Any ED visits in the past 12 months (based on parents reporting any stay)	Post intervention (7‐month follow‐up)	10.1 (reported by study authors)	0.05	No	N/A	OR (0.857); SE (lnOR) (0.461)
Horner 2015	Mean number of ED visits since the previous data collection (8 months)	Post intervention (12‐month follow‐up)	8.9 (approx.)	0.05	Yes ‐ transformed from SMD to OR	SMD (0); SE (0.169)	OR (1.00); SE (0.306)
Howell 2005	Mean number of ED visits in the past 6 weeks	Post intervention (3‐month follow‐up)	4.25	0.05	Yes ‐ transformed from SMD to OR	SMD (‐0.331); SE (0.578)	OR (0.549); SE (1.049)
Levy 2006	Mean urgent care or emergency visits	Post test (at intervention end)	17.36	0.05	Yes ‐ transformed from SMD to OR	SMD (‐0.286); SE (0.174)	OR (0.595); SE (0.318)
McGhan 2003	ED visits in the past year (any)	Post intervention (9‐month follow‐up)	9	0.05	No	N/A	OR (1.283); SE (lnOR) (0.649)
McGhan 2010	ED visits in the past year (any)	Post intervention (12‐month follow‐up)	8.3	0.05	No	N/A	OR (2.64); SE (lnOR) (0.707)
Persaud 1996	Children with ED Visits (20‐week period post intervention)	Post intervention (events in 20‐week period post intervention)	N/A	N/A	No	N/A	OR (0.286); SE (lnOR) (0.737)
Velsor‐Friedrich 2005	Urgent doctor visits (any in the past 12 months)	Post intervention (12‐month follow‐up)	13	0.05	No	N/A	OR (0.683); SE (lnOR) (0.933)
Absence from school
Bruzzese 2011	Mean self‐reported absence in past 2 weeks (any absence)	Post intervention (12‐month follow‐up)	N/A	N/A	No	N/A	SMD (‐0.382); SE (0.121)
Cicutto 2005	Parent‐reported absence (any absence) over a year	Post intervention (12‐month follow‐up)	9.85	0.05	No	N/A	SMD (‐0.256); SE (0.151)
Cicutto 2013	Parent‐reported absence (any absence) over a year	Post intervention (12‐month follow‐up)	7.7	0.05	Yes – transformed from odds ratio to SMD	OR (0.660); SE (lnOR) (0.129)	SMD (‐0.229); SE (0.071)
Gerald 2006	Absences recorded on school records	Post test (unclear duration)	Clustering accounted for in analytical strategy	Clustering accounted for in analytical strategy	No	N/A	SMD (‐0.199); SE (0.084)
Gerald 2009	Absence from school due to respiratory illness/asthma *December measure used	Post intervention (15‐month follow‐up)	N/A	N/A	Yes – transformed from odds ratio to SMD	OR (1.1667); SE (lnOR) (0.364)	SMD (0.085); SE (0.227)
Howell 2005	School days missed in past 6 weeks	Post intervention (3‐month follow‐up)	3.25	0.05	No	N/A	SMD (0.152); SE (0.635)
McGhan 2003	Any missed school days	Post intervention (9‐month follow‐up)	9	0.05	Yes – transformed from odds ratio to SMD	OR (0.720); SE (lnOR) (0.413)	SMD (‐0.181); SE (0.227)
McGhan 2010	(No) Missed school days (any) over past 12 months	Post intervention (12‐month follow‐up)	8.3	0.05	Yes – transformed from odds ratio to SMD	OR (0.640); SE (lnOR) (0.353)	SMD (0.246); SE (0.195) (note: inverse taken as the intervention favours control)
Persaud 1996	Mean school days of absence based on school records	Post intervention (immediately afterwards)	N/A	N/A	No	N/A	SMD (‐0.236); SE (0.335)
Splett 2006	Mean percentage of days attended	Post intervention (12‐month follow‐up)	Deemed that analysis methods accounted for clustering	Deemed that analysis methods accounted for clustering	No	N/A	SMD (0.019); SE (0.051)
Days of restricted activity
Bruzzese 2011	Mean self‐reported days of restricted activity in past 2 weeks	Post intervention (12‐month follow‐up)	N/A	N/A	No	N/A	SMD (‐0.349); SE (0.120)
Cicutto 2005	Days of limited activity due to asthma	Post intervention (12‐month follow‐up)	9.85	0.05	No	N/A	SMD (‐0.318); SE (0.151)
Cicutto 2013	Percentage of students reporting days of restricted activity	Post intervention (12‐month follow‐up)	7.7	0.05	Yes – transformed from odds ratio to SMD	OR (0.612); SE (lnOR) (0.130)	SMD (‐0.271); SE (0.072)
Unplanned visits to medical providers
Bruzzese 2011	Mean acute care visits in the past 2 months	Post intervention (12‐month follow‐up)	N/A	N/A	Yes – transformed from SMD to OR	SMD (‐0.283); SE (0.120)	OR (0.598); SE (0.217)
Cicutto 2013	Unscheduled care in the past year (reports of)	Post intervention (12‐month follow‐up)	7.7	0.05	No	OR (0.703); SE (lnOR) (0.143)	SMD (‐0.194); SE (0.079)
McGhan 2003	Any unscheduled doctor visits	Post intervention (9‐month follow‐up)	9	0.05	No	OR (0.886); SE (lnOR) (0.426)	SMD (‐0.067); SE (0.235)
McGhan 2010	Unscheduled GP visits (any) over past 12 months	Post intervention (12‐month follow‐up)	8.3	0.05	No	OR (1.169); SE (lnOR) (0.397)	SMD (0.086); SE (0.219)
Splett 2006	Episodic asthma visits to school health office (over 6 months following start of intervention)	Over 6 months following start of intervention	97.6	0.05	No	OR (0.913); SE (lnOR) (0.282)	SMD (‐0.046); SE (0.156)
Daytime symptoms
Atherly 2009	Mean number of days with asthma symptoms	Post intervention (3‐month follow‐up)	45.8	0.05	No	N/A	SMD (‐0.026); SE (0.168)
Bruzzese 2008	Mean days last 2 weeks with asthma symptoms	Post intervention (2‐month follow‐up)	N/A	N/A	No	N/A	SMD (‐0.151); SE (0.418)
Bruzzese 2011	Mean days last 2 weeks with asthma symptoms	Post intervention (12‐month follow‐up)	N/A	N/A	No	N/A	SMD (‐0.210); SE (0.120)
Shah 2001	Number of students reporting attacks in school at follow‐up	Post intervention (6‐month follow‐up)	41.8	0.05	Yes – transformed from odds ratio to SMD	OR (0.647); SE (lnOR) (0.488)	SMD (‐0.240); SE (0.269)
Velsor‐Friedrich 2005	Symptom days in past 2 weeks	Post intervention (12‐month follow‐up)	13	0.05	Yes – transformed from odds ratio to SMD	OR (0.846); SE (lnOR) (0.705)	SMD (‐0.030); SE (0.413)
Night‐time symptoms
Bruzzese 2008	Mean nights woken last 2 weeks with asthma symptoms	Post intervention (2‐month follow‐up)	N/A	N/A	No	N/A	SMD (‐0.433); SE (0.423)
Bruzzese 2011	Mean self‐reported night‐time awakenings in past 2 weeks	Post intervention (12‐month follow‐up)	N/A	N/A	No	N/A	SMD (‐0.388); SE (0.121)
Howell 2005	Mean number of night‐time awakenings in past 6 weeks	Post intervention (3‐month follow‐up)	4.25	0.05	No	N/A	SMD (0.253); SE (0.478)
McGhan 2003	Waking up in past 2 weeks twice or more	Post intervention (9‐month follow‐up)	9	0.05	Yes – transformed from odds ratio to SMD	OR (1.237); SE (lnOR) (0.412)	SMD (0.117); SE (0.227)
Use of reliever therapies
Gerald 2009	Rescue medication use over twice per week *November measure used	Post intervention (15‐month follow‐up)	N/A	N/A	N/A	OR (0.228); SE (lnOR) (0.582)	N/A
McGhan 2003	Number of students with appropriate use of reliever medication	Post intervention (9‐month follow‐up)	9	0.05	N/A	OR (3.48); SE (lnOR) (0.565)	N/A
McGhan 2010	Used short‐acting bronchodilators in past 2 weeks	Post intervention (12‐month follow‐up)	8.3	0.05	N/A	OR (0.878); SE (lnOR) (0.356)	N/A
Splett 2006	Students with access to reliever medication visiting health office (over 6 months following start of intervention) *Note low levels of children with reliever medication	Over 6 months following start of intervention	97.6	0.05	N/A	OR (1.28); SE (lnOR) (0.282)	N/A
Use of corticosteroids and/or use of add‐on therapies
Bruzzese 2011	Use of controller medication	Post intervention (12‐month follow‐up)	N/A	N/A	No	N/A	OR (1.451); SE (lnOR) (0.240)
Horner 2015	Inhaled corticosteroid adherence	Post intervention (5‐month follow‐up)	8.9	0.05	No	N/A	SMD (‐0.605); SE (0.173)
Howell 2005	Inhaled corticosteroid adherence as prescribed (during past week)	Post intervention (3‐month follow‐up)	4.25	0.05	No	N/A	SMD (0.953); SE (0.546)
McGhan 2003	Currently using inhaled steroids	Post intervention (9‐month follow‐up)	9	0.05	No	N/A	OR (1.112); SE (lnOR) (0.418)
McGhan 2010	Currently using inhaled steroids	Post intervention (12‐month follow‐up)	8.3	0.05	No	N/A	OR (0.962); SE (lnOR) (0.376)
Splett 2006	Students with access to controller medication visiting health office (over 6 months following start of intervention) *Note low levels of children with controller medication	Over 6 months following start of intervention	97.6	0.05	N/A	OR (1.703); SE (lnOR) (0.806)	SMD (0.293); SE (0.445)
Lung function
Gerald 2009	Poor peak flow measures (red/amber readings)	Post‐intervention (15‐month follow‐up)	N/A	N/A	No	OR (0.94); SE (lnOR) (0.334)
Horner 2015	Airway inflammation (exhaled nitric oxide, collected using the single‐use RTube collection device, was the biomarker of airway inflammation)	Post intervention (12‐month follow‐up)	8.9	0.05	No	N/A	SMD (‐0.011); SE (0.169)
Shah 2001	Forced expiratory volume in 1 second: forced vital capacity before bronchodilator	Post intervention (3‐month follow‐up)	Deemed that analysis methods accounted for clustering	Deemed that analysis methods accounted for clustering	No	N/A	SMD (0.074); SE (0.127)
Patterson 2005	Forced expiratory volume in 1 second (% predicted change)	Post intervention (2‐month follow‐up)	Deemed that analysis methods accounted for clustering	Deemed that analysis methods accounted for clustering	No	N/A	SMD (‐0.05); SE (0.177)
Velsor‐Friedrich 2005	Peak flow increases as a percentage of pretest peak flow (change)	Post intervention (12‐month follow‐up)	13	0.05	No	N/A	SMD (‐5.905); SE (0.839)
Quality of life						Mean difference (QoL only)	Standardised mean difference (QoL only)
Al‐Sheyab 2012	Arabic version of the Pediatric Asthma Quality of Life Questionnaire (PAQLQ) *because of uncertainty about SD values, derived from t/P value of difference between means	Post intervention (3‐month follow‐up)	Deemed that analysis methods accounted for clustering	Deemed that analysis methods accounted for clustering	No	MD 1.35 (CI 0.96 to 1.74)	SMD (0.299); SE (0.129)
Cicutto 2005	Juniper Pediatric Asthma Quality of Life Questionnaire overall quality of life	Post intervention (2‐month follow‐up)	9.85	0.05	No	MD 0.50 (CI 0.00 to 1.00)	SMD (0.356); SE (0.151)
Cicutto 2013	Juniper Pediatric Asthma Quality of Life Questionnaire overall quality of life	Post intervention (12‐month follow‐up)	7.7	0.05	No	MD 0.40 (CI 0.21 to 0.59)	SMD (0.308); SE (0.064)
Henry 2004	Juniper Pediatric Asthma Quality of Life Questionnaire overall quality of life	Post intervention (6‐month follow‐up)	15.2	0.05	No	MD 0.16 (CI ‐0.22 to 0.54)	SMD (0.128); SE (0.114)
Horner 2008	Juniper Pediatric Asthma Quality of Life Questionnaire overall quality of life	Post intervention (7‐month follow‐up)	10.2	0.05	No	MD 0.05 (CI ‐0.21 to 0.31)	SMD (0.083); SE (0.196)
Howell 2005	Juniper Pediatric Asthma Quality of Life Questionnaire overall quality of life	Post intervention (3‐month follow‐up)	6	0.05	No	MD 0.03 (CI ‐1.71 to 1.77)	SMD (0.020); SE (0.484)
Kintner 2009	Quality of life is defined through the Participation in Life Activities Scale	Immediately post intervention	Deemed that analysis methods accounted for clustering	Deemed that analysis methods accounted for clustering	No	N/A	SMD (0.583); SE (0.263)
Patterson 2005	Change in Juniper Pediatric Asthma Quality of Life Questionnaire overall quality of life	Change in quality of life between baseline and 4 months post intervention	Deemed that analysis methods accounted for clustering	Deemed that analysis methods accounted for clustering	No	MD 0.07 (CI ‐0.26 to 0.40)	N/A
Shah 2001	Juniper Pediatric Asthma Quality of Life Questionnaire overall quality of life; percentage of students with clinically significant improvement	Post intervention (3‐month follow‐up)	Deemed that analysis methods accounted for clustering	Deemed that analysis methods accounted for clustering	No	MD 0.09 (CI ‐0.23 to 0.41)	N/A
Withdrawal
Al‐Sheyab 2012	Withdrew between baseline and outcome collection	Post intervention (3‐month follow‐up)	65.25	0.05	No	N/A	OR (0.511); SE (lnOR) (1.074)
Bartholomew 2006	Lost to follow‐up at post‐test measure	Post intervention (duration unclear)	11.2	0.05	No	N/A	OR (0.237); SE (lnOR) (0.145)
Bruzzese 2008	Withdrew between baseline and outcome collection	Immediate post intervention	N/A	N/A	No	N/A	OR (0.307); SE (lnOR) (1.683)
Bruzzese 2011	Withdrew between baseline and outcome collection	Post intervention (12‐month follow‐up)	N/A	N/A	No	N/A	OR (1.313); SE (lnOR) (0.279)
Cicutto 2005	Withdrew between baseline and outcome collection	Post intervention (6‐month follow‐up)	9.85	0.05	No	N/A	OR (1.788); SE (lnOR) (0.629)
Gerald 2009	Withdrew between baseline and outcome collection	Post intervention (6‐month follow‐up)	N/A	N/A	No	N/A	OR (1.788); SE (lnOR) (0.613)
Horner 2008	Withdrew between baseline and outcome collection	Post intervention (7‐month follow‐up)	10.2	0.05	No	N/A	OR (1.333); SE (lnOR) (0.531)
Horner 2015	Failed to complete final data collection	Post intervention (12‐month follow‐up)	8.9	0.05	No	N/A	OR (0.75); SE (lnOR) (0.486)
Kintner 2009	Withdrew during intervention and between end of intervention and follow‐up	Post intervention (12‐month follow‐up)	13.2	0.05	No	N/A	OR (30.176); SE (lnOR) (1.860)
Levy 2006	Failure to complete outcome evaluation	Post intervention (12‐month follow‐up)	17.36	0.05	No	N/A	OR (0.357); SE (lnOR) (0.3881)
McGhan 2003	Withdrew between baseline and outcome collection	Post intervention (9‐month follow‐up)	9	0.05	No	N/A	OR (1.147); SE (lnOR) (0.5381)
McGhan 2010	Withdrew between baseline and interim outcome collection	Post intervention (6‐month follow‐up)	8.3	0.05	No	N/A	OR (1.007); SE (lnOR) (0.387)
Patterson 2005	Withdrew during intervention	Post intervention – immediately following intervention	7.95	0.05	No	N/A	OR (5.675); SE (lnOR) (1.087)
Shah 2001	Withdrew between baseline and outcome collection	Post intervention (3‐month follow‐up)	45.3	0.05	No	N/A	OR (1.343); SE (lnOR) (0.475)
CI: confidence interval. ED: emergency department. lnOR: log odds ratio. MD: mean difference. N/A: not applicable. OR: odds ratio. PAQLQ: Pediatric Asthma Quality of Life Questionnaire. QoL: quality of life. SD: standard deviation. SE: standard error. SMD: standardised mean difference.

Table 13. Details of data transformations and adjustments made for meta‐analyses

Table 14. Summary of interventions, conditions entered, and model results

Domain (model)	Conditions entered	Sufficient configurations identified that trigger successful implementation
1. Setting and participant features	School health centre; high school; parents direct intervention recipients; teachers direct intervention recipients; school nurses/others direct intervention recipients	Yes
2. Recruitment and retention processes	Additional marketing materials; provision of incentives; provision of catch‐up sessions; provision of reminders	No
3. Curriculum, pedagogy, and intervention emphasis	Focus on establishing alliances with care providers; focus on asthma symptom recognition and management; tailored content; emphasis on personal responsibility; interactive pedagogical style; diverse pedagogical style	No
4. Modifiable intervention processes	Theory driven; run in class time; run in students' free time; school nurse key role in delivery or teaching; personalised or individual 1‐to‐1 instruction	Yes
5. Stakeholder engagement	School asthma policy; child satisfaction; teachers engaged/relationships developed; parents engaged/relationships developed; school nurses engaged/relationships developed	Yes
6. Consolidated model	Theory driven; run in students' free time; child satisfaction; parents engaged/relationships developed; high school	Yes

Table 14. Summary of interventions, conditions entered, and model results

Table 15. Data table for QCA model 6 ‐ consolidated model

	Successful intervention	High school	Child satisfaction	Theory driven	Intervention takes place during students' own free time	Good relationships/engagement with parents
Joseph 2010	0.52	1	0	1	0.33	0
Kouba 2012	0.33	1	0	1	1	0
Dore‐Stites 2007	0.67	0	1	1	0.33	0.75
Joseph 2013	1.00	1	0	1	0.75	1
Mujuru 2011	0.67	0	0	0	0	0.25
Henry 2004	0.83	1	0	0	0	0
Pike 2011	0.67	0	0	0	0	0
Spencer 2000	0.33	0	0	0	0.33	1
Engelke 2013	0.50	0.5	0	0	0.33	1
Splett 2006	0.50	0.5	0	0	0.33	0
Kintner 2012	0.83	1	1	1	1	0.25
Berg 2004	0.83	1	1	1	0.33	0
Howell 2005	0.33	0	0.633333	1	0.33	0.75
Gerald 2006	0.33	0	0	0	0.33	0
Langenfeld 2010	0.33	0	0	0	0.33	0
Al‐Sheyab 2012	0.83	1	0.633333	1	0.33	0
Levy 2006	0.52	0	0	0	0.33	0
Terpstra 2012	1.00	0.66	0	1	1	0.25
Horner 2015	0.67	0	0	1	1	0
Bruzzese 2008	0.94	0.66	1	1	0.33	1
Lee 2011	0.50	0	0	1	0	0
Bruzzese 2004	0.33	1	0.633333	1	0.75	0
Cicutto 2013	0.67	0	0	1	1	0
Brasler 2006	0.00	0.66	0.633333	0	0.75	0
Crane 2014	0.50	0	0	1	1	0
Bruzzese 2011	0.88	1	0	1	0.33	0
Magzamen 2008	0.19	0.75	0	0	1	0
QCA: qualitative comparative analysis.

Table 15. Data table for QCA model 6 ‐ consolidated model

Table 16. Truth table for QCA model 6 ‐ consolidated model

High school	Child satisfaction	Theory driven	Intervention takes place during students' own free time	Good relationships/ engagement with parents	Outcome code (based on consistency score)	Number of studies with membership in causal combination > 0.5	Consistency score with subset relationship (n = 27 in each assessment)	Proportional reduction in inconsistency	Cases
1	1	1	0	0	1	2	1	1	Al‐Sheyab 2012; Berg 2004
1	0	1	1	1	1	1	1	1	Joseph 2013
1	1	1	0	1	1	1	1	1	Bruzzese 2008
1	0	1	0	0	1	2	0.924	0.841	Bruzzese 2011; Joseph 2010
1	1	1	1	0	1	2	0.853	0.752	Bruzzese 2004; Kintner 2012
0	1	1	0	1	1	2	0.815	0.668	Dore‐Stites 2007; Howell 2005
1	0	1	1	0	0	2	0.768	0.595	Kouba 2012; Terpstra 2012
0	0	0	0	1	0	1	0.763	0	Engelke 2013; Spencer 2000
1	0	0	0	0	0	1	0.762	0.615	Henry 2004
0	0	1	1	0	0	3	0.675	0.463	Cicutto 2013; Crane 2014; Horner 2015
0	0	0	0	0	0	5	0.67	0.322	Gerald 2006; Langenfeld 2010; Levy 2006; Mujuru 2011; Pike 2011; Splett 2006
0	0	1	0	0	0	1	0.6	0	Lee 2011
1	0	0	1	0	0	1	0.358	0	Magzamen 2008
1	1	0	1	0	0	1	0	0	Brasler 2006
QCA: qualitative comparative analysis.

Table 16. Truth table for QCA model 6 ‐ consolidated model

Table 17. Complex solution for QCA model 6 ‐ consolidated model

		Consistency score with subset relationship (n = 27 in each assessment)	Proportional reduction in inconsistency	Raw coverage	Unique coverage	Cases
1	CHILDSATTHEORYDRIVENruninstudenttime*GOODRELPAR	0.846	0.756	0.106	0.106	Bruzzese 2008; Dore‐Stites 2007; Howell 2005
2	HIGHSCHOOLCHILDSATTHEORYDRIVEN*goodrelpar	0.845	0.786	0.162	0.063	Al‐Sheyab 2012; Berg 2004; Bruzzese 2004; Kintner 2012
3	HIGHSCHOOLTHEORYDRIVENruninstudenttime*goodrelpar	0.949	0.914	0.177	0.078	Al‐Sheyab 2012; Berg 2004; Bruzzese 2011; Joseph 2010
4	HIGHSCHOOLchildsatTHEORYDRIVENRUNINSTUDENTTIMEGOODRELPAR	1	1	0.064	0.064	Joseph 2013
	M1	0.875	0.823	0.41
QCA: qualitative comparative analysis. [Notation: Upper case = condition is present; Lower case = condition is absent; * = logical and; + logical or; Key: HIGHSCHOOL = High School (lower case not in high school); THEORYDRIVEN = Authors explicitly named theory or presented conceptual model for intervention; RUNINSTUDENTTIME = Substantial component run in students' own time (e.g. lunchtime); GOODRELPAR = Good level of reported in engagement and/or developing relationships with parents; CHILDSAT = Children reported as satisfied; SUCCESSFULIMPLEMENTATION = Implementation of intervention successful]

Table 17. Complex solution for QCA model 6 ‐ consolidated model

Table 18. Intermediate solution for QCA model 6 ‐ consolidated model

		Consistency score with subset relationship (n = 27 in each assessment)	Proportional reduction in inconsistency	Raw coverage	Unique coverage	Cases
1	HIGHSCHOOLCHILDSATTHEORYDRIVEN	0.839	0.791	0.21	0.053	Al‐Sheyab 2012; Berg 2004; Bruzzese 2004; Bruzzese 2008; Kintner 2012
2	HIGHSCHOOLTHEORYDRIVENGOODRELPAR	1	1	0.138	0.064	Bruzzese 2008; Joseph 2013
3	HIGHSCHOOLTHEORYDRIVENruninstudenttime	0.961	0.942	0.235	0.078	Al‐Sheyab 2012; Berg 2004; Bruzzese 2008; Bruzzese 2011; Joseph 2013
4	CHILDSATTHEORYDRIVENruninstudenttime*GOODRELGPAR	0.846	0.756	0.106	0.064	Bruzzese 2008; Dore‐Stites 2007; Howell 2005
	M1	0.862	0.81	0.432
QCA: qualitative comparative analysis. [Notation: Upper case = condition is present; Lower case = condition is absent; * = logical and; + logical or; Key: HIGHSCHOOL = High School (lower case not in high school); THEORYDRIVEN = Authors explicitly named theory or presented conceptual model for intervention; RUNINSTUDENTTIME = Substantial component run in students' own time (e.g. lunchtime); GOODRELPAR = Good level of reported in engagement and/or developing relationships with parents; CHILDSATB = Children reported as satisfied; SUCCESSFULIMPLEMENTATION = Implementation of intervention successful]

Table 18. Intermediate solution for QCA model 6 ‐ consolidated model

Table 19. Summary of results from consolidated model

Consolidated model	Theory driven	Run in children's free time	Child satisfaction	Parents engaged/relationships developed	High school	Successful intervention
Pathway 1	Present	‐	Present	‐	Present	Yes
Pathway 2	Present	‐	‐	Present	Present	Yes
Pathway 3	Present	Absent	‐	‐	Present	Yes
Pathway 4	Present	Absent	Present	Present	‐	Yes
Absent: absence of condition is essential in triggering success. Present: presence of condition is essential in triggering success. ‐ (symbol): presence or absence of condition is not essential in triggering success.

Table 19. Summary of results from consolidated model

Table 20. Summary of QCA results based on intermediate solutions

Model 1. Setting and participant features	School health centre	High school	Parents direct intervention recipients	Teachers direct intervention recipients	School nurses/others direct intervention recipients		Successful intervention
Pathway 1	Present	Present	Present	Absent	‐		Yes
Pathway 2	Absent	Present	Absent	‐	‐		Yes
Pathway 3	Absent	‐	Absent	Absent	Absent		Yes
Pathway 4	Present	Present	Present	‐	Present		Yes
Model 2. Recruitment and retention processes	Additional marketing materials	Provision of incentives	Provision of catch‐up sessions	Provision of reminders			Successful intervention
	‐	‐	‐	‐	‐		No solution found
Model 3. Curriculum, pedagogy, and intervention emphasis	Focus on establishing alliances with care providers	Focus on asthma symptom recognition and management	Tailored content	Emphasis on personal responsibility	Interactive pedagogical style	Diverse pedagogical style	Successful intervention
	‐	‐	‐	‐	‐		No solution found
Model 4. Modifiable intervention processes	Theory driven	Run in class time	Run in students’ free time	School nurse key role in delivery or teaching	Personalised or individual 1‐to‐1 instruction		Successful intervention
Pathway 1	Present	‐	Absent	Absent	‐		Yes
Pathway 2	Present	‐	‐	Present	Absent		Yes
Model 5. Stakeholder engagement	School asthma policy	Child satisfaction	Teachers engaged/ relationships developed	Parents engaged/ relationships developed	School nurses engaged/ relationships developed		Successful intervention
Pathway 1	Absent	‐	‐	Present	Absent		Yes
Pathway 2	‐	Present	‐	‐	Absent		Yes
Model 6. Consolidated model	Theory driven	Run in students’ free time	Child satisfaction	Parents engaged/ relationships developed	High school		Successful intervention
Pathway 1	Present	‐	Present	‐	Present		Yes
Pathway 2	Present	‐	‐	Present	Present		Yes
Pathway 3	Present	Absent	‐	‐	Present		Yes
Pathway 4	Present	Absent	Present	Present	‐		Yes
Absent: absence of condition is essential in triggering success. Present: presence of condition is essential in triggering success. QCA: qualitative comparative analysis. ‐ (symbol): presence or absence of condition is not essential in triggering success.

Table 20. Summary of QCA results based on intermediate solutions

Table 21. Truth table for QCA model 1 ‐ setting and participants

School‐based health centre	High school	Parents directly involved	Teachers received training	School nurses or other stakeholders received training	Outcome code (based on consistency score)	Number of studies with membership in causal combination > 0.5	Consistency score with subset relationship (n = 27 in each assessment)	Proportional reduction in inconsistency	Cases
1	1	0	1	0	1	2	1	1	Bruzzese 2008; Terpstra 2012
1	0	1	0	0	1	1	1	1	Henry 2004
1	1	1	1	1	1	1	1	1	Kintner 2012
0	0	1	0	1	1	1	0.995	0.99	Cicutto 2013
0	0	0	0	0	1	2	0.918	0.588	Crane 2014; Pike 2011
1	0	0	0	0	1	1	0.889	0.811	Al‐Sheyab 2012
1	1	0	0	1	0	2	0.865	0.662	Bruzzese 2004; Bruzzese 2011
1	1	0	0	0	0	4	0.852	0.761	Berg 2004; Joseph 2010; Joseph 2013; Magzamen 2008
0	1	0	0	0	0	4	0.845	0.543	Horner 2015; Langenfeld 2010; Lee 2011; Mujuru 2011
0	0	1	0	0	0	1	0.763	0.136	Levy 2006
0	1	1	0	0	0	1	0.754	0	Gerald 2006
1	0	0	1	0	0	1	0.751	0.647	Kouba 2012
0	1	0	1	0	0	3	0.73	0.56	Dore‐Stites 2007; Howell 2005; Spencer 2000
1	1	1	1	0	0	1	0	0	Brasler 2006
QCA: qualitative comparative analysis.

Table 21. Truth table for QCA model 1 ‐ setting and participants

Table 22. Complex solution for QCA model 1 ‐ setting and participants

		Consistency score with subset relationship (n = 27 in each assessment)	Proportional reduction in inconsistency	Raw coverage	Unique coverage	Cases
1	HIGHSCHOOLschoolbasedhealthparentdirect*anyothdir	0.913	0.861	0.176	0.043	Al‐Sheyab 2012; Henry 2004
2	schoolbasedhealthteacherdirectparentdirect*anyothdir	0.913	0.769	0.294	0.16	Al‐Sheyab 2012; Crane 2014; Pike 2011
3	highschoolschoolbasedhealthTEACHERDIRECTparentdirectANYOTHDIR	0.995	0.99	0.042	0.042	Cicutto 2013
4	HIGHSCHOOLSCHOOLBASEDHEALTHteacherdirectPARENTDIRECTanyothdir	1	1	0.105	0.105	Bruzzese 2008; Terpstra 2012
5	HIGHSCHOOLSCHOOLBASEDHEALTHTEACHERDIRECTPARENTDIRECTANYOTHDIR	1	1	0.074	0.074	Kintner 2012
	M1	0.952	0.901	0.558
QCA: qualitative comparative analysis. [Notation: Upper case = condition is present; Lower case = condition is absent; * = logical and; + logical or; Key: HIGHSCHOOL = High School (lower case not in high school); SCHOOLBASEDHEALTH = School Based Health Centre; TEACHERDIRECT = Teachers received directly received component of intervention; PARENTDIRECT = Parents directly received component of intervention; ANYOTHDIR = School nurses or other stakeholders (apart from children) directly received component of intervention; SUCCESSFULIMPLEMENTATION = Implementation of intervention successful]

Table 22. Complex solution for QCA model 1 ‐ setting and participants

Table 23. Intermediate solution for QCA model 1 ‐ setting and participants

		Consistency score with subset relationship (n = 27 in each assessment)	Proportional reduction in inconsistency	Raw coverage	Unique coverage	Cases
1	HIGHSCHOOLschoolbasedhealthparentdirect	0.904	0.838	0.226	0.093	Al‐Sheyab 2012; Henry 2004
2	HIGHSCHOOLSCHOOLBASEDHEALTHteacherdirect*PARENTDIRECT	1	1	0.105	0.105	Bruzzese 2008; Terpstra 2012
3	schoolbasedhealthteacherdirectparentdirect*anyothdir	0.913	0.769	0.294	0.16	Crane 2014; Pike 2011

4	highschoolTEACHERDIRECTANYOTHDIR	0.778	0.5	0.074	0.042	Cicutto 2013
5	HIGHSCHOOLSCHOOLBASEDHEALTHPARENTDIRECT*ANYOTHDIR	1	1	0.074	0.042	Kintner 2012
	Solution	0.915	0.831	0.608
QCA: qualitative comparative analysis. Overall solution HIGHSCHOOLschoolbasedhealthparentdirect + schoolbasedhealthteacherdirectparentdirectanyothdir + HIGHSCHOOLSCHOOLBASEDHEALTHteacherdirectPARENTDIRECT + (highschoolTEACHERDIRECTANYOTHDIR + HIGHSCHOOLSCHOOLBASEDHEALTHPARENTDIRECTANYOTHDIR) => SUCCESSFULIMPLEMENTATION [Notation: Upper case = condition is present; Lower case = condition is absent; = logical and; + logical or; Key: HIGHSCHOOL = High School (lower case not in high school); SCHOOLBASEDHEALTH = School Based Health Centre; TEACHERDIRECT = Teachers received directly received component of intervention; PARENTDIRECT = Parents directly received component of intervention; ANYOTHDIR = School nurses or other stakeholders (apart from children) directly received component of intervention; SUCCESSFULIMPLEMENTATION = Implementation of intervention successful]

Table 23. Intermediate solution for QCA model 1 ‐ setting and participants

Table 24. Data table for QCA model 3 ‐ curriculum, pedagogy, and intervention emphasis

	Successful intervention	Curriculum reflected forming alliances and monitoring symptoms	Curriculum reflected learning about asthma triggers and monitoring symptoms	Emphasised the intervention as being tailored or personalised	Emphasised developing personal responsibility as aim of the intervention	Pedagogical style focused on interactive methods	Diverse pedagogical style employed
Joseph 2010	0.52	0	1	1	0	0	0
Kouba 2012	0.33	0	0	0	1	0	0
Dore‐Stites 2007	0.67	1	0	0	1	0	0
Joseph 2013	1.00	0	1	1	0	0	1
Mujuru 2011	0.67	0	1	0	0	0	0
Henry 2004	0.83	0	0	0	0	0	0
Pike 2011	0.67	0	1	0	0	0	0
Spencer 2000	0.33	0	0	0	0	1	0
Engelke 2013	0.50	0	0	0	0	0	1
Splett 2006	0.50	0	0	0	0	1	0
Kintner 2012	0.83	0	1	0	0	0	0
Berg 2004	0.83	0	1	1	0	0	0
Howell 2005	0.33	0	1	0	0	0	0
Gerald 2006	0.33	1	0	0	0	0	0
Cheung 2015	0.33	0	0	0	1	0	0
Al‐Sheyab 2012	0.83	0	1	0	1	0	1
Levy 2006	0.52	0	0	0	0	0	1
Terpstra 2012	1.00	1	0	0	1	0	0
Horner 2015	0.67	1	0	0	1	0	0
Bruzzese 2008	0.94	0	1	0	1	0	0
Lee 2011	0.50	0	0	0	0	0	1
Bruzzese 2004	0.33	0	0	1	0	0	0
Cicutto 2013	0.67	1	0	0	0	0	0
Brasler 2006	0.00	0	1	0	0	0	0
Crane 2014	0.50	0	0	0	1	0	0
Bruzzese 2011	0.88	0	0	1	0	0	0
Magzamen 2008	0.19	0	1	0	0	0	0
QCA: qualitative comparative analysis.

Table 24. Data table for QCA model 3 ‐ curriculum, pedagogy, and intervention emphasis

Table 25. Truth table for QCA model 3 ‐ curriculum, pedagogy, and intervention emphasis

Curriculum reflected forming alliances and monitoring symptoms	Curriculum reflected learning about asthma triggers and monitoring symptoms	Emphasised the intervention as being tailored or personalised	Emphasised developing personal responsibility as aim of the intervention	Pedagogical style focused on interactive methods	Diverse pedagogical style employed	Outcome code (based on consistency score)	Number of studies with membership in causal combination > 0.5	Consistency score with subset relationship (n=27 in each assessment); [proportional reduction in inconsistency]	Cases
0	1	1	0	0	1	1	1	1 [1]	Joseph 2013
0	1	0	1	0	0	1	1	0.938 [0.9333]	Bruzzese 2008
0	0	0	0	0	0	0	1	0.833 [0.8]	Henry 2004
0	1	0	1	0	1	0	1	0.833 [0.8]	Al‐Sheyab 2012
1	0	0	1	0	0	0	3	0.778 [0.714]	Dore‐Stites 2007; Horner 2015; Terpstra 2012
0	1	1	0	0	0	0	2	0.677 [0.523]	Berg 2004; Joseph 2010
0	0	1	0	0	0	0	2	0.604 [0.486]	Bruzzese 2004; Bruzzese 2011
0	0	0	0	0	1	0	3	0.507 [0.027]	Engelke 2013; Lee 2011; Levy 2006
1	0	0	0	0	0	0	2	0.5 [0.25]	Cicutto 2013; Gerald 2006
0	1	0	0	0	0	0	6	0.448 [0.287]	Brasler 2006; Howell 2005; Kintner 2012; Magzamen 2008; Mujuru 2011; Pike 2011
0	0	0	0	1	0	0	2	0.417 [0]	Spencer 2000; Splett 2006
0	0	0	1	0	0	0	3	0.389 [0]	Crane 2014; Kouba 2012; Langenfeld 2010
QCA: qualitative comparative analysis.

Table 25. Truth table for QCA model 3 ‐ curriculum, pedagogy, and intervention emphasis

Table 26. Truth table for QCA model 4 ‐ modifiable design features

Theory driven	Personalised or individual sessions	Intervention takes place during lesson time	Intervention takes place during students’ own free time	School nurse involved in delivery of the intervention	Outcome code (based on consistency score)	Number of studies with membership in causal combination > 0.5	Consistency score with subset relationship (n = 27 in each assessment)	Proportional reduction in inconsistency	Cases
1	0	0	0	1	1	2	1	1	Bruzzese 2008; Dore‐Stites 2007
1	0	0	0	0	1	1	1	1	Al‐Sheyab 2012
1	0	1	1	1	1	1	1	1	Kintner 2012
1	1	0	0	0	1	1	0.996	0.993	Bruzzese 2011
1	0	0	1	1	1	1	0.931	0.816	Joseph 2010
1	1	1	0	0	1	2	0.931	0.872	Crane 2014; Terpstra 2012
1	0	1	0	1	1	1	0.903	0.729	Lee 2011
1	1	1	1	0	0	2	0.852	0.729	Bruzzese 2004; Joseph 2013
1	0	0	1	0	0	2	0.833	0.706	Cicutto 2013; Horner 2015
1	1	0	0	1	0	2	0.753	0.602	Berg 2004; Howell 2005
1	1	0	1	0	0	1	0.732	0.481	Kouba 2012
0	0	1	0	0	0	5	0.659	0.035	Engelke 2013; Langenfeld 2010; Levy 2006; Spencer 2000; Splett 2006
0	1	0	0	1	0	0	4	0.638	0.484
0	0	1	1	1	0	0	1	0.5	0
0	0	0	1	1	0	0	1	0.444	0
QCA: qualitative comparative analysis.

Table 26. Truth table for QCA model 4 ‐ modifiable design features

Table 27. Complex solution for QCA model 4 ‐ modifiable design features

		Consistency score with subset relationship (n = 27 in each assessment)	Proportional reduction in inconsistency	Raw coverage	Unique coverage	Cases
1	THEORYDRIVENpersonalorindividualSCHOOLNURSEINSTRUCT	0.926	0.876	0.253	0.148	Bruzzese 2008; Crane 2014; Dore‐Stites 2007; Kintner 2012; Lee 2011; Terpstra 2012
2	THEORYDRIVENPERSONALORINDIVIDUALruninstudenttime*schoolnurseinstruct	0.938	0.866	0.151	0.033	Bruzzese 2011; Joseph 2013
3	THEORYDRIVENpersonalorindividualruninlessons*runinstudenttime	0.999	0.998	0.149	0.001	Al‐Sheyab 2012a; Bruzzese 2008; Dore‐Stites 2007
	M1	0.933	0.883	0.426
QCA: qualitative comparative analysis. [Notation: Upper case = condition is present; Lower case = condition is absent; * = logical and; + logical or; Key: THEORYDRIVEN = Authors explicitly named theory or presented conceptual model for intervention; SCHOOLNURSEINSTRUCT = Substantial component delivered by schools' nurse; PERSONALORINDIVIDUAL = Substantial components delivered that were individually personalised or delivered to individuals; RUNINSTUDENTTIME = Substantial component run in students' own time (e.g. lunchtime); RUNINLESSONS = Substantial component run during lesson time; SUCCESSFULIMPLEMENTATION = Implementation of intervention successful]

Table 27. Complex solution for QCA model 4 ‐ modifiable design features

Table 28. Intermediate solution for QCA model 4 ‐ further modifiable intervention design features

		Consistency score with subset relationship (n = 27 in each assessment)	Proportional reduction in inconsistency	Raw coverage	Unique coverage	Cases
1	THEORYDRIVENpersonalorindividualSCHOOLNURSEINSTRUCT	0.926	0.876	0.253	0.167	Bruzzese 2008; Crane 2014; Dore‐Stites 2007; Kintner 2012; Lee 2011; Terpstra 2012
2	THEORYDRIVENruninstudenttimeschoolnurseinstruct	0.963	0.92	0.258	0.172	Al‐Sheyab 2012; Bruzzese 2011; Joseph 2010
	M1	0.933	0.883	0.425
QCA: qualitative comparative analysis. [Notation: Upper case = condition is present; Lower case = condition is absent; * = logical and; + logical or; Key: THEORYDRIVEN = Authors explicitly named theory or presented conceptual model for intervention; SCHOOLNURSEINSTRUCT = Substantial component delivered by schools' nurse; PERSONALORINDIVIDUAL = Substantial components delivered that were individually personalised or delivered to individuals; RUNINSTUDENTTIME = Substantial component run in students' own time (e.g. lunchtime); RUNINLESSONS = Substantial component run during lesson time; SUCCESSFULIMPLEMENTATION = Implementation of intervention successful] Overall solution THEORYDRIVENruninstudenttimeschoolnurseinstruct + THEORYDRIVENpersonalorindividualSCHOOLNURSEINSTRUCT => PROCOUTSUM

Table 28. Intermediate solution for QCA model 4 ‐ further modifiable intervention design features

Table 29. Truth table for QCA model 5 ‐ stakeholder involvement and engagement

	School asthma policy	Good relationships/ engagement with parents	Good relationships/ engagement with school nurses	Child satisfaction	Outcome code (based on consistency score)	Number of studies with membership in causal combination > 0.5	Consistency score with subset relationship (n = 27 in each assessment)	Proportional reduction in inconsistency	Cases
1	0	1	0	0	1	1	1	1	Joseph 2013
2	0	1	0	1	1	1	0.958	0.939	Bruzzese 2008
3	0	0	0	1	1	4	0.857	0.786	Al‐Sheyab 2012; Berg 2004; Bruzzese 2004; Kintner 2012
4	0	1	1	1	0	2	0.723	0.465	Dore‐Stites 2007; Howell 2005
5	1	0	0	0	0	3	0.674	0.515	Cicutto 2013; Henry 2004; Levy 2006
6	0	0	0	0	0	10	0.615	0.405	Bruzzese 2011; Gerald 2006; Horner 2015; Joseph 2010; Kouba 2012; Lee 2011; Magzamen 2008; Mujuru 2011; Pike 2011; Terpstra 2012
7	0	0	1	0	0	1	0.6	0	Crane 2014
8	1	1	0	0	0	1	0.5	0	Engelke 2013
9	0	1	1	0	0	1	0.488	0	Spencer 2000
10	1	0	1	0	0	2	0.352	0	Langenfeld 2010; Splett 2006
11	1	0	1	1	0	1	0	0	Brasler 2006
QCA: qualitative comparative analysis.

Table 29. Truth table for QCA model 5 ‐ stakeholder involvement and engagement

Table 30. Complex solution for QCA model 5 ‐ stakeholder involvement and engagement

		Consistency score with subset relationship (n = 27 in each assessment)	Proportional reduction in inconsistency	Raw coverage	Unique coverage	Cases
1	anyschpolgoodrelnurCHILDSAT	0.846	0.794	0.243	0.152	Al‐Sheyab 2012; Berg 2004; Bruzzese 2004; Bruzzese 2008; Kintner 2012
2	anyschpolGOODRELPARgoodrelnur	0.979	0.972	0.187	0.095	Bruzzese 2008; Joseph 2013
	M1	0.884	0.849	0.339
QCA: qualitative comparative analysis. [Notation: Upper case = condition is present; Lower case = condition is absent; * = logical and; + logical or; Key: ANYSCHPOL = School asthma policy; GOODRELNUR = Good level of engagement and/or developing relationships with school nurses; GOODRELPAR = Good level of reported in engagement and/or developing relationships with parents; CHILDSAT = Children reported as satisfied; SUCCESSFULIMPLEMENTATION = Implementation of intervention successful]

Table 30. Complex solution for QCA model 5 ‐ stakeholder involvement and engagement

Table 31. Intermediate solution for QCA model 5 ‐ stakeholder involvement and engagement

		Consistency score with subset relationship (n = 27 in each assessment)	Proportional reduction in inconsistency	Raw coverage	Unique coverage	Cases
1	goodrelnur*CHILDSAT	0.846	0.794	0.243	0.152	Al‐Sheyab 2012; Berg 2004; Bruzzese 2004; Bruzzese 2008; Kintner 2012
2	anyschpolGOODRELGPARgoodrelnur	0.979	0.972	0.187	0.095	Bruzzese 2008; Joseph 2010
	M1	0.884	0.849	0.339
QCA: qualitative comparative analysis. [Notation: Upper case = condition is present; Lower case = condition is absent; * = logical and; + logical or; Key: ANYSCHPOL = School asthma policy; GOODRELNUR = Good level of engagement and/or developing relationships with school nurses; GOODRELGPAR = Good level of reported in engagement and/or developing relationships with parents; CHILDSAT = Children reported as satisfied; SUCCESSFULIMPLEMENTATION = Implementation of intervention successful]

Table 31. Intermediate solution for QCA model 5 ‐ stakeholder involvement and engagement

Comparison 1. Effects of school‐based asthma interventions vs usual care

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Exacerbations leading to hospitalisation Show forest plot	6	1873	Std. Mean Difference (Random, 95% CI)	‐0.19 [‐0.35, ‐0.04]

2 Exacerbations leading to emergency department (ED) visits Show forest plot	13	3883	Odds Ratio (Random, 95% CI)	0.70 [0.53, 0.92]

3 Absence from school Show forest plot	10	4609	Std. Mean Difference (Random, 95% CI)	‐0.07 [‐0.22, 0.08]

4 Days of restricted activity Show forest plot	3	1852	Std. Mean Difference (Random, 95% CI)	‐0.30 [‐0.41, ‐0.18]

5 Unplanned visit to hospital or GP due to asthma symptoms Show forest plot	5	3490	Odds Ratio (Random, 95% CI)	0.74 [0.60, 0.90]

6 Experience of daytime and night‐time symptoms ‐ daytime symptoms Show forest plot	5	1065	Std. Mean Difference (Random, 95% CI)	‐0.15 [‐0.33, 0.02]

7 Experience of daytime and night‐time symptoms ‐ night‐time symptoms Show forest plot	4	459	Std. Mean Difference (Random, 95% CI)	‐0.18 [‐0.52, 0.15]

8 Use of reliever therapies, e.g. beta₂‐agonists Show forest plot	2	437	Odds Ratio (Random, 95% CI)	0.52 [0.15, 1.81]

9 Corticosteroid dosage and/or use of add‐on therapies (usage of) Show forest plot	3	614	Odds Ratio (Random, 95% CI)	1.25 [0.88, 1.77]

10 Corticosteroid dosage and/or use of add‐on therapies (appropriate usage of) Show forest plot	2		Std. Mean Difference (Random, 95% CI)	Totals not selected

11 Health‐related quality of life (SMD) Show forest plot	7	2587	Std. Mean Difference (Random, 95% CI)	0.27 [0.18, 0.36]

12 Health‐related quality of life (MD) Show forest plot	8	2950	Mean Difference (IV, Random, 95% CI)	0.35 [0.06, 0.64]

13 Withdrawal from the study Show forest plot	13	3442	Odds Ratio (Random, 95% CI)	1.14 [0.92, 1.43]

Comparison 1. Effects of school‐based asthma interventions vs usual care

Comparison 2. Effects of school‐based asthma interventions vs usual care subgrouped by school type

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Exacerbations leading to emergency department (ED) visits Show forest plot	13		Odds Ratio (Random, 95% CI)	0.70 [0.53, 0.92]

1.1 Secondary/high school	2		Odds Ratio (Random, 95% CI)	0.61 [0.40, 0.92]
1.2 Primary/elementary school	11		Odds Ratio (Random, 95% CI)	0.73 [0.52, 1.02]
2 Absence from school Show forest plot	10		Std. Mean Difference (Random, 95% CI)	‐0.07 [‐0.22, 0.08]

2.1 Secondary/high school	1		Std. Mean Difference (Random, 95% CI)	‐0.38 [‐0.62, ‐0.15]
2.2 Primary/elementary school	7		Std. Mean Difference (Random, 95% CI)	‐0.05 [‐0.27, 0.16]
2.3 Primary/elementary and middle schools	1		Std. Mean Difference (Random, 95% CI)	0.02 [‐0.08, 0.12]
2.4 Middle school	1		Std. Mean Difference (Random, 95% CI)	0.08 [‐0.31, 0.48]
3 Withdrawal from the study Show forest plot	13		Odds Ratio (Random, 95% CI)	1.14 [0.92, 1.43]

3.1 Secondary/high school	3		Odds Ratio (Random, 95% CI)	1.25 [0.76, 2.06]
3.2 Primary/elementary school	8		Odds Ratio (Random, 95% CI)	1.22 [0.94, 1.59]
3.3 Middle school	2		Odds Ratio (Random, 95% CI)	0.59 [0.29, 1.20]

Comparison 2. Effects of school‐based asthma interventions vs usual care subgrouped by school type

Comparison 3. Effects of school‐based asthma interventions vs usual care subgrouped by age of children

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Exacerbations leading to emergency department (ED) visits Show forest plot	13		Odds Ratio (Random, 95% CI)	0.70 [0.53, 0.92]

1.1 Aged 11 to 15, 16 to 18	1		Odds Ratio (Random, 95% CI)	0.59 [0.39, 0.91]
1.2 Aged 11 to 15	1		Odds Ratio (Random, 95% CI)	1.04 [0.17, 6.25]
1.3 Aged 5 to 10, 11 to 15	2		Odds Ratio (Random, 95% CI)	0.64 [0.15, 2.76]
1.4 Aged 5 to 10	9		Odds Ratio (Random, 95% CI)	0.74 [0.51, 1.06]
2 Absence from school Show forest plot	10		Std. Mean Difference (Random, 95% CI)	‐0.07 [‐0.22, 0.08]

2.1 Aged 11 to 15, 16 to 18	1		Std. Mean Difference (Random, 95% CI)	‐0.38 [‐0.62, ‐0.15]
2.2 Aged 5 to 10, 11 to 15	4		Std. Mean Difference (Random, 95% CI)	0.03 [‐0.06, 0.13]
2.3 Aged 5 to 10	5		Std. Mean Difference (Random, 95% CI)	‐0.09 [‐0.34, 0.16]
3 Withdrawal from the study Show forest plot	13		Odds Ratio (Random, 95% CI)	1.14 [0.92, 1.43]

3.1 Aged 11 to 15, 16 to 18	1		Odds Ratio (Random, 95% CI)	1.31 [0.76, 2.27]
3.2 Aged 11 to 15	3		Odds Ratio (Random, 95% CI)	0.82 [0.25, 2.67]
3.3 Aged 5 to 10, 11 to 15	4		Odds Ratio (Random, 95% CI)	1.08 [0.48, 2.43]
3.4 Aged 5 to 10	5		Odds Ratio (Random, 95% CI)	1.19 [0.90, 1.58]

Comparison 3. Effects of school‐based asthma interventions vs usual care subgrouped by age of children

Comparison 4. Effects of school‐based asthma interventions vs usual care subgrouped by child socio‐economic status (SES)

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Exacerbations leading to emergency department (ED) visits Show forest plot	13		Odds Ratio (Random, 95% CI)	0.70 [0.53, 0.92]

1.1 Low SES over 50%	2		Odds Ratio (Random, 95% CI)	0.53 [0.30, 0.94]
1.2 Low SES over 25%	3		Odds Ratio (Random, 95% CI)	0.69 [0.28, 1.69]
1.3 Unclear or not low SES	8		Odds Ratio (Random, 95% CI)	0.76 [0.57, 1.01]
2 Absence from school Show forest plot	10		Std. Mean Difference (Random, 95% CI)	‐0.07 [‐0.22, 0.08]

2.1 Low SES over 50%	2		Std. Mean Difference (Random, 95% CI)	0.01 [‐0.09, 0.11]
2.2 Low SES over 25%	2		Std. Mean Difference (Random, 95% CI)	‐0.23 [‐0.36, ‐0.09]
2.3 Unclear or not low SES	6		Std. Mean Difference (Random, 95% CI)	‐0.02 [‐0.28, 0.24]
3 Withdrawal from the study Show forest plot	13		Odds Ratio (Random, 95% CI)	1.14 [0.92, 1.43]

3.1 Low SES over 50%	1		Odds Ratio (Random, 95% CI)	1.27 [0.90, 1.78]
3.2 Low SES over 25%	4		Odds Ratio (Random, 95% CI)	1.16 [0.61, 2.23]
3.3 Unclear or not low SES	8		Odds Ratio (Random, 95% CI)	1.03 [0.73, 1.45]

Comparison 4. Effects of school‐based asthma interventions vs usual care subgrouped by child socio‐economic status (SES)

Comparison 5. Effects of school‐based asthma interventions vs usual care subgrouped by involvement of school staff in direct delivery of self‐management skills to children

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Exacerbations leading to emergency department (ED) visits Show forest plot	13		Odds Ratio (Random, 95% CI)	0.70 [0.53, 0.92]

1.1 Teachers involved in delivery (with or without school nurses)	2		Odds Ratio (Random, 95% CI)	1.00 [0.55, 1.83]
1.2 School nurses alone involved in delivery	1		Odds Ratio (Random, 95% CI)	0.29 [0.07, 1.21]
1.3 Existing school staff not involved in delivery	10		Odds Ratio (Random, 95% CI)	0.69 [0.51, 0.94]
2 Absence from school Show forest plot	10		Std. Mean Difference (Random, 95% CI)	‐0.07 [‐0.22, 0.08]

2.1 School nurses or teachers involved in delivery	3		Std. Mean Difference (Random, 95% CI)	0.08 [‐0.08, 0.24]
2.2 Existing school staff not involved in delivery	7		Std. Mean Difference (Random, 95% CI)	‐0.16 [‐0.32, ‐0.00]
3 Withdrawal from the study Show forest plot	13		Odds Ratio (Random, 95% CI)	1.14 [0.92, 1.43]

3.1 School nurses involved in delivery	1		Odds Ratio (Random, 95% CI)	5.67 [0.16, 195.90]
3.2 Existing school staff not involved in delivery	12		Odds Ratio (Random, 95% CI)	1.14 [0.91, 1.42]

Comparison 5. Effects of school‐based asthma interventions vs usual care subgrouped by involvement of school staff in direct delivery of self‐management skills to children

Comparison 6. Effects of school‐based asthma interventions vs usual care subgrouped by explicit use of theory

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Exacerbations leading to emergency department (ED) visits Show forest plot	13		Odds Ratio (Random, 95% CI)	0.70 [0.53, 0.92]

1.1 Theoretical framework utilised explicitly	10		Odds Ratio (Random, 95% CI)	0.75 [0.54, 1.04]
1.2 Use of theory not explicit	3		Odds Ratio (Random, 95% CI)	0.56 [0.33, 0.97]
2 Absence from school Show forest plot	10		Std. Mean Difference (Random, 95% CI)	‐0.07 [‐0.22, 0.08]

2.1 Theoretical framework utilised explicitly	6		Std. Mean Difference (Random, 95% CI)	‐0.19 [‐0.35, ‐0.03]
2.2 Use of theory not explicit	4		Std. Mean Difference (Random, 95% CI)	0.08 [‐0.05, 0.20]
3 Withdrawal from the study Show forest plot	13		Odds Ratio (Random, 95% CI)	1.14 [0.92, 1.43]

3.1 Theoretical framework utilised explicitly	12		Odds Ratio (Random, 95% CI)	1.22 [0.97, 1.54]
3.2 Use of theory not explicit	1		Odds Ratio (Random, 95% CI)	0.61 [0.30, 1.26]

Comparison 6. Effects of school‐based asthma interventions vs usual care subgrouped by explicit use of theory

Comparison 7. Effects of school‐based asthma interventions vs usual care subgrouped by whether design included active inclusion or participation of parents

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Exacerbations leading to emergency department (ED) visits Show forest plot	13		Odds Ratio (Random, 95% CI)	0.70 [0.53, 0.92]

1.1 Parents actively included	8		Odds Ratio (Random, 95% CI)	0.82 [0.53, 1.25]
1.2 Not included/unclear	5		Odds Ratio (Random, 95% CI)	0.58 [0.42, 0.81]
2 Absence from school Show forest plot	10		Std. Mean Difference (Random, 95% CI)	‐0.07 [‐0.22, 0.08]

2.1 Parents actively included	7		Std. Mean Difference (Random, 95% CI)	‐0.02 [‐0.23, 0.18]
2.2 Not included/unclear	3		Std. Mean Difference (Random, 95% CI)	‐0.18 [‐0.50, 0.15]
3 Withdrawal from the study Show forest plot	13		Odds Ratio (Random, 95% CI)	1.14 [0.92, 1.43]

3.1 Parents actively included	9		Odds Ratio (Random, 95% CI)	1.21 [0.93, 1.58]
3.2 Not included/unclear	4		Odds Ratio (Random, 95% CI)	0.97 [0.62, 1.53]

Comparison 7. Effects of school‐based asthma interventions vs usual care subgrouped by whether design included active inclusion or participation of parents

Comparison 8. Effects of school‐based asthma interventions vs usual care subgrouped by timing of intervention

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Exacerbations leading to emergency department (ED) visits Show forest plot	13		Odds Ratio (Random, 95% CI)	0.70 [0.53, 0.92]

1.1 Intervention mainly delivered during students' free time	5		Odds Ratio (Random, 95% CI)	0.71 [0.45, 1.13]
1.2 Intervention took place during school day (exact time unclear or variable)	8		Odds Ratio (Random, 95% CI)	0.67 [0.48, 0.92]
2 Absence from school Show forest plot	10		Std. Mean Difference (Random, 95% CI)	‐0.07 [‐0.22, 0.08]

2.1 Intervention mainly delivered during students' free time	2		Std. Mean Difference (Random, 95% CI)	‐0.23 [‐0.36, ‐0.11]
2.2 Intervention took place during school day (exact time unclear or variable)	8		Std. Mean Difference (Random, 95% CI)	‐0.01 [‐0.17, 0.16]
3 Withdrawal from the study Show forest plot	13		Odds Ratio (Random, 95% CI)	1.14 [0.92, 1.43]

3.1 Intervention took place during class time	1		Odds Ratio (Random, 95% CI)	13.57 [0.34, 542.83]
3.2 Intervention mainly delivered during students' free time	4		Odds Ratio (Random, 95% CI)	1.19 [0.65, 2.16]
3.3 Intervention took place during school day (exact time unclear or variable)	8		Odds Ratio (Random, 95% CI)	1.13 [0.89, 1.43]

Comparison 8. Effects of school‐based asthma interventions vs usual care subgrouped by timing of intervention

Comparison 9. Effects of school‐based asthma interventions vs usual care subgrouped by configuration of conditions

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Exacerbations leading to emergency department (ED) visits Show forest plot	13		Odds Ratio (Random, 95% CI)	0.70 [0.53, 0.92]

1.1 PI1 ‐ theory, not in own time, no substantial school nurse involvement	5		Odds Ratio (Random, 95% CI)	0.85 [0.47, 1.52]
1.2 PI2 ‐ theory, not individual, substantial school nurse involvement	0		Odds Ratio (Random, 95% CI)	0.0 [0.0, 0.0]
1.3 Other configuration	8		Odds Ratio (Random, 95% CI)	0.67 [0.47, 0.94]
2 Absence from school Show forest plot	10		Std. Mean Difference (Random, 95% CI)	‐0.07 [‐0.22, 0.08]

2.1 PI1 ‐ theory, not in own time, no substantial school nurse involvement	4		Std. Mean Difference (Random, 95% CI)	‐0.10 [‐0.46, 0.25]
2.2 PI2 ‐ theory, not individual, substantial school nurse involvement	0		Std. Mean Difference (Random, 95% CI)	0.0 [0.0, 0.0]
2.3 Other configuration	6		Std. Mean Difference (Random, 95% CI)	‐0.05 [‐0.21, 0.12]
3 Withdrawal from the study Show forest plot	13		Odds Ratio (Random, 95% CI)	1.14 [0.92, 1.43]

3.1 PI1 ‐ theory, not in own time, no substantial school nurse involvement	4		Odds Ratio (Random, 95% CI)	0.88 [0.55, 1.40]
3.2 PI2 ‐ theory, not individual, substantial school nurse involvement	1		Odds Ratio (Random, 95% CI)	5.67 [0.16, 195.90]
3.3 Other configuration	8		Odds Ratio (Random, 95% CI)	1.23 [0.95, 1.58]

Comparison 9. Effects of school‐based asthma interventions vs usual care subgrouped by configuration of conditions

Comparison 10. Effects of school‐based asthma interventions vs usual care subgrouped by number of consistent conditions (use of theory, parental involvement, not in own time)

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Exacerbations leading to emergency department (ED) visits Show forest plot	13		Odds Ratio (Random, 95% CI)	0.70 [0.53, 0.92]

1.1 No conditions	0		Odds Ratio (Random, 95% CI)	0.0 [0.0, 0.0]
1.2 One condition	3		Odds Ratio (Random, 95% CI)	0.56 [0.33, 0.97]
1.3 Two conditions	7		Odds Ratio (Random, 95% CI)	0.67 [0.49, 0.94]
1.4 Three conditions	3		Odds Ratio (Random, 95% CI)	1.48 [0.65, 3.40]
2 Absence from school Show forest plot	10		Std. Mean Difference (Random, 95% CI)	‐0.07 [‐0.22, 0.08]

2.1 No conditions	0		Std. Mean Difference (Random, 95% CI)	0.0 [0.0, 0.0]
2.2 One condition	3		Std. Mean Difference (Random, 95% CI)	0.02 [‐0.08, 0.11]
2.3 Two conditions	4		Std. Mean Difference (Random, 95% CI)	‐0.16 [‐0.43, 0.11]
2.4 Three conditions	3		Std. Mean Difference (Random, 95% CI)	0.07 [‐0.22, 0.37]
3 Withdrawal from the study Show forest plot	13		Odds Ratio (Random, 95% CI)	1.14 [0.92, 1.43]

3.1 No conditions	0		Odds Ratio (Random, 95% CI)	0.0 [0.0, 0.0]
3.2 One condition	1		Odds Ratio (Random, 95% CI)	0.61 [0.30, 1.26]
3.3 Two conditions	7		Odds Ratio (Random, 95% CI)	1.22 [0.83, 1.80]
3.4 Three conditions	5		Odds Ratio (Random, 95% CI)	1.22 [0.91, 1.64]

Comparison 10. Effects of school‐based asthma interventions vs usual care subgrouped by number of consistent conditions (use of theory, parental involvement, not in own time)

Comparison 11. Adjunct analyses ‐ impact of Implementation on selected outcomes

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Exacerbations leading to emergency department (ED) visits Show forest plot	7		Std. Mean Difference (Random, 95% CI)	‐0.21 [‐0.37, ‐0.04]

1.1 Successful implementation	4		Std. Mean Difference (Random, 95% CI)	‐0.26 [‐0.48, ‐0.04]
1.2 Potential issues in adherence, attrition, or dosage	3		Std. Mean Difference (Random, 95% CI)	‐0.09 [‐0.28, 0.10]
2 Absence from school Show forest plot	7		Std. Mean Difference (Random, 95% CI)	‐0.12 [‐0.28, 0.04]

2.1 Successful implementation	3		Std. Mean Difference (Random, 95% CI)	‐0.28 [‐0.39, ‐0.18]
2.2 Potential issues in adherence, attrition, or dosage	4		Std. Mean Difference (Random, 95% CI)	0.04 [‐0.09, 0.18]

Comparison 11. Adjunct analyses ‐ impact of Implementation on selected outcomes

Comparison 12. Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Exacerbations leading to hospitalisation ‐ standardised mean difference Show forest plot	3	719	Std. Mean Difference (Random, 95% CI)	‐0.20 [‐0.36, ‐0.03]

2 Exacerbations leading to hospitalisation ‐ odds ratio Show forest plot	3	1154	Odds Ratio (Random, 95% CI)	0.71 [0.37, 1.36]

3 Exacerbations leading to hospitalisation ‐ harmonised effect sizes Show forest plot	6	1873	Std. Mean Difference (Random, 95% CI)	‐0.19 [‐0.35, ‐0.04]

4 Exacerbations leading to emergency department (ED) visits ‐ standardised mean difference Show forest plot	4	736	Std. Mean Difference (Random, 95% CI)	‐0.22 [‐0.38, ‐0.05]

5 Exacerbations leading to emergency department (ED) visits ‐ odds ratio Show forest plot	9	3147	Odds Ratio (Random, 95% CI)	0.74 [0.47, 1.16]

6 Exacerbations leading to emergency department (ED) visits ‐ harmonised effect sizes Show forest plot	13	3883	Odds Ratio (Random, 95% CI)	0.70 [0.53, 0.92]

7 Absence from school ‐ standardised mean difference Show forest plot	6	2720	Std. Mean Difference (Random, 95% CI)	‐0.10 [‐0.30, 0.11]

8 Absence from school ‐ odds ratio Show forest plot	4	1889	Odds Ratio (Random, 95% CI)	0.91 [0.59, 1.42]

9 Absence from school ‐ harmonised effect sizes Show forest plot	10	4609	Std. Mean Difference (Random, 95% CI)	‐0.08 [‐0.22, 0.07]

10 Days of restricted activity ‐ standardised mean difference Show forest plot	2	536	Std. Mean Difference (Random, 95% CI)	‐0.34 [‐0.52, ‐0.15]

11 Days of restricted activity ‐ odds ratio Show forest plot	1		Odds Ratio (Random, 95% CI)	Subtotals only

12 Days of restricted activity ‐ harmonised effect sizes Show forest plot	3	1852	Std. Mean Difference (Random, 95% CI)	‐0.30 [‐0.41, ‐0.18]

13 Experience of daytime and night‐time symptoms ‐ daytime symptoms ‐ standardised mean difference Show forest plot	3	762	Std. Mean Difference (Random, 95% CI)	‐0.15 [‐0.33, 0.04]

14 Experience of daytime and night‐time symptoms ‐ daytime symptoms ‐ odds ratio Show forest plot	2	303	Odds Ratio (Random, 95% CI)	0.71 [0.32, 1.55]

15 Experience of daytime and night‐time symptoms ‐ daytime symptoms ‐ harmonised effect sizes Show forest plot	5	1065	Std. Mean Difference (Random, 95% CI)	‐0.15 [‐0.32, 0.02]

16 Experience of daytime and night‐time symptoms ‐ night‐time symptoms ‐ standardised mean difference Show forest plot	3	323	Std. Mean Difference (Random, 95% CI)	‐0.36 [‐0.58, ‐0.14]

17 Experience of daytime and night‐time symptoms ‐ night‐time symptoms ‐ odds ratio Show forest plot	1	136	Odds Ratio (Random, 95% CI)	1.24 [0.56, 2.72]

18 Experience of daytime and night‐time symptoms ‐ night‐time symptoms ‐ harmonised effect sizes Show forest plot	4		Std. Mean Difference (Random, 95% CI)	‐0.18 [‐0.52, 0.15]

19 Use of reliever therapies, e.g. beta₂‐agonists ‐ odds ratio Show forest plot	2	437	Odds Ratio (Random, 95% CI)	0.52 [0.15, 1.81]

20 Corticosteroid dosage and/or use of add‐on therapies (usage of) Show forest plot	3	614	Odds Ratio (Random, 95% CI)	1.25 [0.88, 1.79]

21 Corticosteroid dosage and/or use of add‐on therapies (appropriate usage of) Show forest plot	2		Std. Mean Difference (Random, 95% CI)	Totals not selected

22 Health‐related quality of life ‐ standardised mean difference Show forest plot	7	2502	Std. Mean Difference (Random, 95% CI)	0.27 [0.18, 0.36]

23 Health‐related quality of life (MD) Show forest plot	8	2950	Mean Difference (IV, Random, 95% CI)	0.35 [0.06, 0.64]

24 Unplanned visit to hospital or GP due to asthma symptoms ‐ standardised mean difference Show forest plot	1	280	Std. Mean Difference (Random, 95% CI)	‐0.28 [‐0.52, ‐0.05]

25 Unplanned visit to hospital or GP due to asthma symptoms ‐ odds ratio Show forest plot	4	1316	Odds Ratio (Random, 95% CI)	0.78 [0.62, 0.98]

26 Unplanned visit to hospital or GP due to asthma symptoms ‐ harmonised effect sizes Show forest plot	5	1596	Odds Ratio (Random, 95% CI)	0.74 [0.60, 0.90]

27 Withdrawal from the study Show forest plot	13	3442	Odds Ratio (Random, 95% CI)	1.14 [0.92, 1.43]

Comparison 12. Effects of school‐based asthma interventions vs usual care, including disaggregated effect sizes