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

Abstract

Background

There is a paucity of empirical studies examining the latent structure of depression symptoms within clinical populations.

Objective

The current study aimed to evaluate the latent structure of DSM-IV major depression utilising dimensional, categorical, and hybrid models of dimensional and categorical latent variables in a large treatment-seeking population.

Methods

Latent class models, latent factor models, and factor mixture models were fit to data from 1165 patients currently undergoing online treatment for depression.

Results

Model fit statistics indicated that a two-factor model fit the data the best when compared to a one-factor model, latent class models, and factor mixture models.

Conclusions

The current study suggests that the structure of depression consists of two underlying dimensions of depression severity when compared to categorical or a mixture of both categorical and dimensional structures. For clinical samples, the two latent factors represent psychological and somatic symptoms.

Introduction

The extant psychiatric nosologies (Diagnostic and Statistical Manual of Mental Disorders [DSM-IV] and International Classification of Diseases [ICD-10]) are based on a categorical framework, wherein mental disorders are defined as discrete entities, with clear boundaries that demarcate normality from pathology. In recent years, however, a burgeoning empirical literature has accumulated indicating that significant improvements in the validity and utility of psychiatric assessment and diagnosis could be achieved by incorporating a dimensional–spectrum conceptualisation [1], [2], [3], [4], [5]. This concept assumes that disorders are best represented by underlying dimensions or traits, whereby different individuals can vary along the underlying dimension depending on their level of severity of depression. There are no natural cut-offs used to describe someone as depressed or not, but instead “fuzzy” boundaries exist that describe who should receive and who could benefit from clinical attention (which could be through the use of a variety of proven effective treatments). Furthermore the level of attention and the specific type of treatment that is required can be determined by examining the degree of severity experienced by each individual.

Despite the recent literature advocating that dimensional approaches be applied to psychopathology, there are several notable criticisms with this approach. Primarily, there are concerns that dimensional approaches may be perceived as cumbersome and complex, and have less clinical utility and efficiency. These concerns are reflected by the argument that dimensional methods may hamper diagnostic decision making, disrupt communication between clinicians, researchers, and patients, and result in poor diagnostic reliability [6]. The primary use of diagnostic categories can therefore be seen as a mechanism for clinicians, health policy makers, health planners, and researchers to simplify an often complex clinical presentation and facilitate decision making. Consequently, as a means of improving the use of diagnostic categories, a primary goal of any classification system should be to focus on validating the existing diagnostic categories and examining the presence of more relevant categories that better describe how these disorders occur in nature.

To inform the debate regarding the relative validity and appropriateness of dimensional versus categorical conceptualisations of psychopathology, latent variable modelling can be utilised to test and compare competing models [7]. Of particular relevance to the current study, the structure of depression has been examined in a number of general population surveys using a variety of latent variable modelling techniques (e.g., taxometrics) [8], [9], [10]. These studies have sought to statistically examine whether depressive symptom data are better modelled using dimensional or categorical approaches. The consensus amongst the epidemiological literature to date is that depressive symptoms are best represented as dimensional constructs, with statistical fit statistics indicating superior fit of dimensional models in comparison to categorical models [11].

There is a paucity of evidence examining the latent structure of depression symptoms within clinical populations. It is critical to examine the latent structure of depression in treatment settings given that significant differences exist between general population and clinical samples. For instance, individuals in clinical samples are likely to display greater symptom severity as well as, elevated levels of comorbidity and impairment than those in the community [12]. Furthermore, there is little evidence examining hybrid models that combine both categorical and dimensional concepts using factor-mixture models [13]. These models attempt to find a bridge between both approaches by assuming that homogenous categories or classes of depressive individuals exist but, within each class, individuals can also vary along a latent dimension indicating differences in terms of severity [14]. This way, reliability and clinical communication are maintained through the use of descriptive categories while diagnostic information and variability within classes are maintained through the use of dimensions. Since revisions for DSM-5 and ICD-11 are aimed at improving the clinical validity and utility of the diagnostic criteria, particularly with respect to incorporating more dimensional approaches of psychopathology within the existing categorical framework, this research is especially timely [15], [16]. The aim of this paper was to examine alternative measurement models reflecting different conceptualisations of DSM-IV depression symptom structure in a clinical population. In particular, we compared dimensional (factor analysis), categorical (latent class analysis), and hybrid models reflecting a combination of continuous and categorical latent variables (factor mixture model). To the best of our knowledge, this is the first study of its kind to evaluate these models in a clinical sample of depressed patients.