Introduction
Classification systems of mental disorders provide patient models for the most likely symptoms of a typical patient with a specific illness. The profile of depression spans negativity bias, anhedonia, impairments in learning and memory, executive dysfunction, psychomotor change, and increased stress sensitivity (Hasler, Drevets, Manji, & Charney, Reference Hasler, Drevets, Manji and Charney2004). However, specific symptoms may vary substantially from patient to patient. Distinct subtypes of major depressive disorder (MDD) have been observed—such as melancholic and non-melancholic depression (Gold & Chrousos, Reference Gold and Chrousos2002; Parker, Reference Parker2000, Reference Parker2007)—where certain symptoms are associated with a specific subtype of depression. The current classification system for mental disorders (DSM-IV) indicates that classification systems work best when there is a clear distinction between disorder categories (DSM-IV; APA, 2000). Further understanding how these subtypes differ from one another may provide insights that will help improve accuracy of classification systems. The current study aims to explore the constructs of melancholic and non-melancholic depression and how they differ from one another drawing on both clinical and neuropsychological data.
While current classifications suggest major depressive disorder (MDD) is a homogeneous disorder that changes on an index of severity (Benazzi, Reference Benazzi2008; Judd, Schettler & Akisal, Reference Judd, Schettler and Akiskal2002; Sadek & Bona, Reference Sadek and Bona2000; Shankman & Klein, Reference Shankman and Klein2002), subtypes of depression are also diagnosed on the basis of certain symptoms being present in one subtype but not other. For instance, psychomotor retardation is characteristic of melancholic depression (see Parker, Reference Parker2007). However, the degree to which other symptoms are specific to subtype (such as anxiety) or dependent on severity remains a matter of debate (see Angst, Gamma, Benazzi, Adjacic, & Rossler, Reference Angst, Gamma, Benazzi, Ajdacic and Rossler2007; Coryell, Reference Coryell2007; Gold & Chrousos, Reference Gold and Chrousos2002; Judd et al., Reference Judd, Schettler and Akiskal2002; Parker, Reference Parker2000; Parker & Hadzi-Pavlovic, Reference Parker and Hadzi-Pavlovic1996).
The homogenous or unitarian position (see Judd et al., Reference Judd, Schettler and Akiskal2002) places subtypes on a continuum such that the manifestation of certain features is dependent on severity. It takes the position that subthreshold forms of the disorder are least severe, followed by non-melancholic depression, then melancholic depression and psychotic depression; considered to be the most severe. In this model, specific symptoms become more prominent as severity increases. Alternatively, the heterogeneous position (see Parker, Reference Parker2005) suggests that symptoms, such as psychomotor disturbance, indicate that subtypes are biologically distinct and are not a function of severity. Thus an individual may have a biological predisposition to a certain type of depression with specific features. There has been a lack of clarity on this issue, in part due to the presence of other features such as anxiety, which may be present in severe cases of the disorder regardless of subtype (Parker, Reference Parker2000).
Although past research has supported the homogenous position (see Kendall, Reference Kendall1976; Klein, Reference Klein1976; Parker & Hadzi-Pavlovic, Reference Parker and Hadzi-Pavlovic1996; Parker et al., Reference Parker, Hall, Boyce, Hadzi-Pavlovic, Mitchell and Wilhelm1991), Parker (Reference Parker2000) has argued that this is due to over inclusion of non-specific variables in research studies. These are variables such as depressed mood and sleep disturbances that do not distinguish between the different subtypes. The inclusion of such variables in studies reduces the impact of specifying variables such as psychomotor disturbance (Parker, Reference Parker2000). Furthermore, the objective evidence for psychomotor disturbance has been recommended, rather than relying on subjective reports from the patient, which may be ubiquitous across all subtypes (Parker, Reference Parker2000, Reference Parker2007). This consideration highlights the need to use objective rating scales such as the CORE (Hickie, Reference Hickie, Parker and Hazdi-Pavlovic1996). Psychomotor disturbance and anhedonia may relate to contributions from the dopaminergic neurotransmitter system, in contrast to other symptoms, such as sleep and appetite disturbance, which may have more of a serotonergic and/or noradrenergic basis (Malhi, Parker, & Greenwood, Reference Malhi, Parker and Greenwood2005; Stein, Reference Stein2008). This work provides a neurobiological basis for the heterogeneity proposal.
In addition to psychomotor disturbance, certain neuropsychological deficits have been reported in melancholic patients in comparison to non-melancholic patients. Higher level cognitive functions such as working memory (Austin et al., Reference Austin, Mitchell, Wilhelm, Parker, Hickie, Brodaty and Hadzi-Pavlovic1999) and attention shifting (Austin et al., Reference Austin, Mitchell, Wilhelm, Parker, Hickie, Brodaty and Hadzi-Pavlovic1999; Murphy et al., Reference Murphy, Sahakian, Rubinsztein, Michael, Rogers, Robbins and Paykel1999) are worse in melancholic depression in comparison to non-melancholic depression, while lower level functions such as verbal fluency do not differentiate these subtypes (Austin et al., Reference Austin, Mitchell, Wilhelm, Parker, Hickie, Brodaty and Hadzi-Pavlovic1999). While patients with melancholia consistently present with a more severe depression (see Schotte, Maes, Cluydts, & Cosyns, Reference Schotte, Maes, Cluydts and Cosyns1997), not all studies have been able to replicate the association between cognitive impairment and severity (see Stordal et al., Reference Stordal, Lundervold, Egeland, Mykletun, Asbjornsen, Landro and Lund2004; Fossati et al., Reference Fossati, Amar, Raoux, Ergis and Allilaire1999). Hence, it is unclear whether neuropsychological function is a result of subtype or severity; or whether the processes are interdependent.
The current study aims to to employ a data-driven approach to determine whether subtypes of depression could be identified on the basis of clinical symptoms and neuropsychological performance. To our knowledge, this is the first study to integrate both symptom and neuropsychological data to identify subtypes of depression.
Method
Participants
A total of 128 participants with a primary diagnosis of MDD were recruited from the community in collaboration with the Brain Resource International Database (Gordon, Cooper, Rennie, Hermens, & Williams, Reference Gordon, Cooper, Rennie, Hermens and Williams2005). All participants gave written informed consent in accordance with Australian National Health and Medical Research Council (NHMRC) ethical guidelines. Exclusion criteria included a history of brain injury (causing loss of consciousness for 10 min or more), neurological disorder, or other serious medical or genetic condition. All participants were medication free for at least five half-lives. All participants met criteria for MDD as determined by trained research officers using the Mini International Neuropsychological Interview (MINI: Sheehan et al., Reference Sheehan, Lecrubier, Sheehan, Amorium, Janavs, Weiller and Dunbar1998), a structured psychiatric interview based on the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) criteria. A primary diagnosis of MDD was required for inclusion in the study, and participants were categorized with or without melancholic or non-melancholic depression using the MINI, and the CORE rating scale (Parker & Hadzi-Pavlovic, Reference Parker and Hadzi-Pavlovic1996) was administered to confirm such categorization. Depressed participants with co-morbid anxiety disorder were included to maximize the generalizability of the sample and to further explore the relationship between anxiety, severity, and subtype. Participants were excluded if they had any current substance (including alcohol) dependence disorder on the basis of the MINI and AUDIT questionnaire (Babor, Higgins-Biddle, Saunders, & Monteiro, Reference Babor, Higgins-Biddle, Saunders and Monteiro2001). Secondary diagnoses included generalized anxiety disorder (n = 45), panic disorder (n = 29), and post-traumatic stress disorder (n = 14).
Procedure
Depression severity was assessed using the Hamilton Rating Scale for Depression (HRSD; Hamilton, Reference Hamilton1960; Williams, Reference Williams1988) and observable psychomotor disturbance was measured by the CORE Assessment of Psychomotor Change (CORE; Parker & Hadzi-Pavlovic, Reference Parker and Hadzi-Pavlovic1996). The MINI, HRSD and CORE were administered by trained research officers during the clinical interview. The Depression, Anxiety, and Stress Scale (DASS; Lovibond & Lovibond, Reference Lovibond and Lovibond1995) was completed by participants at the completion of the clinical interview. All participants subsequently completed the IntegNeuro, a standardized neuropsychological test battery that has established reliability and validity statistics (Kemp, Reference Kemp2009; Kemp et al., Reference Kemp, Stephan, Hopkinson, Sumich, Paul, Clark and Williams2005; Paul et al., Reference Paul, Lawrence, Williams, Clark, Cooper and Gordon2005; Silverstein et al., Reference Silverstein, Jaeger, Donovan-Lepore, Wilkniss, Savitz, Malinovsky and Dent2010; Williams et al., Reference Williams, Simms, Clark, Paul, Rowe and Gordon2005).
Measures
Hamilton rating scale for depression
Depression severity was examined using the clinician administered Hamilton Rating Scale for Depression (Hamilton, Reference Hamilton1960; Williams, Reference Williams1988), a semi-structured interview assessing the severity of individual symptoms across a 17-item scale. The total score was used as a measure of depression severity.
Depression, anxiety, and stress scales
Depression, anxiety, and stress symptoms were established using the Depression, Anxiety and Stress Scale (DASS; Lovibond & Lovibond, Reference Lovibond and Lovibond1995). The DASS is a 21-item self-report measure where items are rated on a four-point scale (0–3). Depression symptoms were measured by the total score from the DASS-D (depression) scale, anxious arousal was measured by the total score from the DASS-A (anxiety) scale, and anxious apprehension was measured by total score from the DASS-S (stress) scale. The depression scale assesses symptoms of depression, measuring anhedonia, dysphoria, hopelessness, worthlessness, demotivation, and lack of interest or enthusiasm, consistent with DSM-IV criteria of MDD mood disorders. The anxiety (or anxious arousal) scale measures autonomic arousal and subjective experience of anxiety (including dryness of mouth, difficulty breathing, pounding heart, sweaty palms), consistent with panic disorder and PTSD. The stress (or anxious apprehension) scale measures nervous tension, difficulty relaxing, and irritability, consistent with generalized anxiety disorder.
The CORE assessment of psychomotor change
Psychomotor disturbance was measured by the clinician administered CORE Assessment of Psychomotor Change (Hickie, Reference Hickie, Parker and Hazdi-Pavlovic1996). The CORE measure includes 18 items indicating observable features of psychomotor disturbance, which are rated at the end of a clinical interview. Each item is rated on a four-point scale (0–3) and the scale addresses three signs of psychomotor disturbance: non-interactiveness, retardation, and agitation. The total score was used as a measure of psychomotor disturbance.
Neuropsychological measures
The cognitive tests were administered using pre-recorded task instructions (via headphones). Task instructions included a computerized visual demonstration followed by a “test trial” before acquiring the data. Responses were given via a touch screen computer or .wav files for spoken answers. If a participant failed the test trial, the task instructions were repeated and elaborated upon. Cognitive functions of interest included estimated verbal memory recall, attention, numerical and spatial working memory and executive functioning. These domains of function have previously been identified to differ between melancholic and non-melancholic depression (see Austin et al., Reference Austin, Mitchell, Wilhelm, Parker, Hickie, Brodaty and Hadzi-Pavlovic1999; Quinn et al., under review). As such these functions were chosen to clarify the contribution of neuropsychological function to subtype and severity. To examine and control for differences on general ability a test of premorbid intelligence was also applied. The individual tasks chosen for assessment are as follows:
Memory recall
Memory recall was assessed using the verbal recall and recognition task. This task assesses verbal learning and memory encoding, memory retrieval and recognition and involves similar processes to that of the California Verbal Learning Test (Tombaugh & Hubley, Reference Tombaugh and Hubley2001) and the Rey Auditory Verbal Learning and Memory task (Geffen, Moar, O'Hanlon, Clark, & Geffen, Reference Geffen, Moar, O'Hanlon, Clark and Geffen1990; Rey, Reference Rey1964). In this task, participants are presented with a list of 12 words, which they are asked to memorize and recall after each presentation. Words are between four and seven characters in length and were closely matched on concreteness, number of letters and frequency. After four trials a distractor list is given (trial 5) and a further two trials are completed using the original list. The measure included in this study was trial 6, where participants ability for immediate memory recall is assessed.
Attention
Attention was assessed using the time estimation task. The time estimation task requires participants to indicate the length of time (in seconds) a black circle turns to color (green) thus giving an estimate of controlled attention. The response is indicated via the touch screen providing numbers from 1 through to 12 representing number of seconds in a sequence across the screen. Attention was measured by the standard deviation of the proportional bias in estimation time. Time is estimated from the absolute value of the average difference between the actual duration of the stimulus and the participants estimate weighted by the length of the stimulus.
Working memory
Numerical working memory was measured via the reverse digit span task. The participant hears a series of digits (each presented for 500 ms, separated by 1-s intervals). The participant is required to repeat the digits back in the reverse order. The number of digits in the sequence is gradually increased from 3 to 9 with 2 repetitions at each level. The measure taken is the total numbers of digits recalled in the correct order. Spatial working memory was measured via the span of visual memory task which is adapted from the Corsi Blocks task (Milner, Reference Milner1974) and the Dot Location task (Roth & Crosson, Reference Roth and Crosson1985). In this task, nine squares on a touch screen light up in a random order. Four seconds later, participants hear a tone which indicates they have to reproduce the sequence by pressing squares on the touch screen. The measure is the total number of correct squares in the sequence without error.
Executive functioning
Executive functioning was measured by the executive maze task in which a dot-based maze is presented on the screen. This task is a computerized adaptation of the Austin Maze (Crowe et al., Reference Crowe, Barclay, Brennan, Farkas, Gould, Katchmarsky and Vayda1999). The participant is required to find by trial and error a hidden path through the maze and the sequential steps taken to achieve this. The measure taken in this task was the total number of trials necessary before completion.
Premorbid intelligence
Premobid intelligence was measured with the spot the real word test, a computerized adaptation of the spot the word test (Baddeley, Emslie, & Nimmo-Smith, Reference Baddeley, Emslie and Nimmo-Smith1993). This test assesses premorbid intelligence and involves differentiating between real words and nonsense words. Real word and nonsense word pairs are presented on the touch screen and the participant has to specify which is the “real” word by pressing the touch screen. The measure is the total number of correct responses.
Statistical Analyses
All statistical analyses were performed using SPSS (Version 17: SPSS Inc, Chicago, IL). Significant effects were set at p < .05, while findings were labeled as trends if p < .1 and p > .05. Participants were excluded from the study at the point of statistical analysis if they were identified as an outlier deemed to be 1.5 times the interquartile range on the premorbid intelligence test, on level of education and also on age. An initial series of chi square and t tests were conducted to examine differences between groups.
Data driven approach to categorization
A cluster analysis was used to determine whether subtypes of depression were present on the basis of clinical symptoms and neuropsychological function using the DASS questionnaire, CORE checklist, and neuropsychological battery. Cluster analysis was chosen for its ability to identify groupings, patterns of attributes or categories that can be interpreted as a meaningful set by identifying how each category differs from the others. A two-step cluster analysis was used using Schwarz Bayesian Information Criterion (BIC) and log-likelihood distance to determine the number of clusters. The two-step cluster analysis classifies each case (depressed patients) into a category or group and each category is part of a larger set of categories that together define the set of phenomena (subtypes of depressed patients). Each cluster has a score on each variable, but the value of the scores varies across clusters. It is the pattern of variation across clusters within a cluster solution that serves as the interpretive key. A χ2 analysis was conducted from the results of the cluster analysis to examine whether there was a relationship between the assigned cluster and subtype.
Predicting Clinical Depression
A backward stepwise binary logistic regression was used to assess the relative contribution of all clinical and neuropsychological variables to the diagnosis of melancholic depression. Odds ratios were used to assess the strength of the relationship between melancholic depression and the predictor variables. Standardized and unstandardized regression coefficients have been reported for depression severity indicating differences per standard deviation change and per unit change respectively. A backward stepwise linear regression was used to assess the relative contribution of clinical and neuropsychological variables to depression severity. The coefficient weights were used to assess the strength of the relationship between depression severity and the predictor variables. Linear regression was chosen as both dependant and independent variables were continuous and either increase or decrease in a linear manner with respect to severity.
Results
Participant Characteristics
There were no significant differences for group on gender (χ 2(1, N = 124) = 0.976; p = .323), age (t(1,122) = −1.470; p = .144) years of education (t(1,121) = −1.296; p = .197), or premorbid intelligence (F(1,122) = −0.299; p = .766). Significant differences were observed between groups (melancholic depression, MEL; non-melancholic depression, NMEL) on the HRSD (t(1,122)=−3.818; p < .001), DASS-A (F(1,115.200) = −2.653; p = .009), the DASS-S (t(1,119.003) = −2.338; p = .021), and the CORE total score (F(1,94.878) = −4.109; p < .001) such that patients with melancholic depression displayed increased scores on all measures than those with non-melancholic depression. Groups also showed a trend toward differing on the DASS-D (t(1, 122) = −1.826; p = .070), consistent with the finding for HRSD. Melancholic depression was associated with the diagnosis of comorbid anxiety (χ 2(2) = 8.116; p = .004), such that 77% of patients with melancholic depression had a comorbid anxiety disorder in contrast to 47% of patients with non-melancholic depression. Consistent with this, patients with melancholic depression displayed increased DASS anxiety scores relative to those with non-melancholic depression (MEL = 14.677; NMEL = 10.627). Participants with melancholic depression were also more likely to have a diagnosis of panic disorder (χ 2(2) = 6.067; p = .011). Groups did not differ on GAD, PTSD, or substance abuse. See Table 1 for current comorbidity data relevant to subtypes (i.e., melancholic and non-melancholic depression).
Gender, age, and comorbidity data for melancholic vs. non-melancholic depression
Cluster Analysis: Exploratory approach to Subtype
Two clusters were auto-determined using SPSS with the optimal result indicated by the lowest BIC value. For 2 clusters, the BIC was 811.39, the BIC change was −44.41, the ratio of BIC change was 1.00, and the ratio of distance measures was 1.98. Cluster membership was assigned using BIC, an approximation to the log factor for the model of interest compared to the saturated model. The means and standard deviations indicating the centroids for each cluster along with the comparison data for the diagnostic categories (i.e., melancholic and non-melancholic depression) are listed in Table 2.
Means and standard deviations of clusters
Cluster 1 was indicated by lower than mean performance on neuropsychological function (in order of importance: numerical working memory, executive functioning, spatial working memory, memory recall and attention) and also higher than mean levels of symptom severity (anxiety, stress, depression, and psychomotor disturbance).
Cluster 2 was the opposite of cluster 1; indicating higher than mean performance on neuropsychological function (attention, executive functioning, spatial working memory, numerical working memory, and memory recall) and also lower than mean levels of symptom severity (anxiety, stress, depression, and psychomotor disturbance). While lower scores on memory recall, numerical working memory, and spatial working memory indicate poorer performance, higher scores on attention and executive functioning indicate poorer performance. Higher scores on the clinical measures (psychomotor disturbance, depression, anxiety, and stress) also indicate increased severity. The variable importance plots (Figures 1 and 2) display the level of importance each variable holds to a cluster with respect to the other.
Variable importance plot for cluster 1. This figure shows the level of importance each variable contributes to defining cluster 1. This profile is characterized by impairments in working memory and executive functioning.
Variable importance plot for cluster 2. This figure shows the level of importance each variable contributes to defining cluster 2. Distinct from cluster 1, this profile highlights a role for (improved) attention and (reduced) anxiety in determining membership for cluster 2. Profiles of clusters 1 and 2 are distinct; based on the importance of individual variables in defining each cluster.
Chi-Square: Relationship Between Cluster and Subtype
A relationship between subtype and cluster was found to be significant (χ 2(1) = 7.495, p < .01) with 58% of patients with melancholic depression being assigned to cluster 1 and 66% of patients with non-melancholic depression assigned to cluster 2. See Table 3 for comparison of frequencies and percentages relating to the subtypes and clusters.
Frequencies and percentages of classification: Clusters and subtype
Logistic Regression: Melancholic Depression
Four predictors (stress, psychomotor disturbance, memory recall, and attention) contributed significantly to the model for melancholic depression (see Table 4 for odds ratios, and unstandardized coefficients for each significant predictor). Higher levels of stress and psychomotor disturbance and poorer attention were associated with melancholic depression in comparison to non-melancholic depression. By contrast, memory recall impairment was associated with non-melancholic depression. The odds that a patient will be melancholic are multiplied by 5.329 for each unit increase in the standard deviation on the measure of attention and by 1.281 for each unit increase on the CORE. The odds that a patient will be non-melancholic will be increased by 1.273 for each point decrease on the memory recall task.
Predictors contributing to the model for melancholic depression, their unstandardized coefficients, the associated variance explained by each predictor, and statistics
*Positive value indicative of melancholia vs. non-melancholia. Here melancholic patients have increased memory recall scores whereas non-melancholic patients have decreased memory recall scores.
Linear Regression: Depression Severity
Three predictors (anxiety, psychomotor disturbance, and attention) contributed significantly to the model of depression severity (see Table 5 for standardized and unstandardized coefficients, explained by each significant predictor and variance). Higher levels of anxiety, psychomotor disturbance, and greater variability in attention were associated with depression severity.
Predictors contributing to the model for depression severity, their standardized coefficients, the associated variance explained by each predictor, and statistics
Discussion
The results from the cluster analysis reveal two types of depression relevant to clinical and neuropsychological features. The first cluster indicates a subtype in which clinical and neuropsychological features are more severe than average whereas the second cluster indicates a subtype where these features are less severe than average. No specific symptom was a clear marker of either subtype and segregation was based on severity. Results from the χ2 analysis revealed that these clusters were associated with the melancholic and non-melancholic subtypes of depression, such that melancholia is associated with more severe clinical and neuropsychological features in comparison to the non-melancholic subtype. However, the logistic regression revealed that anxious apprehension, psychomotor disturbance and variability in attentional capacity contributed to the diagnosis of melancholic depression, while poorer performance on the memory recall task contributed to non-melancholia. The linear regression revealed increased levels of anxiety, psychomotor disturbance, and attention variability contributed to depression severity.
The utility of cluster analysis has been questioned due to suggestions that the technique will always produce clusters such that the number of meaningful clusters should be interpreted with reservation (Parker, Reference Parker2000). In the current study, we did not assign a specific number of clusters to be drawn from the data and yet the two clusters were obtained that contributed to the two subtypes of interest. By using a χ2 analysis, we were able to test the applicability of cluster analysis to diagnosis. Despite a significant relationship being observed between subtype and cluster; only 58% of melancholic patients assigned to cluster 1 and 66% of non-melancholic patients assigned to cluster 2. While it is likely that there are other contributors to the types of depression including temperament, personality, physical symptoms, and genetic vulnerability, findings provide support for the heterogeneity of depression. The results of the cluster analysis segregate the groups by severity; however the clusters were further characterized by specific neuropsychological features. The key difference is that participants diagnosed with non-melancholia cluster displayed poorer performance on the memory recall task. On this task they were in fact more severe than melancholic participants. This finding is paradoxical in the context of the homogeneous model of depression. If depression were a homogenous disorder we would expect the more severe cluster to display poorer memory recall skills. Whereas severity is clearly an appropriate indicator to separate melancholia and non-melancholia, greater diagnostic accuracy may be gained from looking at specific features such as memory recall for non-melancholic patients and attention capacity in melancholic patients. The variables that contributed to severity were anxious arousal, psychomotor disturbance and variability in attention, key characteristics of melancholic depression. This supports Parker's (Reference Parker2000) observation that although melancholic depression is generally considered a more severe depression, its differentiation should be based on symptoms associated with subtype. In the current study we observed that melancholic depression is associated with higher levels of anxious apprehension, psychomotor disturbance, and poorer attention, whereas non-melancholic depression was associated with memory recall deficits. Severity was associated with anxious arousal, psychomotor disturbance, and poorer attention, all of which contribute to the diagnosis of melancholic depression.
Several limitations of the study are worth noting. First, the study only focuses on the diagnoses of melancholia and non-melancholia. Here, we provide support for a heterogeneous model of depression on the basis of clinical and neuropsychological features; however, further research is needed to determine whether distinctive features can be characterized for other subtypes of depression such as atypical and post-partum depression. Second, the HDRS is biased for melancholic participants in terms of increased severity and does not cover all the core depressive symptoms as outlined in the DSM-IV. Regardless, our study identifies specific neuropsychological features such as sustained attention and memory recall that may further contribute to diagnosis of a specific subtype. Third, our patient sample was characterized by secondary diagnoses of generalized anxiety disorder (GAD), panic disorder, and post-traumatic stress disorder, which may also contribute to the neuropsychological impairment observed. However, the prevalence of GAD and PTSD did not differ between subtypes, and although patients with melancholia were more likely to have a diagnosis of panic disorder and increased anxiety, this contributed to depression severity. We propose therefore that these demographic differences do not impact on the major conclusion drawn from our results: that depression is a heterogeneous disorder over and above the impact of depression severity.
Our study highlights the need for classification systems to take into account a greater range of phenotypic characteristics including neuropsychological function. Whilst greater validity in diagnosis can be achieved in this manner our results also support a complementary role for severity assessment, particularly in cases where phenotypic indicators are less clear. Further research is warranted into other subtypes of depression to gain greater clarity into the underlying features of specific subtypes. In summary, we conclude that clinical and neuropsychological variables contribute to a heterogeneous model of depression where subtype is a primary function of specific symptoms and severity is a secondary basis for differentiation.
Acknowledgments
We acknowledge the data and support provided by BRAINnet; www.BRAINnet.net, under the governance of the BRAINnet Foundation. BRAINnet is the scientific network that coordinates access to the Brain Resource International Database for independent scientific purposes. CQ was supported by an Australian Postgraduate Award and AK, by a National Health and Medical Research Council Career Development Award Fellowship (571101). This research was supported by was supported by a National Health and Medical Research Council (NHMRC) Project Grant (464863) and an Australian Research Council Discovery Grant (DP0987332). AH has been awarded research funding from the NHMRC, Australian Rotary, Perpetual Trustees, Eli Lilly Australia, Janssen-Cilag Australia and the Schizophrenia Fellowship of NSW. AH has also received consultancy fees from Organon Australia, Eli Lilly and Lundbeck Australia. AH has received payments for educational sessions run for Astra Zeneca, Janssen Cilag, Eli Lilly, and Organon. AH has also run educational sessions for several medical education companies including Wellmark Australia, Reed Business Information, and CME LLC. Johnson and Johnson Pharmaceutical Research and Development, RED Europe also supported the project from which data in the current study was drawn. Sponsors played no role in analysis and interpretation of data; or in the decision to submit the paper for publication. The authors, CQ and AK, had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
