Elsevier

Journal of Affective Disorders

Volume 190, 15 January 2016, Pages 395-406
Journal of Affective Disorders

Research report
The identification of symptom-based subtypes of depression: A nationally representative cohort study

https://doi.org/10.1016/j.jad.2015.10.040Get rights and content

Highlights

  • Depression could best be described in terms of both severity and anxiety comorbidity.

  • Four subtypes were found: severe depression with anxiety, moderate depression with anxiety, moderate depression without anxiety and mild depression.

  • In particular anxiety was a distinguishing feature within moderate depression.

Abstract

Background

In recent years, researchers have used various techniques to elucidate the heterogeneity in depressive symptoms. This study seeks to resolve the extent to which variations in depression reflect qualitative differences between symptom categories and/or quantitative differences in severity.

Methods

Data were used from the Netherlands Mental Health Survey and Incidence Study-2, a nationally representative face-to-face survey of the adult general population. In a subsample of respondents with a lifetime key symptom of depression at baseline and who participated in the first two waves (n=1388), symptom profiles at baseline were based on symptoms reported during their worst lifetime depressive episode. Depressive symptoms and DSM-IV diagnoses were assessed with the Composite International Diagnostic Interview 3.0. Three latent variable techniques (latent class analysis, factor analysis, factor mixture modelling) were used to identify the best subtyping model.

Results

A latent class analysis, adjusted for local dependence between weight change and appetite change, described the data best and resulted in four distinct depressive subtypes: severe depression with anxiety (28.0%), moderate depression with anxiety (29.3%), moderate depression without anxiety (23.6%) and mild depression (19.0%). These classes showed corresponding clinical correlates at baseline and corresponding course and outcome indicators at follow-up (i.e., class severity was linked to lifetime mental disorders at baseline, and service use for mental health problems and current disability at follow-up).

Limitations

Although the sample was representative of the population on most parameters, the findings are not generalisable to the most severely affected depressed patients.

Conclusions

Depression could best be described in terms of both qualitative differences between symptom categories and quantitative differences in severity. In particular anxiety was a distinguishing feature within moderate depression. This study stresses the central position anxiety occupies in the concept of depression.

Introduction

Depression is a heterogeneous syndrome. Affected individuals vary markedly in their symptom profiles and response to treatment. Recognition of such heterogeneity in depression has long prompted researchers to identify meaningful and valid symptom-based subtypes to guide research on aetiology and to aid the development of more tailored intervention and treatment programmes (Baumeister and Parker, 2012). In the past decade, most researchers have used latent class analysis (LCA) to elucidate the heterogeneity in depressive symptoms.

This technique, often described as the categorical variant of factor analysis (FA), explains heterogeneity in symptomatology by assuming a number of discrete underlying groups (latent classes) of persons with similar symptom profiles. According to Clark et al. (2013) this has been the predominant view of psychopathology because classifying individuals into diagnostic categories or subtypes currently is a clinical necessity and required by insurance companies and other reporting agencies (Muthén, 2006). Alternatively, the underlying structure of depression is viewed as dimensional or continuous in nature with each individual having some amount of the disorder. This view has its counterpart in FA. Despite that both opposite views regarding the latent structure of depression still exist (Wright et al., 2013), increasing evidence shows that variations in depression are best viewed as qualitatively distinct subtypes as well as differences in severity along an underlying continuum. One solution to this debate is to use a hybrid model, which allows the underlying structure to be simultaneously categorical and dimensional. This paper addresses this issue.

The studies that have used LCA to identify subtypes of depression are summarised in Appendix A. We included studies among adults which employed a structured clinical interview to assess psychiatric symptoms and syndromes, and which utilised data from the general population or from a naturalistic cohort of current or remitted depressed patients and healthy controls. Purely clinical studies describing depressive subtypes among patients in treatment were excluded (such as described by Van Loo et al. (2012)), because of the risk of selection bias (i.e. many people with depression do not seek specialised treatment as a result of which clinical studies might only capture a limited range of depression severity and symptom categories). The studies described in Appendix A were performed in the period 1989–2015, the majority in the past 5 years. Analyses were based on data from 12 different surveys, mainly performed in North America and Western Europe. Most studies contained a subsample of respondents with at least one key symptom of depression; the depression typology was based on symptoms reported during their worst lifetime episode. Some studies used the aggregated depressive symptoms groups (for instance appetite, weight, sleep and psychomotor disturbance), while most researchers acknowledge that a differentiation of opposite depressive symptoms – for example appetite increase versus appetite decrease; weight gain versus weight loss – can help identify certain subtypes of depression, in particular melancholic and atypical depression.

Almost all studies found evidence for a depression typology based on symptom categories (e.g. typical/melancholic/cognitive and atypical/psychosomatic depression) and symptom severity within these categories (e.g. mild and severe depression). The studies which restricted their analyses to a subsample of respondents with a diagnosis of major or minor depression (Lamers et al., 2010, Lamers et al., 2012a, Lamers et al., 2012b; Major Depressive Disorder Working Group of the Psychiatric GWAS Consortium, 2013) tended to find the least number of classes: typical, atypical and moderate. The studies not confined to a subsample with a depressive symptom or diagnosis (Chen et al., 2000, Eaton et al., 1989, Kendler et al., 1996, Mezuk and Kendler, 2012, Moreno and Andrade, 2010) found a larger number of classes. Here, subthreshold subtypes could be identified as well. The studies which based their analyses on a subsample of respondents with at least one key symptom of depression (Alexandrino-Silva et al., 2013, Carragher et al., 2009, Lee et al., 2014, Prisciandaro and Roberts, 2009, Rodgers et al., 2013, Rodgers et al., 2014, Sullivan et al., 1998, Sullivan et al., 2002) generally found a number of in-between classes: typical, atypical, moderate and mild. Differences in study sample, design, diagnostic instrument and statistical method used thus hinder a refined comparison of the qualitative distinctions of the previously found subtypes.

The LCAs used in the studies portrayed so far are not equipped to describe differences in severity within subtypes. Studies that investigated the factor structure of depression found evidence for at least one dimension. Prisciandaro and Roberts (2009) found a two-factor model consisting of a ‘cognitive–affective’ and ‘somatic’ dimension, based on population data. However, in an earlier study they concluded that depression was best characterised by one dimension (Prisciandaro and Roberts, 2005). A similar result was reported by Sunderland et al. (2013), based on a treatment seeking clinical population, who found a two-factor solution with a high correlation (0.89) between the two latent factors representing ‘psychological symptoms’ and ‘somatic symptoms’. However, a systematic review of the available clinical studies on the factor structure of depression (Van Loo et al., 2012) revealed massive heterogeneity between the studies. Factor solutions differed both in numbers (mostly one-four factors) and content. Only to some degree, the analyses suggested the presence of a general factor representing both core symptoms of depression and some varying additional factors. Whereas FAs are equipped to describe differences in severity along one or more underlying continua, they are not made for describing qualitatively distinct subtypes. In the current study we will explore whether factor mixture modelling (FMM), a relatively new and promising technique, which combines aspects of LCA and FA (Lubke and Muthén, 2005, Clark et al., 2013), is more suitable to describe both differences in symptom categories and in depression severity in the general population. FMM has already been shown to be successful in describing the underlying psychopathology of other mental disorders, such as attention-deficit/hyperactivity disorder (Lubke et al., 2007), panic disorder (Roberson-Nay and Kendler, 2011) and alcohol use disorder (Jackson et al., 2014), and of major depression in a treatment seeking clinical population (Sunderland et al., 2013).

Some studies described in Appendix A that tried to validate the distinct subtypes of depression by examining their demographic and clinical correlates found indications for differential, potential aetiological mechanisms. Atypical depression which is characterised by increased appetite and weight gain, for example, appears to have its origins in inflammatory and metabolic abnormalities (Lamers et al., 2010, Lamers et al., 2013, Hickman et al., 2014, Rudolf et al., 2014). While most studies were confined to the identification and validation of distinct subtypes of depression, to our knowledge, few, if any, have also evaluated the usefulness of this knowledge in predicting the course and outcome of depression. We aim to do both by analysing data from the first two waves of the Netherlands Mental Health Survey and Incidence Study-2 (NEMESIS-2), a nationally representative survey of the general population aged 18–64 years. We created a subsample of respondents who had a lifetime key symptom of depression at baseline and who participated in both waves.

To summarise, our aim was to replicate and expand existing knowledge by studying subtypes of depression and their association with potential correlates, course and outcome indicators. First, three latent variable techniques (LCA, FA and FMM) were applied to identify subtypes of depression. Subtyping was based on disaggregated depressive symptoms and some additional symptoms – such as irritability and anxiety which are assumed to be important for differentiating depressive subtypes (Van Loo et al., 2012) – reported during the worst lifetime depressive episode. Second, after having determined which technique best described the data, the corresponding typology was described by reference to covariates cited in the literature (e.g. demographic characteristics, comorbid mental and somatic disorders, clinical characteristics of major depressive disorder, and familial vulnerability to mental illness), in order to explore potential mechanisms underlying the typology. Third, by using data from the second wave, we investigated differences in clinical course and outcome (i.e. disability and service use) between the various subtypes.

Section snippets

Methods

NEMESIS-2 is a psychiatric epidemiological cohort study of the Dutch general population aged 18–64 years. It is based on a multistage, stratified random sampling of households, with one respondent randomly selected in each household.

In the first wave (T0), performed from November 2007 to July 2009, a total of 6646 persons were interviewed (response rate 65.1%; average interview duration: 95 min). This sample was nationally representative, although younger subjects were somewhat underrepresented (

Best fitting model

The best fitting model, based on the BIC-values, entropy and interpretability, in a LCA with 13 depressive symptoms was a 4-class model. Although this model did not add much to the 3-class model (the LMR-LRT was not significant), it was conceptually more meaningful.

A series of EFAs with the same depressive symptoms were conducted estimating 1–3 factor models. As the 2-factor and 3-factor solutions entailed a useless factor with very few (≤3) items loading on it (that is, appetite and weight),

Discussion

This is the first study which used different latent variable techniques (LCA, FA, FMM) to identify the best symptom-based subtyping of depression and to subsequently assess its value in predicting course and outcome of depression. The best-fitting model was a flexible LCA with four classes, in which we were able to account for the conditional dependence within classes in a more parsimonious way than in FMM. The classes varied both in symptom category and in severity: severe depression with

Research findings

It is difficult to make direct comparisons between depressive subtypes found in this and other studies, due to differences in study sample, design, diagnostic instrument and, in particular, statistical method used. Earlier studies often used LCA to describe depressive subtypes (see Appendix A for an overview of findings). Almost all of these studies found evidence for a typology of depression based on differences in symptom categories (e.g. typical versus atypical), and within these categories

Limitations

A limitation of this study is the conditional branching inherent in the CIDI which may have led to an underestimation of certain (i.e. atypical) symptoms. Skip rules were used in the interview for questions assessing changes in appetite, weight, sleep and psychomotor disturbance. Accordingly, if one symptom was present (for example, decreased appetite), the question to assess its reverse (increased appetite) was not administered. However, evidence exists that opposite depressive symptoms can be

Acknowledgements

NEMESIS-2 is conducted by the Netherlands Institute of Mental Health and Addiction (Trimbos Institute) in Utrecht. Financial support has been received from the Ministry of Health, Welfare and Sport, with supplementary support from the Netherlands Organization for Health Research and Development (ZonMw) and the Genetic Risk and Outcome of Psychosis (GROUP) investigators.

FL is supported by a FP7-Marie Curie CIG (PCIG12-GA-2012-334065).

References (44)

  • S. Rudolf et al.

    Elevated IL-6 levels in patients with atypical depression but not in patients with typical depression

    Psychiatry Res.

    (2014)
  • P.F. Sullivan et al.

    The subtypes of major depression in a twin registry

    J. Affect. Disord.

    (2002)
  • M. Sunderland et al.

    Factor mixture analysis of DSM-IV symptoms of major depression in a treatment seeking clinical population

    Compr. Psychiatry

    (2013)
  • R. Armitage

    Sleep and circadian rhythms in mood disorders

    J. Affect. Disord.

    (2007)
  • Asparouhov, T., Muthen, B., 2011. Using Bayesian priors for more flexible latent class analysis. In: Proceedings of the...
  • H. Baumeister et al.

    Meta-review of depressive subtyping models

    J. Affect. Disord.

    (2012)
  • S.L. Clark et al.

    Models and strategies for factor mixture analysis: an example concerning the structure underlying psychological disorders

    Struct. Equ. Model.

    (2013)
  • W.W. Eaton et al.

    DSM-III major depressive disorder in the community. A latent class analysis of data from the NIMH Epidemiologic Catchment Area Programme

    Br. J. Psychiatry

    (1989)
  • R. de Graaf et al.

    The Netherlands Mental Health Survey and Incidence Study-2 (NEMESIS-2): design and methods

    Int. J. Methods Psychiatr. Res.

    (2010)
  • J.M. Haro et al.

    Concordance of the Composite International Diagnostic Interview Version 3.0 (CIDI 3.0) with standardized clinical assessments in the WHO World Mental Health Surveys

    Int. J. Methods Psychiatr. Res.

    (2006)
  • R.J. Hickman et al.

    C-reactive protein is elevated in atypical but not nonatypical depression: data from the National Health and Nutrition Examination Survey (NHANES)

    J. Behav. Med.

    (2014)
  • Y. Huang et al.

    DSM-IV personality disorders in the WHO World Mental Health Surveys

    Br. J. Psychiatry

    (2009)
  • Cited by (0)

    View full text