Deconstructing depression by machine learning: the POKAL-PSY study

Abstract

AbstractUnipolar depression is a prevalent and disabling condition, often left untreated. In the outpatient setting, general practitioners fail to recognize depression in about 50% of cases mainly due to somatic comorbidities. Given the significant economic, social, and interpersonal impact of depression and its increasing prevalence, there is a need to improve its diagnosis and treatment in outpatient care. Various efforts have been made to isolate individual biological markers for depression to streamline diagnostic and therapeutic approaches. However, the intricate and dynamic interplay between neuroinflammation, metabolic abnormalities, and relevant neurobiological correlates of depression is not yet fully understood. To address this issue, we propose a naturalistic prospective study involving outpatients with unipolar depression, individuals without depression or comorbidities, and healthy controls. In addition to clinical assessments, cardiovascular parameters, metabolic factors, and inflammatory parameters are collected. For analysis we will use conventional statistics as well as machine learning algorithms. We aim to detect relevant participant subgroups by data-driven cluster algorithms and their impact on the subjects’ long-term prognosis. The POKAL-PSY study is a subproject of the research network POKAL (Predictors and Clinical Outcomes in Depressive Disorders; GRK 2621).