Sub-Grouping Non-Melancholic Major Depression Using Both Clinical and Aetiological Features

Abstract

Objective: In previous papers we have considered the extent to which two contrasting analytic approaches, examining reported clinical symptom variables alone and aetiological variables alone, assist definition of subgroups of non-melancholic major depression. Here, we address the same objective but combine both sets of variables, and contrast the combined solution with each of the contributing ones. Method: We study a sample of 185 subjects with a putative non-melancholic major depressive disorder, with analyses involving 13 aetiological and 38 symptom variables.

Results: A four-class subgrouping was derived by use of a cluster analytic technique, with ‘neurotic depression’, non-anxious ‘depressed’, ‘situational’ and ‘residual’ groups. The largest group comprised ‘neurotic depression’ subjects, with characteristics compatible with a spectrum disorder encompassing both clinical features as well as an underlying temperament and personality style marked by anxiety. Conclusions: Comparative advantages and properties of the three differing analytic approaches to defining ‘meaningful’ non-melancholic major depressive subgroupings are considered. As a ‘neurotic depressive’ class has been consistently identified across those three approaches, but with quite varying numbers of subjects circumscribed, it is clearly a ‘fuzzy’ entity which may benefit from a dimensional approach to its measurement. As many of the non-melancholic groupings appear secondary to a substantive predisposing factor such as anxiety or disordered personality functioning, the clinical importance and treatment utility in identifying and circumscribing such classes are clearly supported.

Currently, non-melancholic depression is subdivided in classificatory systems (e.g. DSM-IV, ICD-10) [1, [2]] on the basis of severity and duration criteria. Previously, subdivision was on the basis of broad aetiological variables, with items such as ‘neurotic’ and ‘reactive’ depression weighting contributions of personality style and life event stressors. Again, there have been attempts to subdivide nonmelancholic depression on the basis of the dominant clinical aspect, with Roth and Barnes [3] finding support for ‘anxious’ and ‘hostile’ depressive subgroups. All of these approaches presumably reflect the failure to identify distinct syndromes on the basis of clinical features only [4].

In attempting to develop a clinically useful system for subtyping non-melancholic depression, we have undertaken two contrasting analytic approaches. The ‘top down’ one hypothesised [5] an a priori aetiological model, with subgroups reflecting the relative dominance of personality style, disordered personality functioning and anxiety (as premorbid variables) and precipitating life event stressors. A latent class analysis assessing such variables generated a four-class solution. As life-event stressors had similar item probabilities across all four classes, we found no support in our sample for a ‘reactive’ or ‘situational’ depression class. However, we did identify: (i) an ‘anxious worrier’ class with anxiety reflected at the personality/temperament level (and with Cluster C personality style characteristics over-represented); (ii) a non-anxious class where members tended to have an unstable and volatile personality (or Cluster B style); (iii) a class where members had a high rate of lifetime anxiety disorders but lacked a distinct anxious personality style; and (iv) a residual class.

In addition to that approach, we also undertook [6] a ‘bottom up’ analytic approach where we examined currently reported clinical symptoms (of depression and anxiety) solely, and sought to determine if sub-types or syndromes could be identified by patterning of such ‘surface’ features only. A cluster analysis suggested four subgroups, with members of the first two clusters (i.e. labelled ‘anxiety’ and ‘irritability’) able to be viewed as having ‘spectrum conditions’, in that the manifest symptom pattern related to similar personality/temperament styles. We also identified a third subgroup where severity of depressed mood appeared the clearest defining feature, and a fourth ‘residual'subgroup. In this report, we study the same sample, but now analyse the combined set of those previously refined variables to determine if the combination of both the aetiological and clinical symptom sets provides a more coherent and clinically useful solution than those derived from one set only, and so allows any clearer definition of the contributing ‘spectrum conditions’.

Method

As the study design and measures have already been detailed in the Journal [7], we summarise details.

Subjects

The subjects were a subset of inpatients and outpatients of our tertiary referral Mood Disorders Unit (MDU) and outpatients of our research psychiatrists. All were required to have had a major depressive episode, as defined by DSM-III-R [8] criteria, and present for less than 2 years. The final sample involved 185 subjects diagnosed as having a nonmelancholic depression either by DSM-III-R criteria or by our MDU ‘clinical’ definitions [9].

Assessment of subjects

Patients completed a self-report questionnaire which addressed many historical issues but which also required the patient to rate exposure to a series of life-event stressors prior to the onset of their current episode and their impact. The patient completed two self-report measures of early experiences of parenting, the Parental Bonding Instrument (PBI) [10] and the Measure of Parenting Style (MOPS) [11]. A research psychologist then sought additional details on the current and previous episodes of depression, and generated a lifetime diagnosis of anxiety disorders using the computerised Composite International Diagnostic Interview [12]. The Costello and Comrey [13] trait anxiety measure was completed, together with the Beck Depression Inventory measuring self-reported depression severity [14]. Details were sought about depression, anxiety and alcoholism in any first-degree relative. The interviewing psychiatrist assessed more complex clinical features, obtained information on suicidal and self-injurious behaviours, completed the 17-item Hamilton [15] depression measure, and assigned one of the four MDU clinical diagnoses (i.e. psychotic, endogenous, neurotic or reactive depression). Additionally, the psychiatrist rated the severity of both acute and chronic life-events stressors according to DSM-III-R anchor points and also derived a global assessment of functioning (GAF) score.

The psychiatrist was required to make a number of judgements about personality style, with composite scale scores developed corresponding to the DSM-IV cluster model of the personality disorders (namely, ‘eccentric’, ‘dramatic’, ‘sensitive’), as well as an additional ‘anxious’ personality style. In addition to rating personality style, the psychiatrist was required to make a judgement as to whether there was evidence of ‘inadequate’ or ‘disordered’ personality functioning, using both clinical judgement and the more structured criteria generated by Millon [16], which assess disordered function across a range of parameters (e.g. ‘inflexible’) and domains (e.g. intimate relationships). These were aggregated into total ‘parameter’ and ‘domain’ scores on those measures. A clinical judgement of ‘interpersonal rejection sensitivity’ was also made. The psychiatrist undertook a developmental history, examining early markers of later anxiety such as behavioural inhibition and school phobia, and then judged lifetime ‘trait anxiety’ by asking the patient if, when not depressed, they generally viewed themselves as ‘nervy’, ‘a worrier’, ‘tense’ and ‘anxious’. In addition to the patient's self-report score for severity of antecedent life-event stressors, the psychiatrist generated a clinical estimate rating on a similar scale. Subsequently, consensus discussions between the psychiatrists considered a vignette of the patient presented by the interviewing psychiatrist (to judge the severity and context of life-event stressors experienced by the patient prior to the onset of the episode) and so generated a ‘consensus antecedent stressor’ score.

Statistical analyses

We analysed a data set of 51 variables refined in the two previous studies. The former study derived 13 aetiological variables, comprising: (i) four personality style scores (‘eccentric’, ‘dramatic’, ‘sensitive’ and ‘anxious’) scores, and with the first three corresponding to the DSM ‘cluster’ concept; (ii) two measures of dysfunctional personality (‘inadequate personality’, and disordered personality as quantified by the total ‘parameter’ score); (iii) four trait anxiety variables (nervy, a worrier, tense, and anxious); (iv) the lifetime presence of any anxiety disorder; as well as (v) two antecedent severity of stressor scores (as quantified by DSM-III-R criteria, and the consensus antecedent stressor rating). As noted, the same 38 refined manifest symptoms examined in the clinical feature study (and which had subsequently generated factors labelled ‘somatic anxiety’, ‘depressed mood’, ‘irritability’ and ‘anhedonia/fatigue) were added as individual variables to form the total item set.

Methods

The large number of variables discounted use of a latent class analysis (used in our study of aetiological variables), while the ratio of variables to subjects (nearly 1:4) suggested that a cluster analysis would be appropriate. Thus, in seeking to group individuals into classes we undertook a K-means cluster analysis of the 51 variables, with imposed three-class, four-class, five-class, and six-class solutions. In all solutions, there was a cluster which, when examined against residual variables in our data set, corresponded to the historical concept of ‘neurotic depression’ (as detailed shortly). Its principal contributing features (from the 51-item set) were high trait anxiety, an anxious worrying personality style, anxiety and fatigue symptoms, and irritability, and 96 (52%) were assigned to this cluster. As we sought to determine if any composite solution involving both aetiological and clinical variables was superior to each individual approach alone (and for which four classes had been derived in each of those analyses), we now focus on the four cluster solution. The other three clusters were labelled: (i) ‘depressed’ (with 27 subjects or 15% of the sample) due to high weighting from depressive, anhedonic and fatigue items; (ii) ‘situational’ (with 21 subjects or 11% of the sample) due to higher ratings of consensually judged antecedent life-event stressors; and (iii) ‘residual’ (with 41 subjects or 25%), with high loadings from several items indicative of low self-esteem and of high self-criticism.

When comparison was made between the subjects allocated to these four clusters with those allocated to differing classes in the separate analytic sets, the current allocation was more strongly correlated with the solution involving clinical variables alone than with the solution involving aetiological variables alone (levels of agreement were 68% and 61%, respectively).

As a ‘neurotic depressive’ group was most clearly identifiable across all three approaches (i.e. clinical variables alone, aetiological variables alone, and the composite set), albeit variably labelled, we examined the extent to which those allocated to that class by each individual approach were assigned to that cluster by the composite approach. Twenty-seven of the 30 (90%) assigned by aetiological variables to the ‘anxious worrier’ class, and 50 of the 55 (91%) assigned by the clinical variables to the ‘anxiety’ class were assigned to our current ‘neurotic depression’ cluster. However, the ‘class’ defined in our current analyses comprised far more subjects (n == 96) than the equivalent classes derived by the other two approaches (30 and 55, respectively), with inspection identifying that it subsumed significant percentages of those viewed as having an ‘irritable/hostile’ depression, as well as the anxious worriers or anxiety class subjects.

Properties of the composite four-cluster solution

While we provided preliminary labels for the groups identified in the current cluster analysis, analyses were then undertaken against remaining study variables to determine if the labels could be validated. Family histories of anxiety, depression and alcoholism were non-differentiating. Those in the ‘neurotic depression’ cluster tended to be more likely (i.e. 57% vs 44% to 29% for the other groups) to receive a clinical diagnosis of ‘neurotic’ (vs ‘reactive’) depression, while the alternative clinical diagnosis of ‘reactive depression’ tended to be commonest (i.e. 48% vs 37% to 33% for the other groups) for those in the ‘situational’ cluster.

Tables 1, 2 and 3 allow further profiling of cluster members. As noted, those assigned to the ‘neurotic depression’ cluster were the most clearly distinguishable. Their members (see Table 1) had the lowest mean age and (perhaps as a consequence of the younger age) were more likely to be single. They were the youngest at onset of their first depressive episode and they returned significantly higher Hamilton severity scores. They were most likely to have had a lifetime anxiety disorder, whether it preceded their initial depression or was concurrent with it. They scored significantly higher total scores on current anxiety symptoms, and had the highest rates of behavioural inhibition and school phobia (possible forerunners to later anxiety), as well as of school truancy. In terms of our trait anxiety ‘temperament’ variables, they had the highest rates in reporting themselves as generally (i.e. when not depressed) being nervy, worriers, tense and anxious, although this profile was not supported by Costello-Comrey trait anxiety scores.

Table 1.

Sociodemographic and diagnostic variables compared across composite ‘top down’ and ‘bottom up’ clusters

Table 2.

Parenting variables compared across composite ‘top down’ and ‘bottom up’ clusters

Table 3.

Personality and life-event stressors and other variables compared across composite ‘top down’ and ‘bottom up’ clusters

When early developmental family risk variables were examined, similar findings were generated for the MOPS as for the PBI measure, but with the former being slightly more distinct and therefore reported in Table 2. Thus, those in the ‘neurotic depression’ cluster consistently reported the highest levels of adverse parenting (significant for higher levels of maternal and paternal overprotection), and had the highest rates of exposure to a number of early adverse developmental factors (a mother who was ‘nervy/worrier’, a depressed mother, a mother providing dysfunctional parenting, a dysfunctional parental marriage, the parents being more likely to have separated, subject raised by a single parent and to have had multiple parenting figures). Table 3 data demonstrate that subjects demonstrated high rates of disordered personality functioning and were judged as high on interpersonal rejection sensitivity. In terms of DSM personality disorder ‘style’, this group received the highest Cluster C (most distinct), Cluster A and Cluster B scores, as well as returning significantly higher scores on the ‘anxious’ personality style. In response to stress, they had the highest rates of reporting having a ‘short fuse’. They had the highest rates of cigarette smoking, illicit drug use (while only marihuana data are reported, similar trends were evident for amphetamines, LSD, cocaine and heroin, with respective rates of 25%, 21%, 17% and 5%), and dependence on anxiolytic medication. Additionally, they had the highest rate of suicide attempts (both for the current episode and previously), while also having the highest current and past self-injury rates.

The second group was termed ‘situational depression’, with (as noted earlier) members having the highest diagnostic rates of ‘reactive depression’. However, this label is only weakly supported by analysis of life-event stress scores. Members did show a non-significant trend to report the highest number of total stressors in the 12 months preceding their depression and they received the highest consensus rating score for severity of antecedent stressors, but these trends were slight. What perhaps distinguishes this group best is the suggestion of minimal long-standing predispositional factors. Thus, they returned the lowest rates of exposure to adverse developmental factors (significantly lowest for total number of events), they were significantly unlikely to be judged as having a personality disturbance prior to their initial depressive episode, and unlikely to be rated as having an ‘inadequate personality’ or a style of interpersonal rejection sensitivity. In addition, they returned the significantly lowest levels of disordered function on the domain and parameter measures, and had the lowest scores on the three DSM clusters. They were least likely to have a lifetime or concurrent anxiety disorder and had the lowest rates of suicide and self-injurious attempts. In addition, they had the highest mean age, were most likely to be in an ongoing relationship, and had a negligible family history of depression. Thus, while the term ‘situational depression’ is used, this group is perhaps more noteworthy in showing little evidence of a predisposing at-risk style, while only tending to be more likely to receive a diagnosis of ‘reactive depression’ and return higher life-event stressor scores. Any formalised ratio of ‘predisposing life-event stress: predispositional style’ (not attempted here) would then be likely to view members of this cluster as being most likely to develop depression as a consequence of the life-event stress rather than as a consequence of any identified predisposing risk style.

The third (labelled ‘depressed’) group was not clearly distinguishable from other groups. In the cluster analysis, this group was defined most distinctly by having high scores on the depression, anhedonic and fatigue items. Members were not able to be distinguished from members of the other groups on sociodemographic variables. In terms of early adverse developmental experiences, they did show a non-significant trend to report PBI-defined paternal ‘affectionless control’ while they were also most likely to report their father as being ‘dysfunctional’ in their early years. There was a non-significant trend for them to report physical/verbal abuse and also to have the highest rate of sexual abuse by other than a parent in the first 16 years. On measures assessing personality dysfunction they returned high scores, akin to those returned by subjects in the ‘neurotic depression’ group. For all direct and indirect measures of anxiety, they returned lower scores than those in the ‘neurotic depression’ group. They may then represent a group most distinguishable (particularly from the ‘neurotic depression’ group) by having a ‘non-anxious depression’ in conjunction with a disordered personality style.

The fourth group is labelled a ‘residual’ one as group members could not be identified by any distinct feature in the cluster analysis nor across the list of corroborative variables, while any such analysis as used in this study will generate at least one residual group.

Discussion

Whether using only aetiological variables (in our ‘top down’ approach), manifest clinical symptoms of anxiety and depression (in our ‘bottom up’ approach), or the combination as assessed in this report, we have identified a ‘class’ or group of subjects that generates a very similar cross-sectional and longitudinal clinical ‘profile’. Such subjects appear most clearly distinguished by having high levels of trait and state ‘anxiety’. Thus, they rate much higher on anxiety symptoms during their depressive episode, are more likely to rate as high on trait anxiety characteristics (e.g. being a ‘worrier’), are most likely to meet diagnostic criteria for having had a lifetime anxiety disorder, are most likely to have become dependent on anxiolytic medication and are most likely to provide evidence of behavioural inhibition in childhood. They are rated clinically as high on disordered personality function, and are most likely to have made suicide attempts and to have injured themselves. While they return the highest scores on self-report (Beck) and clinician-rated (Hamilton) ‘depression’ measures, their total scores are likely to be elevated by affirming the anxiety items in those measures, so that their ‘depression’ per se may not necessarily be more severe.

Despite the similar clinical profile of this group across all three analytic approaches, each approach identified quite differing total numbers (30, 55 and 96, respectively), suggesting that it is probably best to view group membership as ‘fuzzy’ and perhaps attempt to define group members dimensionally, whether this is in terms of the extent (e.g. ‘very’, ‘moderately’, ‘slightly’ or ‘not at all’ conforming) to which they conform to some derived prototypic definition or by severity. Alternatively, conceptualising members as having a ‘spectrum disorder’ has some utility in further seeking to define and circumscribe a ‘class’. Cassano et al. [17] invokes such a term to encompass any continuum between core symptoms (i.e. axis I components) and associated features, when the latter can include behavioural patterns, temperament and personality (i.e. axis II features), as well as other features, including sequelae. T h e finding that members identify themselves as showing a defined temperament style of behavioural inhibition in their early years (effectively being ‘shy’) and that they meet lifetime criteria for an anxiety disorder (in this study, before their first depressive episode), suggests that there is early disorder expression, albeit manifested as shyness and anxiety. Their disordered personality functioning appears again related to their ‘anxious’ temperament/personality, and it is reasonable to assume that their higher rates of drug dependence, cigarette smoking, self-injury and suicide attempts are driven by their anxiety levels. The clinical picture is compatible with both previous concepts of ‘neurotic depression’ [18, [19]] that have involved a substantive anxiety component and consistent identification of an ‘anxiety’ subclass in nonmelancholic depression [5].

The clinical utility of defining such a ‘class’ is considerable, and two examples are offered. First, if a percentage of non-melancholic depressed patients have a primary anxiety disorder and/or an anxious temperament, then therapeutic intervention may better be directed at their ‘anxiety’ rather than necessarily at their ‘depression’. ‘Depression’ in such subjects is then (like other expressions of psychopathology such as drug dependence and self-injury) often no more than a secondary consequence of the anxiety, with therapeutic benefit emerging more from addressing the ‘primary’ anxiety. Second, it would be important to determine if members of this group are selectively responsive to certain interventions (in comparison to those in other ‘groups’), be they psychopharmacological (e.g. the selective serotonin re-uptake inhibitors) or non-physical (e.g. cognitive–behaviour therapy) treatments. Our capacity to generate any hierarchy of effective treatments across the whole heterogeneous non-melancholic depressive class is likely to remain limited but, more importantly, it lacks common sense. It is far more appropriate and important to circumscribe and identify one or more ‘meaningful’ classes and pursue treatment efficacy studies in those more homogeneous subsamples. Such studies may benefit from flexibility in defining ‘neurotic depressive’ group members (ranging from our ‘tight’ aetiologically based definition, which assigned few subjects, to the more ‘inclusive’ composite approach adopted here, where one-half of the sample were so assigned).

In this composite analysis, we failed to identify one ‘group’ identified in our earlier analyses, that is subjects distinguished by irritability, having a ‘short fuse’ when under stress, and having other markers of a volatile personality style (akin to the DSM Cluster B description). As they scored reasonably highly on anxiety measures, we postulated [6] that they may ‘externalise’ their anxiety in the form of irritability, and conform to the literature description of a ‘hostile’ non-melancholic class. Significant percentages (i.e. 62% and 59% of those, respectively, assigned to such a class in the aetiological and clinical feature-focused analyses) were subsumed in our current ‘neurotic depressive’ class. Thus, our composite analytic strategy identified a very broad ‘neurotic depressive’ class, subsuming two possible subclasses (effectively, the ‘anxious’ and ‘hostile, irritable’ subgroups) previously identified in our analyses of the separate variable sets. More detailed studies will be required to consider the validity and comparative advantages of those competing models, and again the implications such distinctions will have for treatment studies.

In our current analyses, we found some support for a non-anxious, non-melancholic depressive class, a subgroup also suggested in our analysis of clinical features alone but not so suggested in the aetiological-based analyses. Thus, this group (if it exists at all) would appear defined only by current clinical features. Those with a non-melancholic depression may then have a significant number of current anxiety symptoms or have a non-anxious depression, and with the former expression appearing related to an anxious temperament. Whether the latter is determined by absence of such a temperament style, by differing aetiological factors, by phase of illness or by other influences (e.g. our illness duration criterion excluding those with dysthymia) can only be speculated about at this time.

While we found weak support for a ‘situational’ or ‘reactive’ depressive subgroup in this study, it is important to note that it was not suggested on the basis of any more distinct or severe influence of antecedent life event stressors. Instead, we have argued for such a construct on the basis of such subjects not having any substantive predisposing variables (e.g. anxiety, disordered personality functioning), so that life-event stressors were the default operative component in a model that hypothesised only a certain number of contributing factors. It is then equally possible that group membership was defined by some other variable that we have failed to identify.

Conclusions and implications

The present study was undertaken to determine whether a composite analysis of both aetiological and clinical variables provided a more meaningful identification of non-melancholic depressive subgroups than either component alone. Results fail to provide a clear-cut answer, as the composite approach generates a much more encompassing subgroup of those with a ‘neurotic depressive’ disorder, effectively subsuming two subgroups of ‘anxious’ and ‘hostile, irritable’ depressive expressions that were suggested in the individual analytic approaches. The three analytic approaches suggest that, while at least one subgroup of ‘neurotic depression’ appears identifiable, it (and any other subgroups) are likely to be ‘fuzzy’ disorders.

Our next task will be to refine or further develop a number of measures addressing constructs suggested by our several analytic approaches as having some utility (and reflecting anxiety, personality style, disordered personality functioning and manifest symptoms) and then undertake similar cross-sectional analyses in an independent study group. In addition, we will need to undertake longitudinal studies of the identified subgroups to determine whether subclassification has implications in regard to natural history and treatment specificity. As noted earlier, the nonmelancholic depressive disorders are clearly a heterogeneous class and the spectrum model (assuming a surface pattern driven by an admixture of temperament and personality, or cognitive and behavioural patterns) would appear a useful one for consideration by researchers and clinicians. The key implication for the clinician is that there may be considerable utility in viewing these disorders as commonly being secondary conditions (whether to anxiety or disordered personality functioning or other factors) and then considering whether and how the primary determining condition might be addressed clinically.