Simultaneous Decomposition of Depression Heterogeneity on the Person-, Symptom- and Time-Level: The Use of Three-Mode Principal Component Analysis

Response trajectories during escitalopram treatment of patients with major depressive disorder

Depression is a leading global cause of disability, yet about half of patients do not respond to initial antidepressant treatment. This treatment difficulty may be in part due to the heterogeneity of depression and corresponding response to treatment. Unsupervised machine learning allows underlying patterns to be uncovered, and can be used to understand this heterogeneity by finding groups of patients with similar response trajectories. Prior studies attempting this have clustered patients using a narrow range of data primarily from depression scales. In this work, we used unsupervised machine learning to cluster patients receiving escitalopram therapy using a wide variety of subjective and objective clinical features from the first eight weeks of the Canadian Biomarker Integration Network in Depression-1 trial. We investigated how these clusters responded to treatment by comparing changes in symptoms and symptom categories, and by using Principal Component Analysis (PCA). Our algorithm found three clusters, which broadly represented non-responders, responders, and remitters. Most categories of features followed this response pattern except for objective cognitive features. Using PCA with our clusters, we found that subjective mood state/anhedonia is the core feature of response with escitalopram, but there exists other distinct patterns of response around neurovegetative symptoms, activation, and cognition.

Failure to Identify Robust Latent Variables of Positive or Negative Valence Processing Across Units of Analysis

BACKGROUND

The heterogeneous nature of mood and anxiety disorders highlights a need for dimensionally based descriptions of psychopathology that inform better classification and treatment approaches. Following the Research Domain Criteria approach, this investigation sought to derive constructs assessing positive and negative valence domains across multiple units of analysis.

METHODS

Adults with clinically impairing mood and anxiety symptoms (N = 225) completed comprehensive assessments across several units of analysis. Self-report assessments included nine questionnaires that assess mood and anxiety symptoms and traits reflecting the negative and positive valence systems. Behavioral assessments included emotional reactivity and distress tolerance tasks, during which skin conductance and heart rate were measured. Neuroimaging assessments included fear conditioning and a reward processing task. The latent variable structure underlying these measures was explored using sparse Bayesian group factor analysis.

RESULTS

Group factor analysis identified 11 latent variables explaining 31.2% of the variance across tasks, none of which loaded across units of analysis or tasks. Instead, variance was best explained by individual latent variables for each unit of analysis within each task. Post hoc analyses 1) showed associations with small effect sizes between latent variables that were derived separately from functional magnetic resonance imaging and self-report data and 2) showed that some latent variables are not directly related to individual valence system constructs.

CONCLUSIONS

The lack of latent structure across units of analysis highlights challenges of the Research Domain Criteria approach and suggests that while dimensional analyses work well to reveal within-task features, more targeted approaches are needed to reveal latent cross-modal relationships that could illuminate psychopathology.

Treatment response classes in major depressive disorder identified by model-based clustering and validated by clinical prediction models

The identification of generalizable treatment response classes (TRC[s]) in major depressive disorder (MDD) would facilitate comparisons across studies and the development of treatment prediction algorithms. Here, we investigated whether such stable TRCs can be identified and predicted by clinical baseline items. We analyzed data from an observational MDD cohort (Munich Antidepressant Response Signature [MARS] study, N = 1017), treated individually by psychopharmacological and psychotherapeutic means, and a multicenter, partially randomized clinical/pharmacogenomic study (Genome-based Therapeutic Drugs for Depression [GENDEP], N = 809). Symptoms were evaluated up to week 16 (or discharge) in MARS and week 12 in GENDEP. Clustering was performed on 809 MARS patients (discovery sample) using a mixed model with the integrated completed likelihood criterion for the assessment of cluster stability, and validated through a distinct MARS validation sample and GENDEP. A random forest algorithm was used to identify prediction patterns based on 50 clinical baseline items. From the clustering of the MARS discovery sample, seven TRCs emerged ranging from fast and complete response (average 4.9 weeks until discharge, 94% remitted patients) to slow and incomplete response (10% remitted patients at week 16). These proved stable representations of treatment response dynamics in both the MARS and the GENDEP validation sample. TRCs were strongly associated with established response markers, particularly the rate of remitted patients at discharge. TRCs were predictable from clinical items, particularly personality items, life events, episode duration, and specific psychopathological features. Prediction accuracy improved significantly when cluster-derived slopes were modelled instead of individual slopes. In conclusion, model-based clustering identified distinct and clinically meaningful treatment response classes in MDD that proved robust with regard to capturing response profiles of differently designed studies. Response classes were predictable from clinical baseline characteristics. Conceptually, model-based clustering is translatable to any outcome measure and could advance the large-scale integration of studies on treatment efficacy or the neurobiology of treatment response.

Severity, course trajectory, and within-person variability of individual symptoms in patients with major depressive disorder

BACKGROUND

Depression shows a large heterogeneity of symptoms between and within persons over time. However, most outcome studies have assessed depression as a single underlying latent construct, using the sum score on psychometric scales as an indicator for severity. This study assesses longitudinal symptom-specific trajectories and within-person variability of major depressive disorder over a 9-year period.

METHODS

Data were derived from the Netherlands Study of Depression and Anxiety (NESDA). This study included 783 participants with a current major depressive disorder at baseline. The Inventory Depressive Symptomatology-Self-Report (IDS-SR) was used to analyze 28 depressive symptoms at up to six time points during the 9-year follow-up.

RESULTS

The highest baseline severity scores were found for the items regarding energy and mood states. The core symptoms depressed mood and anhedonia had the most favorable course, whereas sleeping problems and (psycho-)somatic symptoms were more persistent over 9-year follow-up. Within-person variability was highest for symptoms related to energy and lowest for suicidal ideation.

CONCLUSIONS

The severity, course, and within-person variability differed markedly between depressive symptoms. Our findings strengthen the idea that employing a symptom-focused approach in both clinical care and research is of value.

Why so GLUMM? Detecting depression clusters through graphing lifestyle-environs using machine-learning methods (GLUMM)

BACKGROUND

Key lifestyle-environ risk factors are operative for depression, but it is unclear how risk factors cluster. Machine-learning (ML) algorithms exist that learn, extract, identify and map underlying patterns to identify groupings of depressed individuals without constraints. The aim of this research was to use a large epidemiological study to identify and characterise depression clusters through "Graphing lifestyle-environs using machine-learning methods" (GLUMM).

METHODS

Two ML algorithms were implemented: unsupervised Self-organised mapping (SOM) to create GLUMM clusters and a supervised boosted regression algorithm to describe clusters. Ninety-six "lifestyle-environ" variables were used from the National health and nutrition examination study (2009-2010). Multivariate logistic regression validated clusters and controlled for possible sociodemographic confounders.

RESULTS

The SOM identified two GLUMM cluster solutions. These solutions contained one dominant depressed cluster (GLUMM5-1, GLUMM7-1). Equal proportions of members in each cluster rated as highly depressed (17%). Alcohol consumption and demographics validated clusters. Boosted regression identified GLUMM5-1 as more informative than GLUMM7-1. Members were more likely to: have problems sleeping; unhealthy eating; ≤2 years in their home; an old home; perceive themselves underweight; exposed to work fumes; experienced sex at ≤14 years; not perform moderate recreational activities. A positive relationship between GLUMM5-1 (OR: 7.50, P<0.001) and GLUMM7-1 (OR: 7.88, P<0.001) with depression was found, with significant interactions with those married/living with partner (P=0.001).

CONCLUSION

Using ML based GLUMM to form ordered depressive clusters from multitudinous lifestyle-environ variables enabled a deeper exploration of the heterogeneous data to uncover better understandings into relationships between the complex mental health factors.

Predicting long-term depression outcome using a three-mode principal component model for depression heterogeneity

BACKGROUND

Depression heterogeneity has hampered development of adequate prognostic models. Therefore, more homogeneous clinical entities (e.g. dimensions, subtypes) have been developed, but their differentiating potential is limited because neither captures all relevant variation across persons, symptoms and time. To address this, three-mode Principal Component Analysis (3MPCA) was previously applied to capture person-, symptom- and time-level variation in a single model (Monden et al., 2015). This study evaluated the added prognostic value of such an integrated model for longer-term depression outcomes.

METHODS

The Beck Depression Inventory (BDI) was administered quarterly for two years to major depressive disorder outpatients participating in a randomized controlled trial. A previously developed 3MPCA model decomposed the data into 2 symptom-components ('somatic-affective', 'cognitive'), 2 time-components ('recovering', 'persisting') and 3 person-components ('severe non-persisting depression', 'somatic depression' and 'cognitive depression'). The predictive value of the 3MPCA model for BDI scores at 3-year (n=136) and 11-year follow-up (n=145) was compared with traditional latent variable models and traditional prognostic factors (e.g. baseline BDI component scores, personality).

RESULTS

3MPCA components predicted 41% and 36% of the BDI variance at 3- and 11-year follow-up, respectively. A latent class model, growth mixture model and other known prognostic variables predicted 4-32% and 3-24% of the BDI variance at 3- and 11-year follow-up, respectively.

LIMITATIONS

Only primary care patients were included. There was no independent validation sample.

CONCLUSION

Accounting for depression heterogeneity at the person-, symptom- and time-level improves longer-term predictions of depression severity, underlining the potential of this approach for developing better prognostic models.

Decomposing the heterogeneity of depression at the person-, symptom-, and time-level: latent variable models versus multimode principal component analysis

BACKGROUND

Heterogeneity of psychopathological concepts such as depression hampers progress in research and clinical practice. Latent Variable Models (LVMs) have been widely used to reduce this problem by identification of more homogeneous factors or subgroups. However, heterogeneity exists at multiple levels (persons, symptoms, time) and LVMs cannot capture all these levels and their interactions simultaneously, which leads to incomplete models. Our objective is to briefly review the most widely used LVMs in depression research, illustrating their use and incompatibility in real data, and to consider an alternative, statistical approach, namely multimode principal component analysis (MPCA).

METHODS

We applied LVMs to data from 147 patients, who filled out the Quick Inventory of Depressive Symptomatology (QIDS) at 9 time points. Compatibility of the results and suitability of the LVMs to capture the heterogeneity of the data were evaluated. Alternatively, MPCA was used to simultaneously decompose depression on the person-, symptom- and time-level and to investigate the interactions between these levels.

RESULTS

QIDS-data could be decomposed on the person-level (2 classes), symptom-level (2 factors) and time-level (2 trajectory-classes). However, these results could not be integrated into a single model. Instead, MPCA allowed for decomposition of the data at the person- (3 components), symptom- (2 components) and time-level (2 components) and for the investigation of these components' interactions.

CONCLUSIONS

Traditional LVMs have limited use when trying to define an integrated model of depression heterogeneity at the person, symptom and time level. More integrative statistical techniques such as MPCA can be used to address these relatively complex data patterns and could be used in future attempts to identify empirically-based subtypes/phenotypes of depression.