FormalPara Key Points for Decision Makers
Table 1
Full size table

1 Introduction

1.1 Background/Rationale

Serious mental illness (SMI) comprises a set of chronic conditions that affect mood, emotion, cognition and motivation and all aspects of life including employment, relationships, housing, and personal care [1]. UK healthcare policy defines SMI as schizophrenia, bipolar disorder, and other non-organic psychoses [2]; 0.72% of the population in England will receive a diagnosis of schizophrenia at some point in their lives [3], and 1–2% will receive a diagnosis of bipolar disorder [4]. People with SMI have poorer physical health compared to the general population, with a higher prevalence of common chronic disorders and multi-morbidity, and a lower life expectancy by around 20 years, and often die of preventable physical illnesses [5,6,7,8,9]. People with SMI have high rates of healthcare utilisation, including accident and emergency (A&E) attendances and hospital admissions, for both physical and mental health problems [10,11,12,13], and the multiplicity of co-morbid health conditions raises total healthcare costs [14].

A number of studies have examined the direct costs of SMI to the healthcare system as well as the wider indirect costs to society [15,16,17,18,19,20,21,22]. Direct costs include the use of hospital services, drugs, staff time, ambulances, and community care; for schizophrenia alone these costs can range from 0.5 to 3.5% of national healthcare expenditures [19]. Indirect costs include reduced labour supply, premature mortality, reduced health-related quality of life, lost output, lost tax revenue, transfer payments, and unpaid care by family or friends [21]. The total monetary costs in England for schizophrenia and psychoses have been estimated at £11.8 billion in 2012, and for bipolar disorder, they have been estimated at £5.2 billion in 2007 (equivalent to £6.04 billion in 2012).

Studies have used different costing methodologies to estimate direct costs of mental health treatment and the main cost drivers thereof. Cost estimation studies tend to use a gross-costing methodology that often reflects reimbursement amounts, whereas micro-costing methods involve the enumeration of all resources consumed by a patient. A systematic review by Wolff et al. [23] examining the association between mental healthcare hospital costs and patient characteristics considered costing approaches in mental healthcare more broadly, not just for SMI. It found that studies differed in their estimation of resource use and valuation. The paper also identified six patient characteristics as key cost drivers of hospital costs, one being the presence of psychotic or affective symptoms. Studies using a micro-costing approach [24, 25] focused on examining differences in characteristics between high-cost users and the rest of the cost distribution. These show that high-cost users of mental healthcare (incurring the top 10% of costs) account for a disproportionately large share of healthcare costs (74–87%), and that the majority have a diagnosis of SMI (particularly schizophrenia).

Despite the considerable disease burden and costs for this patient group, there has been little empirical cost estimation research specifically focusing on the SMI population, using linked individual-level data and micro-costing approaches, and analysing data for more than 1 year. A further gap in the literature has been the estimation of multi-sectoral costs for people with SMI examining which elements of the care pathway are the main cost drivers across sectors. Most studies examining cost drivers of patient care for this population group have focused on inpatient care and/or on total costs. Understanding where the largest components of cost fall across different sectors, as well as identifying the main cost drivers, is important to help design more cost-effective treatment pathways and more integrated care for this patient group.

We address these gaps in the evidence by seeking to (1) develop a costing methodology for multi-sectoral healthcare costs for people with SMI in England based on the best micro-level data currently available, (2) estimate annual healthcare costs for people with SMI in England across primary and secondary care, and (3) identify predictors of costs across each sector.

The key contributions of this study are that (1) it is the first study to use linked administrative records for the full care pathway of patients with SMI in the English National Health Service (NHS), covering primary care, emergency departments, inpatient admissions, and community mental health services; (2) it covers 3 financial years (2011/12–2013/14), where previous studies have tended to estimate costs for a single year; (3) it employs a detailed bottom-up costing methodology; and (4) it provides evidence on the key clinical and socio-economic cost drivers across different healthcare sectors.

2 Methods

2.1 Study Design

This observational cohort study used healthcare service administrative data from primary care, inpatient admissions, A&E presentations, and community mental health services linked at the individual patient level. Annual costs of healthcare utilisation in each of these sectors was estimated using a bottom-up approach for 3 financial years from 2011/12 to 2013/14, applying national average unit costs for the 2013/14 financial year from the perspective of the NHS.

To provide some indication of the proportion of healthcare costs related to mental health as distinct from physical health for this population, we identified a subset of costs that could be clearly labelled as “related”, comprising all specialist mental healthcare costs (inpatient and outpatient) and mental health drug costs in primary care.

2.2 Data Source and Sample

Data for this study were accessed through the Clinical Practice Research Datalink (CPRD), a dataset of records from primary care practices in the UK [26]. Data on all patients at participating practices are included unless a patient requests to have her/his data excluded. Practices in England can also allow their records to be linked to hospital records in the Hospital Episodes Statistics (HES) and community mental health service records in the Mental Health Minimum Dataset (MHMDS).Footnote 1 CPRD data are broadly representative of the English population regarding age and gender [27]. CPRD provided anonymised primary care data of patients identified as having SMI based on Read codesFootnote 2 in routinely recorded CPRD data, linked at the individual patient level to data from HES and MHMDS for those patients. Supplementary Table 1 provides the Read codes used to identify diagnostic categories of SMI; see the electronic supplementary material. Supplementary Figure 1 outlines how the sample was formulated for this study. Supplementary Table 2 provides an overview of the sources of utilisation data and unit costs. Supplementary Table 3 outlines the mean visit duration for primary care visits of different types.

2.3 Primary Care Costs

Primary care utilisation data were extracted from CPRD records based on Read codes for three types of primary care activity: consultations, drugs prescribed, and diagnostic tests.

Unit costs for consultations were obtained from national estimates of the cost per minute of general practitioner (GP) and practice nurse time for 2014 [29]. We used visit duration data rather than a standard visit cost because of evidence suggesting that consultations for patients with mental health problems are longer than average [30,31,32,33]. We considered multiple visits in a day for a patient to a single staff member as duplicates, but allowed for visits to different staff members on the same day. See the electronic supplementary material for additional detail.

The costs of drug prescriptions were derived from the Prescription Costs AnalysisFootnote 3 for 2014, which applied the British National Formulary (BNF) edition 66 [34] classification system. Prescription volume was taken from the Therapy section of the CPRD dataset. We used data at the BNF subparagraph level where available, otherwise at the most specific BNF level available (paragraph, section, or chapter level).Footnote 4

National average unit costs for diagnostic tests were taken from the NHS Reference Costs for the financial year 2013/14 [35], calculated retrospectively each year to identify the full cost of diagnostic imaging, diagnostic services, pathology services, and outpatient procedures. Utilisation volume of each test was taken from Read terms recorded in the Test section of the CPRD dataset. These were classified following clinicians’ advice into groups that corresponded to the NHS Reference Costs categories.Footnote 5 Supplementary file Test_unit_allocations_and_costs.xlsx lists the Read terms and how they were classified into Reference Cost categories (see the electronic supplementary material).

2.4 Community Mental Health Service Costs

We used a subset of the activity recorded in MHMDS as measures of utilisation of community mental health services. Specifically, we counted healthcare professional contacts (consultations), Care Programme Approach (CPA) reviews, electroconvulsive therapy (ECT) sessions, and day attendances, and assigned unit costs reported by the Personal Social Services Research Unit (PSSRU) and the Electronic Staff Record [36].

The cost of consultations depends on the staff type and the duration of the contact (for details on how we dealt with missing and zero duration see the electronic supplementary material). Based on advice regarding usual practice, we restricted the number of consultations per day to a maximum of two for each type of staff member to avoid duplicate recordings. Regarding CPA reviews, MHMDS provides only the date and not the staff involved or duration. We assumed, following clinical advice, that reviews (which are supposed to be multidisciplinary) involve a psychiatrist and a nurse and last 30 min, a lower bound estimate. Cost per ECT session was based on past costings updated to 2013/14 values using the hospital and community health services pay and price inflation series [37]. Day attendance cost was based on the cost of day care provided by local authority social services for people with mental health problems [29].

2.5 Hospital Costs

Admissions to specialist mental health inpatient facilities are recorded in MHMDS, and admissions to general hospitals (including admissions to non-specialist mental health providers) are recorded in HES. We costed these two separately using national average costs for the type of admission.

2.5.1 Admissions to Specialist Mental Health Inpatient Facilities

The cost of a specialist mental health admission depends on its duration and the category of mental health problem or needs (termed the “cluster”) to which the patient is assigned, since the Reference Costs of admitted days vary by the 21 mental health clusters that exist in the classification system [35]. Admissions recorded in MHMDS cover all admissions to specialist mental health inpatient facilities and list duration, but not the cluster allocation. Instead, we used the closest start of a cluster episode Footnote 6 or date of Mental Health Clustering ToolFootnote 7 to determine the cluster. We excluded admissions with the same start and end dates as these were likely to represent recording errors, and for the 14% of admissions without an assigned cluster, we used a weighted average cost across all clusters. For further detail, see the electronic supplementary material.

2.5.2 General Hospital Costs

We identified inpatient hospital admissions and A&E presentations from HES records. The classification system for hospital activity in England is Healthcare Resource Groups (HRGs), which are similar to Diagnosis-Related Groups (DRGs) used in other jurisdictions. In the electronic supplementary material, we explain how inpatient episodes of care and A&E attendances are grouped into HRGs reflecting type of activity (inpatient or A&E HRGs), additional high-cost inpatient procedures, and long inpatient stays. Costs for these activities were taken from the hospital-reported Reference Costs.

The allocation of episodes to HRGs was carried out using the costing grouper software which is freely available on the NHS Digital website [40]. We grouped all the activity using the 2012/13 version of the HRGs design and the corresponding 2012/13 Reference Costs to obtain costs for each activity, then used the hospital and community health services pay and price inflation index [41] to express costs in 2013/14 prices. For episodes with missing or invalid essential data fields, the grouper returns an error HRG code. After minor adjustments and imputation of some non-essential fields, the error rate for our sample was reduced to about 7% for inpatient episodes and 10% for A&E presentations. Although HES includes admissions to specialist mental health facilities, these were not costed because these were captured through MHMDS, but mental health conditions treated by non-specialist mental health providers (HRGs starting with “WD”) were included in general hospital costs. Using the inpatient episode start and end dates, costs were apportioned into the financial years 2011/12–2013/14. Where an episode of care extended over 2 years, we attributed costs to each year based on the proportion of episode days in each of the 2 years.

2.6 Explanatory Variables

Read codes in CPRD data were used to identify patient characteristics. SMI diagnoses were grouped according to clinical advice into categories of “schizophrenia and other psychoses” and “bipolar disorder and affective psychosis”, with some patients having diagnostic codes from both groups at different times. A variable was formulated to indicate the number of years since diagnosis when the patient entered observation in the study. The 2010 Index of Multiple Deprivation (IMD) quintile for the patient’s area of residence served as an indicator of socioeconomic area deprivation. Based on the primary care practice location, two variables indicating the distance from the practice to the nearest acute hospital and inpatient mental health facility and another indicating the rurality category of the practice were derived.Footnote 8

2.7 Statistical Methods

For each patient, event costs were aggregated into annual costs by sector and year, effectively obtaining a panel dataset of annual costs in primary care, specialist mental healthcare (including inpatient and community-based care), and general hospital care (including A&E), with total costs as the sum of these three. To investigate patterns of total and sector costs across patient and practice characteristics, we compared mean costs and examined the distribution of costs in each group, and conducted multivariate regression analysis. Explanatory variables were based on factors associated with utilisation of healthcare in prior research [24, 25, 42]. We used a generalised linear model (GLM) with individual random effects, which allows for the non-normal distribution of healthcare costs and models the mean directly, avoiding the need to transform the data. The choices of distributional family and link function were informed by the Park test [43] and Pregibon link test [44]. All analyses were conducted in Stata v14 [45].

2.8 Sensitivity Tests

As an alternative to using the consultation duration data in CPRD and MHMDS, we estimated costs using the average duration for different types of consultations derived from the 2006/07 UK General Practice Workforce Survey that are applied by Curtis [29]. For mental health inpatient admissions, we present for comparison the results of costing bed days using the weighted average cost per bed day across all clusters from the national Reference Costs, rather than applying the cluster-specific cost.

To account for potential sex differences in the severity of schizophrenia [46] and bipolar disorder [47], we estimated the model including an interaction term between sex and diagnostic grouping.

3 Results

3.1 Sample

The sample consists of 34,484 year-observations across 13,846 adults with SMI for the financial years 2011/12–2013/14. Two-thirds of the sample (66.6%) were observed for all 3 years, 15.8% were observed for 2 years and 17.6% were observed for one full financial year. Supplementary Figure 1 shows the numbers of patients at each stage of the sample formation.

The first data column of Table 1 shows the distribution of characteristics across the sample. The mean (median) age was 52 (50) years, the majority (71%) were white, around a third had at least one physical comorbidity, two-thirds had comorbid depression, three-quarters were current or ex-smokers, and 16% had been recently diagnosed with SMI. People in the sample were more likely to live in urban areas (89%) than rural areas, and in more deprived areas.

Table 1 Patient characteristics and mean annual healthcare cost by sector and total
Full size table

3.2 Resource Use and Costs

The mean annual healthcare cost per patient was £4988.87 (range £0–243,831), with £2575.73 (52%) identifiable as related to SMI. Figure 1 shows that total costs decreased with age until middle age, and then increased for patients aged over 65, while the proportion of costs related to SMI steadily decreased with age. For patients 19–35 years old, 64% of total costs were related to SMI, while for patients aged over 65, the proportion was 36%, reflecting a greater emphasis on management of physical comorbidities.

Fig. 1
figure 1

Mean annual cost over age groups, comparing total costs and related costs. CI confidence interval, SMI serious mental illness

Full size image

Table 1 shows that primary care costs (£938, range £0–29,720) made up approximately 19% of total mean annual costs, general hospital costs (£1717, range £0–233,318) comprised 34%, and specialist mental healthcare costs (£2334, range £0–176,502) accounted for 47%. Table 2 outlines the annual utilisation of the different resource units making up these costs.

Table 2 Resource utilisation within each sector per year
Full size table

Table 1 also provides mean cost per year for subgroups of patients: mean total costs increase with age, level of deprivation, and number of comorbidities, and are higher for white than black or minority ethnicity patients, and for patients diagnosed with SMI more recently. Patients with both schizophrenia and bipolar-type disorders had higher costs than those with schizophrenia-type alone, with bipolar-type associated with the lowest costs.

3.3 Association of Characteristics with Costs

The Pregibon test supported a log link for GLM (p = 0.104) with a Poisson distribution supported by the Park test (p = 0.483). The association of each patient and practice characteristic with the difference in total annual costs (marginal effects from random effects GLM) is presented in Table 3. After adjusting for other observable characteristics, higher costs for patients of white ethnicity (compared with black and minority ethnicities), those with higher socioeconomic deprivation, and those with physical comorbidities were still seen, as was the pattern of costs across diagnostic categories.

Table 3 Association of patient and practice characteristics with differences in costs by sector (average marginal effects, £ per year)
Full size table

Table 3 shows how these patterns differ across healthcare sectors. White ethnicity and having SMI diagnoses from both categories were consistently associated with higher costs across all sectors. Socioeconomic deprivation was associated with higher primary care and mental health costs. Older patients had higher primary care and general hospital costs, but lower mental health costs. Physical comorbidity was associated with higher costs in primary care and hospital, but those with three or more Charlson index comorbidities had lower specialist mental healthcare costs. Distance from hospital care and inpatient mental healthcare was associated with lower costs in some sectors. Comorbid depression and smoking were associated with higher primary care costs only, and male patients had lower primary care costs.

3.4 Sensitivity Tests

As shown in Table 4, if we used average primary care visit duration instead of the visit-specific duration data in CPRD, the mean annual primary care cost was higher (£985 instead of £938). Changing the costing of mental health admissions in MHMDS from cluster-specific to weighted cluster average costs had little effect, with mean annual mental health costs increasing from £2334 to £2354.

Table 4 Effect of alternate specifications on sectoral cost distributions
Full size table

The set of models with an interaction term between sex and diagnostic grouping found the interaction between male sex and “schizophrenia and other psychoses” to be positive and statistically significant in the models for total costs and mental health costs. The marginal effects from this model presented in Supplementary Table 4 show a larger difference in costs between the base category of “bipolar disorder and affective psychoses” and “schizophrenia and other psychoses” for males than females.

4 Discussion

Estimates of healthcare costs for people with SMI are required to inform resource allocation, to model resource consumption in health technology assessments, and for service planning. Previous analyses of healthcare costs for people with SMI have focused on specific sectors of the healthcare system, offering only a partial picture of overall resource use. Furthermore, there are no published costing algorithms for SMI care in the English NHS, which may explain the limited availability of costing studies in this population. To address these issues, we have developed a micro-costing methodology to estimate the cost of healthcare services used by people with SMI across primary, general hospital and mental healthcare settings in England. We applied this methodology to linked individual-level data to study the determinants of annual healthcare costs for a sample of people with SMI over a 3-year period. To our knowledge, this is the first study to use linked patient-level data from electronic health records to study cost variation within the SMI population.

We estimate average annual care costs of £4989 per patient (in 2013/14 terms, equivalent to £5777 in 2019 terms adjusted using the consumer price index). [48] This is substantially higher than the average annual cost of £1631 for a member of the general public of similar age to our study average (52 years) [49]. Assuming a point prevalence of SMI of 0.9% and a total population of 54.3 million in 2014 [50], the total annual cost to the English NHS is £2.44 billion (equivalent to £2.82 billion in 2019 terms). Previous estimates for schizophrenia and bipolar disorder separately were higher (approximately £2.6 billion and £1.9 billion, respectively, in 2013/14 terms) [51, 52], but were prevalence-based rather than based on individual-level data linked across NHS services.

Costs are highest in the first 12 months after diagnosis and then fall by approximately 11–13% in subsequent years, which is consistent with the patterns observed for other health conditions [53]. This may be due to higher service use during the treatment initiation period following diagnosis, reduced risk of subsequent crisis events, dropout from treatment, or gradual adaptation by patients to their health condition. Moreover, treatment costs can be predicted by several patient characteristics such as local area deprivation and ethnicity. While most of these characteristics have a consistent directional effect across sectors, older age is associated with higher costs in primary care but lower costs in specialised mental healthcare, suggesting a cross-sectoral shift in the cost burden over a patient’s lifetime. Age and diagnostic category were also strong predictors of cost in Wolff et al. [23], albeit the studies included in that review did not include only SMI patients and focused only on inpatient care.

The primary strength of our study is the use of linked longitudinal electronic health records covering primary, secondary and community care, which allows us to implement a micro-costing approach to obtain detailed resource use estimates. These data have only recently become available and provide a comprehensive view of contemporaneous care arrangements in the English routine care setting.

There are a number of limitations to our study that arise from the specific data used. The data do not cover (non-admitted) outpatient care provided by general hospitals, drugs prescribed by community mental health services, or all social care provided in the community. These cost categories are likely to be important as payers are increasingly starting to pool budgets across sectors. The data did not include costs of seeking healthcare (such as travel) or care outside the NHS, nor measures of disease severity, so we cannot ascertain whether severity mediates the association between costs and factors such as socioeconomic status. While GP practices participating in CPRD have been shown to be broadly representative in terms of the age–sex composition and socio-economic profile of their patient population, we cannot ascertain the representativeness of our specific sample, which may be compromised by the more transient nature of care relationships for people with SMI, who are often living in precarious circumstances, and the requirement to be registered with the practice for at least 1 year prior to sample inclusion. The NHS requires patients to be registered with a GP, so we are unlikely to be missing patients managed solely by specialist mental health services. In MHMDS, there was an imperfect correlation between dates assigned for episodes of care (cluster episodes) and dates when healthcare activities were recorded. Moreover, in some instances, patients were recorded to be in hospital while receiving outpatient care. We made the assumption that cluster episode dates were more likely to be wrong and that recorded activities were genuine, which may have led to overestimated costs.

The measurement of visit costs was imprecise due to the absence of band-specific hourly costs for nurses in the PSSRU cost estimates (with all nurses in primary care classified as band 5), and recording errors of visit duration in CPRD. However, our use of the timing of file opening and closing in CPRD does produce estimates that correlate well with duration data available from other sources, and has now been incorporated into the PSSRU unit cost estimates [41]. This analysis is conducted on data from 2011/12 to 2013/14, using unit costs for 2013/14. While we have inflated the average costs to current values using national consumer price inflation figures for the health domain [48], other changes may have occurred over that period not in line with inflation.

5 Conclusions

Estimates of the direct cost of SMI are important for decision makers seeking to improve technical and allocative efficiency of care services. Information from this study can help decision makers in planning services and benchmarking as well as resource allocation. Our study demonstrates that the cost of healthcare resource use across primary, secondary and community care are determined by a range of patient characteristics that are observable in electronic health records. Specialised mental healthcare is the main component of annual costs for people with SMI.