Research reportPredicting future depression in adolescents using the Short Mood and Feelings Questionnaire: A two-nation study
Introduction
Experienced by up to 5% of the adolescent population at any one time (Sawyer et al., 2010), depression is associated with many adverse outcomes in adulthood, including lowered educational attainment (Kessler et al., 1995); increased risk of substance use and abuse (McKenzie et al., 2010c, O'Donnell et al., 2006, Patton et al., 2002); and increased risk of suicide (Costello et al., 2003). There has recently been a renewal of interest in the study of this disorder at the level of individual symptoms, particularly in the identification of depression in community settings (Lowe et al., 2010, Rhew et al., 2010). Although few overall studies of depressive symptoms of adolescents exist (Patton et al., 2000), several studies have attempted prediction of later episodes of depression from specific symptoms such as feelings of worthlessness (Pine et al., 1999, van Lang et al., 2007, Van Voorhees et al., 2008, Wilcox and Anthony, 2004). Predicting the development of psychiatric disorders in adolescence by detecting key early symptoms may aid in the screening and treatment of such disorders (McGorry et al., 2000, van Lang et al., 2007).
Unfortunately, few large prospective cohort studies in young people with the scope to examine the predictive value of specific depressive symptoms have been conducted (van Lang et al., 2007). Understanding the development of depressive symptoms in early adolescence requires repetition of symptom measures over critical periods. The findings around predicting depression several years into the future have so far been mixed. For example, prediction of depression at ages 14 and 21, employing symptoms reported at age five, had very limited accuracy (Najman et al., 2008). Symptoms reported in late adolescence, however, were associated with depression in young adulthood, assessed six (van Lang et al., 2007) or seven (Pine et al., 1999) years later. Similarly, symptoms of internalising disorder, such as timidity, sadness, and tiredness, exhibited at ages 13 and 15 were associated with psychiatric disorders, including depression, up to 40 years later (Colman et al., 2007).
Although early adolescence is important for the emergence of depression, there has been little research into using depressive symptoms present in this life stage to predict subsequent depression over shorter time periods (Pine et al., 1999). One recent study found that feelings of sadness and moodiness were highly predictive of adolescent depression 12 months later (Van Voorhees et al., 2008). The present study is concerned with predicting high levels of depressive symptoms in early adolescence, utilising a large sample of Australian and American adolescents, over a 12 month period. We employ approaches based upon the quality or type, as well as the quantity, of depression symptoms present. The former approach makes use of classification and regression trees (CART) (Breiman et al., 1984, Gruenewald et al., 2008, Strobl et al., 2009), as well as logistic regression. Classification trees make few assumptions of the data, are generally readily interpretable by clinicians and researchers (Gruenewald et al., 2008), and can be used to identify salient combinations of symptoms.
Identification of symptom patterns that are predictive of high levels of adolescent depression may aid insight into the underlying aetiology of depression, as well as inform the development of more efficient epidemiological tests for adolescent depression. Such tests can aid in planning prevention and early intervention programmes in school and other settings (Rhew et al., 2010, van Lang et al., 2007).
We hypothesise that the presence of depressive symptoms at the first data collection point will be predictive of high levels of symptoms 12 months later. Our examination of which specific combinations of symptoms are most predictive extends earlier research that concentrated on the total number of symptoms and/or on individual symptoms (Pine et al., 1999, Rhew et al., 2010, van Lang et al., 2007, Wilcox and Anthony, 2004). The purpose of our study is to identify such combinations of symptoms, and compare their predictive performance with that of an approach based upon the number of symptoms. Finally, we discuss how such approaches can be applied to screening for depression.
Section snippets
Recruitment and data collections
We employed a large sample of 5769 students on whom information was collected as part of a binational study of youth development, in the US state of Washington (WA), and the Australian state of Victoria (VIC). In the first stage of a two-stage cluster sampling procedure, schools were randomly selected from a stratified sampling frame of all of the schools in WA (public, private, and alternative) and VIC (Catholic, independent, and government). In the second stage of sampling, single intact
Analysis of individual items
A binary coding of “not true” and “sometimes true” versus “true” is often employed in research involving the SMFQ (Angold et al., 1995). Our initial CART analyses of baseline SMFQ items indicated, however, that the merging of “sometimes true” and “true” was more highly associated (had higher Gini coefficients, described below) with SMFQ caseness at 12 months, than was the merging of “not true” and “sometimes true”. The former coding was therefore employed in all analyses of individual items.
The
Discussion
Our study contributes new information about the utility of early adolescent symptoms in predicting subsequent high levels of depression symptoms. A relatively low level of depression (cut-off of seven or higher on the total SMFQ score, and a subset of four items) can provide efficient prediction of later elevated levels of adolescent depression. The quality or type of symptoms is also important. For example, low self-worth was more predictive of high levels of depression symptomatology at 12
Role of funding source
Data collection for this research was supported through a grant from the National Institute on Drug Abuse (DA-012140-05), National Institutes of Health, United States Department of Health and Human Services (RF Catalano, Principal Investigator). Data analysis was supported by grants from the Australian Research Council and Australian Health Management. DP McKenzie is supported through a Postdoctoral Public Health Fellowship from the National Health and Medical Research Council of Australia
Conflict of interest
All authors declare that they have no conflicts of interest.
Acknowledgements
We sincerely thank the participants and associated research staff for their participation in the study.
References (69)
- et al.
A case study found that a regression tree outperformed multiple linear regression in predicting the relationship between impairments and Social and Predictive Activities scores
Journal of Clinical Epidemiology
(2005) - et al.
Logistic regression had superior performance compared with regression trees for predicting in-hospital mortality in patients hospitalized with heart failure
Journal of Clinical Epidemiology
(2010) - et al.
Dimensions of psychopathology in the medically ill: a latent trait approach
Psychosomatics
(2000) - et al.
A 2-item screen for depression in rehabilitation inpatients
Archives of Physical Medical Rehabilitation
(2005) - et al.
Should screening for depression among children and adolescents be demedicalized?
Journal of the American Academy of Child and Adolescent Psychiatry
(2009) - et al.
Anxiety and depression screening for youth in a primary care population
Ambul. Pediatr.
(2008) - et al.
A 4-item measure of depression and anxiety: validation and standardization of the Patient Health Questionnaire-4 (PHQ-4) in the general population
Journal of Affective Disorders
(2010) - et al.
Can we improve the diagnostic efficiency and predictive power of prodromal symptoms for schizophrenia?
Schizophr. Res.
(2000) - et al.
Hazardous or harmful alcohol use in Royal Australian Navy veterans of the 1991 Gulf War: identification of high risk subgroups
Addict. Behav.
(2006) - et al.
Pessimism, worthlessness, anhedonia and thoughts of death identify DSM-IV major depression in hospitalized medically ill
Psychosomatics
(2010)
Self-disparagement as feature and forerunner of depression: findings from the Stirling County Study
Compr. Psychiatry
Screening in early childhood for risk of later mental health problems: a longitudinal study
Journal of Psychiatric Research
Predicting female depression across puberty: a two-nation longitudinal study
Journal of the American Academy of Child and Adolescent Psychiatry
On the relationship between neuroticism, self-esteem, and depression: results from the National Comorbidity survey
Compr. Psychiatry
Predictors of future depression in early and late adolescence
Journal of Affective Disorders
Child and adolescent clinical features as forerunners of adult-onset major depressive disorder: retrospective evidence from an epidemiological sample
Journal of Affective Disorders
Models developed by three techniques did not achieve acceptable prediction of binary trauma outcomes
Journal of Clinical Epidemiology
Introduction to Machine Learning
Development of a short questionnaire for use in epidemiological studies of depression in children and adolescents
International Journal of Methods in Psychiatric Research
Depression scale scores in 8-17-year-olds: effects of age and gender
Journal of Child Psychology and Psychiatry
Modifiable risk factors predicting major depression disorder at four year follow-up: a decision-tree approach
BMC Psychiatry
Classification and Regression Trees
Applying recursive partitioning to a prospective study of factors associated with adherence to mammography screening guidelines
American Journal of Epidemiology
Forty-year psychiatric outcomes following assessment for internalizing disorder in adolescence
American Journal of Psychiatry
Prevalences and development of psychiatric disorders in childhood and adolescence
Archives of General Psychiatry
Outcome in schizophrenia and related disorders compared between developing and developed countries: a recursive partitioning reanalysis of the WHO DOSMD data
British Journal of Psychiatry
Relationship between dehydropiandrosterone sulphate levels and cognitive function in women
Journal of Clinical Endocrinology and Metabolism
An Introduction to the Bootstrap
Persuasion and Healing: a Comparative Study of Psychotherapy
Diverse pathways to positive and negative affect in adulthood and later life: an integrative approach using recursive partitioning
Developmental Psychology
Early identification and intervention in depressive disorders: towards a clinical staging model
Psychother. Psychosom.
Stability in depressed mood in adolescence: results from a 6-year longitudinal study
Journal of Youth and Adolescence
Applied Logistic Regression
Social consequences of psychiatric disorders, I: Educational Attainment
American Journal of Psychiatry
Cited by (46)
Machine learning-based prediction for self-harm and suicide attempts in adolescents
2023, Psychiatry ResearchSymptom centrality and infrequency of endorsement identify adolescent depression symptoms more strongly associated with life satisfaction
2021, Journal of Affective DisordersCognitive predictors of adolescent social anxiety
2021, Behaviour Research and TherapyIs social media screen time really associated with poor adolescent mental health? A time use diary study
2020, Journal of Affective DisordersKey Hospital Anxiety and Depression Scale (HADS) items associated with DSM-IV depressive and anxiety disorder 12-months post traumatic brain injury
2018, Journal of Affective DisordersCitation Excerpt :ROC analysis was also employed to determine the cut-point, known as the empirical cut-point, for the HADS depression and anxiety subscales, for the current study (Dahm et al., 2013; Schwarzbold et al., 2014; Whelan-Goodinson et al., 2009). Previous studies have employed a wide range of statistical and computational procedures including artificial neural networks (Lu et al., 2015; Pourahmad et al., 2016), stepwise regression (Green et al., 2010; Hunt et al., 2017; Izal et al., 2010) and recursive partitioning or classification trees (Exarchos et al., 2012; McKenzie et al., 2010, 2015, 2011; Ponsford et al., 2016) to identify a small number of key items. General comparisons of different techniques have not found any one technique to be consistently better than the others (Witten et al., 2017) and so classification or decision trees were chosen on the basis of familiarity and ease of interpretation.