Original ArticlesToward a generalizable model of symptoms in major depressive disorder☆
Introduction
Lively debate about the nosology of clinical depression continues even in the face of substantial progress toward consensus regarding diagnostic criteria (Snaith 1993). Unanswered questions include: Which symptoms are specific to clinical depression Dohrenwend et al 1980, Zimmerman et al 1986, Feldman 1993? Are there reliably identifiable subtypes, still clearly suffering from depression, but with a symptom profile distinct from other clinically depressed individuals Nelson and Charney 1981, Young et al 1986, Startup et al 1992, Friedman et al 1963, Fleck et al 1995? Can pretreatment symptoms be used to select optimal treatments (prescriptive indicators), to monitor treatment efficacy, or to plan long-term management for prevention of relapse Raskin et al 1970, Raskin 1976, Rush et al 1981, Katz et al 1984, Feighner et al 1983, Nelson et al 1984, O’Brien and Glaudin 1988, van Praag 1992? Are there associations between clinically observed symptoms and biological abnormalities in depressed subjects Katz et al 1982, Costello 1970?
Answering these questions requires measures at the molecular level (i.e., symptom profiles), rather than at the molar level (i.e., total or global rating of severity) Costello 1992, van Praag 1992, Pichot 1972. A critical prerequisite for progress in this area is consensus on a) the array of symptoms that comprise the depressive syndrome and b) the use of adequate measures of these symptoms. Factor analytic studies of depression symptom rating scales, which might be expected to help achieve consensus on the symptoms of depression, have instead produced discrepant results.
These problems are illustrated in studies of the two most widely used symptom-rating instruments in clinical psychiatric research: the clinician-rated Hamilton Depression Rating Scale (HDRS) Hamilton 1960, Hamilton 1967 and the patient-rated Beck Depression Inventory (BDI) Beck et al 1961, Beck et al 1979. For example, factor analyses of the various versions of the HDRS (Grundy et al 1994) in varying samples have obtained varying numbers of factors, from one to seven (e.g., Cleary and Guy 1977, Steer et al 1987, Weckowicz et al 1971, Gibbons et al 1993, Hamilton 1967, O’Brien and Glaudin 1988, Rhoades and Overall 1983). Also, whereas Beck and colleagues argue that the BDI contains three intercorrelated factors (Beck et al 1988), published results show a range of one to seven factors (e.g., Berndt 1979, Cropley and Weckowicz 1966, Foelker GA et al 1987, Langevin and Stancer 1979, Lips and Ng 1986, Clark et al 1983, Steer et al 1987, Teri 1982, Weckowicz et al 1967, Zemore and Eames 1979, Feldman 1993).
Inconsistencies between factor analytic studies appear to be attributable to three problems: a) inconsistent content coverage between instruments, b) inadequate or inconsistent factor analytic procedures (including data sets that are not suitable for factor analysis), and c) variation in the subject populations studied.
The versions of the BDI and HDRS used in this study each measure eight of the nine criterion symptoms for major depressive disorder (MDD) specified in DSM-III-R (American Psychiatric Association 1987) and DSM-IV (American Psychiatric Association 1994), but they differ in which symptom is omitted, in the weight (i.e., relative number of items) given to each criterion symptom, and in the noncriterion symptoms included. On the 17-item version of the HDRS, concentration is omitted, whereas the BDI omits agitation/retardation. The HDRS includes four symptoms not in the DSM criteria for adult MDD: anxious/irritable (item 10), somatic anxiety (item 11), hypochondriasis (item 15), and loss of insight (item 16). Three noncriterion symptoms are included in the BDI: future pessimism (item 2), irritability (item 11), and worries about health (item 20). It bears noting that both the HDRS and BDI cover irritability and health worries even though these are not criterion symptoms in adults (by DSM).
Three items are required to uniquely define a factor. Therefore, symptom weighting has a direct bearing on the factor structure theoretically observable in a given instrument. For example, the HDRS contains one item rating the worthlessness or guilt symptom, which is not sufficient to define a guilt factor, whereas the BDI contains seven items, sufficient to define two such factors. In contrast, the HDRS includes three insomnia items, whereas the BDI contains only one, making it theoretically possible to identify a sleep disturbance factor in the former but not in the latter.
Although much of the factor analytic literature has focused on the problem of how many factors comprise a given instrument, a considerably more important question is how many factors comprise the universe of depression-related symptoms. Analysis of a single instrument, particularly one known to have gaps in symptom coverage, will not answer this question. Also, given the need for a sufficient number of items to uniquely define each factor, a factor analytic study ideally will include several instruments, each spanning all or parts of the hypothesized symptom universe Cattell 1978, Comrey 1988, Carroll 1985, Boyle 1985. In this situation, factors that do not appear on any single instrument are likely to appear (e.g., Jacobson et al 1978). This consideration led us to include several depression symptom-rating instruments in this study.
Two methodological problems underlie the continuing uncertainties regarding the factor structure of the HDRS and the BDI. The first is that many of the analyses have been inadequate from a technical standpoint. Boyle (1985) discussed this problem and provided excellent recommendations for practice, so this topic will not be covered here.
The second is more a conceptual than a technical issue: should symptom-rating scales be considered (and scored as) unidimensional measures of depression severity, or as (multifactor) profiles of depressive syndrome symptomatology? There is substantial evidence to suggest that the BDI and HDRS are multifactor inventories, yet they continue to be scored as though each measures a homogeneous construct of severity. One resolution is to retain in the instrument only those items that comprise a unifactorial score Gibbons et al 1985, Gibbons et al 1993, Clark et al 1983, Maier and Philipp 1985, Bech et al 1981. An alternative is to use the multifactor results as a basis for multiple subscale scores on the instrument (e.g., the Profile of Mood States, POMS) (McNair et al 1971).
An approach that combines the above viewpoints is to investigate whether there is a hierarchical structure in the symptom ratings Katz et al 1984, Tanaka and Huba 1984, Byrne and Baron 1993, Byrne et al 1995, McConville and Cooper 1992. That is, some or all of the multiple primary factors (symptoms) may be intercorrelated, with a higher-order factor that accounts for these correlations Feldman 1993, Gotlieb 1984, Dohrenwend et al 1980. An exploratory study of this possibility involves oblique factor rotation followed by extraction of any second-order factor(s) needed to account for the primary-factor intercorrelations. If a hierarchical structure appears to exist, the symptoms or items most associated with the higher-order factor are likely to define a general depression severity scale, while the residual variance in the multiple first-order factors defines a profile of independent symptom scores. This hierarchical approach was adopted in this study. Exploratory methods were used because the goal of this study is to explore the structure of correlations in these data, rather than confirm an a priori hypothesis about the structure.
Samples with varying distributions of diagnoses, including various kinds of coexisting conditions, may account for discrepancies across studies Mendels et al 1972, Feldman 1993, Ramos-Brieva and Cordero-Villafafila 1988, Startup et al 1992. The distribution of ratings is likely to differ between depressed patients with coexisting conditions and those without (Fleck et al 1995). This may bias correlations from what they might be in a single diagnostic population and also may limit the external validity of the results, since these would apply only to a population with a similar mix of concurrent diagnoses. Also, the presence of medicated patients may suppress a correlation because the medication limits the variation in an abnormality (e.g., insomnia). These considerations led us to include only medication-free patients with MDD and no Axis I coexisting diagnoses.
An accumulating body of research raises concerns about the psychometric characteristics of the HDRS and BDI Hedlund and Vieweg 1979, Bech 1981, Clark et al 1983, Boyle 1985, Maier and Philipp 1985, Gibbons et al 1993, Snaith 1993, Grundy et al 1994. In addition to concerns about symptom coverage Moran and Lambert 1983, Vredenburg et al 1985, Clark et al 1983, criticisms focus on the use of a single total score in apparently multifactor instruments Raskin 1976, Bech 1981 and the poor psychometric performance of some items Gibbons et al 1985, Rehm and O’Hara 1985.
Possible alternatives to the HDRS and BDI for clinical and research purposes are the parallel clinician (IDS-C) and self-report (IDS-SR) forms of the Inventory of Depressive Symptomatology (Rush et al 1986, 1996). The IDS consists of 30 items that span all of the DSM-III-R and DSM-IV criterion symptoms and the symptoms for DSM-IV melancholic subtype, for Research Diagnostic Criteria (RDC) (Spitzer et al 1978) endogenous subtype, and for the atypical subtype (American Psychiatric Association 1994). The IDS also incorporates most of the noncriterion items on the BDI and/or HDRS. Item wording of the IDS-C and IDS-SR differs only in the point of view of the rater. When factor-analyzed alone, each IDS form yields three factors, the first two of which are similar across factors (cognition/mood, anxiety/arousal), plus a third factor that is inconsistent (physiological dysregulation on the IDS-C, sleep disturbance on the IDS-SR).
This study has two primary goals: 1) to establish a generalizable model of the symptoms observed in outpatients with MDD, and 2) to compare symptom coverage of the IDS to that of the HDRS and BDI. If the IDS provides a more complete profile and overlaps with the HDRS and BDI, this instrument might serve as an alternative to the HDRS and BDI.
Section snippets
Subjects
Cases were drawn from the data set of 552 subjects with current or remitted MDD whose data were reported in Rush et al (1996). In addition to the criteria used to select subjects for Rush et al (1996)—absence of coexisting Axis I diagnoses, good medical health, free of psychotropic medication for at least 2 weeks, all four instruments completed within 2 days of each other—subjects also had to have complete ratings for all items on all four instruments.
These data were collected as part of a
Subjects
Of the original 552 included in Rush et al (1996), 353 had current or remitted MDD. Of these, 324 had complete item ratings on the 88 items actually used in the analysis (see below). The characteristics of the final sample are summarized in Table 1. Fourteen patients had remitted MDD. In addition, 13 patients had both dysthymic disorder and MDD.
Adequacy of the data set
All but two of the 94 items had moderate (.3) to high (.9) correlations with at least one item; the two exceptions were HDRS item 16 (loss of insight)
Correspondence between factors and symptoms
The array of item content in the four instruments we analyzed could theoretically yield factors that correspond to the nine criterion symptoms for MDD as listed in DSM-IV. In fact, the factors obtained correspond to many but not all of the DSM-IV criterion symptoms. Conversely, some factors identified clusters of intercorrelated symptoms that are not currently considered criterion symptoms. In addition, items related to other symptoms were either divided among several factors or aggregated into
Acknowledgements
We wish to thank M. Basco, R. Jarrett, and M. Trivedi for access to data on research patients they recruited and evaluated, as well as the team of clinicians who provided symptom ratings under their direction. Also, thanks to David Savage for his expert typing and editing assistance. We are grateful to Kenneth Z. Altshuler, MD, Stanton Sharp Distinguished Chair, Professor, and Chairman of the Department of Psychiatry for his administrative support.
References (106)
- et al.
Psychometric properties of the Beck Depression InventoryTwenty-five years of evaluation
Clin Psychol Rev
(1988) - et al.
The Beck Depression InventoryTesting and cross-validating a second-order factorial structure for Swedish nonclinical adolescents
Behav Res Ther
(1995) - et al.
Are diagnostic criteria, time of episode and occupational impairment important determinants of the female:male ratio for major depression?
J Affect Disord
(1994) - et al.
Application of modern psychometric theory in psychiatric research
J Psychiatr Res
(1985) - et al.
Exactly what does the Hamilton Depression Rating Scale measure?
J Psychiatr Res
(1993) - et al.
The structure of moods
Pers Indiv Diff
(1992) - et al.
Use of the Hamilton Rating Scale for classification of depressive disorders
Compr Psychiatry
(1982) - et al.
A new validation of the Hamilton Rating Scale for Depression
J Psychiatr Res
(1988) - et al.
Item characteristics of the Hamilton Rating Scale for Depression
J Psychiatr Res
(1985) - et al.
Differential effects of cognitive therapy and pharmacotherapy on depressive symptoms
J Affect Disord
(1981)
Diagnostic and Statistical Manual of Mental Disorders
Diagnostic and Statistical Manual of Mental Disorders
Period analysis of sleep onset in depressed outpatients and normal controls
Rating scales for affective disordersTheir validity and consistency
Acta Psychiatr Scand
The Hamilton Depression ScaleEvaluation of objectivity using logistic models
Acta Psychiatr Scand
Quantitative rating of depressive states—Correlation between clinical assessment, Beck’s self-rating scale and Hamilton’s objective rating scale
Acta Psychiatr Scand
Cognitive approaches to depression and suicide
Hopelessness and suicidal behavior. An overview
JAMA
Cognitive Therapy of Depression
An inventory for measuring depression
Arch Gen Psychiatry
Taking items out of contextDimensional shifts with the short form of the Beck Depression Inventory
Psychol Rep
Recognising psychiatric symptomsRelevance to the diagnostic process
Br J Psychiatry
Self-report measures of depressionSome psychometric considerations
Br J Clin Psychol
The Beck Depression InventoryTesting and cross-validating a hierarchical factor structure for nonclinical adolescents
Meas Eval Counsel Devel
Depression rating scales
Arch Gen Psychiatry
Exploratory factor analysisA tutorial
Human Cognitive Abilities
On methodology in the study of cognitive abilities
Multivar Behav Res
The scree test for the number of factors
Multivar Behav Res
The Scientific Use of Factor Analysis in Behavioral and Life Sciences
The core symptoms of depression in medical and psychiatric patients
J Nerv Ment Dis
Factor analysis of the Hamilton Depression Scale
Drugs Exp Clin Res
Factor-analytic methods of scale development in personality and clinical psychology
J Consult Clin Psychol
The structure of depression found in the general population
Psychol Med
Classification and psychopathology
Research on symptoms versus research on syndromes—Arguments in favour of allocating more research time to the study of symptoms
Br J Psychiatry
The dimensionality of clinical depression
Aust J Psychol
Nonspecific psychological distress and other dimensions of psychopathology
Arch Gen Psychiatry
Comparison of alprazolam, imipramine, and placebo in the treatment of depression
JAMA
Distinguishing depression and anxiety in self-reportEvidence from confirmatory factor analysis on nonclinical and clinical samples
J Consult Clin Psychol
Factorial structure of the 17-item Hamilton Depression Rating Scale
Acta Psychiatr Scand
Confirmatory factor analysis of the short form Beck Depression Inventory in elderly community samples
J Clin Psychol
Syndromes and themes of psychotic depressionA factor analysis
Arch Gen Psychiatry
Common factor analysis versus component analysisSome well and little known facts
Multivar Behav Res
Depression and general psychopathology in university students
J Abnorm Psychol
A psychometric investigation of the standard and short form Beck Depression Inventory
Psychol Rep
The Hamilton Rating Scale for Depression: One scale or many?
Clin Psychol Sci Pract
A rating scale for depression
J Neurol Neurosurg Psychiatry
Development of a rating scale for primary depressive illness
Br J Soc Clin Psychol
Cited by (80)
Sex differences in lever pressing and running wheel tasks of effort-based choice behavior in rats: Suppression of high effort activity by the serotonin transport inhibitor fluoxetine
2021, Pharmacology Biochemistry and BehaviorCitation Excerpt :Major depressive disorder often is marked by motivational and psychomotor symptoms such as motor slowing, fatigue, lassitude, loss of energy, and effort-related deficits (Stahl, 2002; Demyttenaere et al., 2005; Salamone, 2006; Treadway and Zald, 2011; Treadway et al., 2012; Fava et al., 2014). The severity of such motivational symptoms of depression is correlated with problems in social function, employment, treatment outcomes (Tylee, 1999; Stahl, 2002), and overall depression severity (Gullion and Rush, 1998). There is evidence of a link between exertion of effort, physical activity, and depression in both animal and human studies (Todder et al., 2009; Bardi et al., 2012).
Evaluating the evidence for biotypes of depression: Methodological replication and extension of Drysdale et al. (2017)
2019, NeuroImage: ClinicalCitation Excerpt :Both instruments are widely used and represent gold-standard rating scales for depression in clinical research (Cusin et al., 2009). Responses to the two different instruments are highly correlated, and their specific items are related to mostly overlapping symptoms (Rush et al., 1996; Gullion and Rush, 1998). In general, replication of effects using different instruments provides evidence that the found association is characteristic of the pathology rather than the specific instrument.
Partial reversal of the effort-related motivational effects of tetrabenazine with the MAO-B inhibitor deprenyl (selegiline): Implications for treating motivational dysfunctions
2018, Pharmacology Biochemistry and BehaviorThe 52 symptoms of major depression: Lack of content overlap among seven common depression scales
2017, Journal of Affective DisordersCitation Excerpt :A way forward for now is the assessment of a broad range of symptoms; the 44-item Symptoms of Depression Questionnaire provides a good example and aims to capture the full range of clinical presentations of depression (Pedrelli et al., 2014). Roughly 2 decades ago, Gullion and Rush (1998) (p. 959) noted that “a critical prerequisite for progress […] is consensus on a) the array of symptoms that comprise the depressive syndrome and b) the use of adequate measures of these symptoms”. We are nowhere near to agreeing which symptoms are good depression symptoms, or which scales are good depression scales.
- ☆
This research was supported in part by a Mental Health Clinical Research Center Grant (MH-41115) from NIMH to the University of Texas Southwestern Medical Center.