Data-driven subtypes of major depressive disorder

Patients with major depressive disorder constitute a heterogeneous group: depressive patients differ considerably in terms of clinical presentation, course of illness, and underlying mechanisms. These differences complicate ‘one size fits all’ solutions in research and treatment assignments.

In her dissertation, Hanna van Loo aimed to reduce these differences by searching for data-driven subtypes of major depressive disorder: subgroups of depressed patients with important similarities as identified by statistical data-analyses. She investigated theoretical, methodological and empirical aspects relevant for this search, by focusing on three questions: what sort of categories are we looking for, what methods are suited to identify them and what do the data show us?

To answer these questions, she first applied advanced statistical learning methods to large datasets in the Netherlands and the United States so as to identify groups of patients with a high or low risk for a severe course of depression. These studies resulted in three preliminary data-driven subtypes predicting a severe, moderate and mild course of illness: the most severe subtype predicted significantly more future episodes of depression, hospitalizations, and disability. Second, she performed several theoretical studies of psychiatric comorbidity – the fact that many psychiatric patients have more than one psychiatric disorder – to promote the understanding of classifications of depression using insights from the philosophy of science. The results of these studies demonstrate the potential of data-driven subtypes of depression as bases for clinically relevant classifications and provide several starting-points for future research.