249. A head-to-head comparison of two biologically-based subtyping solutions for major depressive disorder

Abstract

Abstract Background Major depressive disorder (MDD) is a heterogeneous psychiatric syndrome, and many patients do not respond fully to an initial trial of a selective serotonin reuptake inhibitor (SSRI), the most prescribed first-line antidepressant class. The neurobiological basis of heterogeneity in SSRI responses is poorly understood, and conventional approaches to assessing treatment response typically do not differentiate between improvements in anhedonia, mood, insomnia, and other symptom domains, which vary across individuals. Motivated by these imperatives, there is renewed interest in efforts to delineate novel subtypes of depression, which could open avenues for identifying subtype-specific biomarkers and informing treatment selection decisions. However, the field has progressed rapidly in recent years and there are now numerous MDD subtyping models explaining different aspects of heterogeneity. How do these models relate to one another? Aims & Objectives We aimed to develop and compare two neurobiological subtypes of depression defined by distinct patterns of abnormal connectivity in depression-related brain networks. Method We leveraged two independent resting-state fMRI datasets from male and female adults with MDD (Dataset 1 [D1] n = 328 and Dataset 2 [D2] n = 130). Participants in D1 were treated with repetitive transcranial magnetic stimulation (rTMS), while those in D2 were treated with escitalopram. In each dataset, we used regularized canonical correlation analysis, optimized using a nested cross-validation scheme, to identify brain-behaviour dimensions representing associations between baseline resting-state functional connectivity (RSFC) and either baseline depressive symptoms (D1) or subsequent improvements in specific depressive symptoms (D2). We performed hierarchical clustering on these components to identify discrete subtypes defied by baseline depressive symptoms (D1) or symptom-specific SSRI response (D2). Finally, we projected the D2 data with the D1 subtyping model generated using baseline severity and vice versa to compare how distinct subtypes predict response to escitalopram or rTMS in both models. Results In D1, leveraging baseline symptom severity, we identified three significant (p < .05) latent dimensions that clustered into four MDD subtypes that differed on baseline anhedonia, mood, anxiety and somatic symptoms. In D2, leveraging symptom-specific SSRI improvement, we identified four significant (p < .05) latent dimensions that clustered into three MDD subtypes, that differed on SSRI-induced improvements in mood, anhedonia, anxiety, and neurovegetative symptoms. rTMS and escitalopram response rates significantly differed for both models (p < .05). For the four baseline MDD subtypes, subtype 1 responded optimally to both interventions and subtypes 2 and 4 responded preferentially to escitalopram and rTMS, respectively. For the three symptom-improvement MDD subtypes and escitalopram response, subtype 1 experienced significant anxiety improvement, and subtype 2 had the greatest overall response. Subtype 3 had modest mood improvements, and minimal anxiety/neurovegetative improvements to escitalopram, but the best overall improvement to rTMS. Discussion & Conclusions Different symptom dimensions and subtypes respond optimally to escitalopram relative to rTMS. Our results could inform two testable hypotheses: first, that individuals assigned to our baseline symptoms subtype 1 could stand to particularly benefit from SSRI+rTMS combination therapy; and second, individuals assigned to symptom-improvement subtype 3 should potentially skip first-line antidepressants altogether in favour of rTMS.