Interactive Ensemble clustering for mixed data with application to mood disorders

Abstract

Mental disorders are among the most elusive conditions in medicine and defy simple models, be they biological, psychological, social, or any simplistic admixture. In contrast to current classifications in other areas of medicine, those used in psychiatry, including the Diagnostic and Statistical Manual of Mental Disorders (DSM), rely on clinical manifestations (signs) and subjective reports (symptoms) rather than on the underlying causes and mechanisms.

Three landmark naturalistic studies funded by the National Institute of Mental Health (NIMH) provided some sobering statistics in this respect: psychiatric interventions are effective in less than 25% of patients presenting with an acute episode.17, 27 Diagnoses of mental health conditions are currently characterized by the following:

• Based on little objective evidence (almost arbitrary)
• No biological markers
• Co-morbidity (one person having multiple diagnoses)
• Heterogeneity within diagnosis (two patients with the same diagnosis can have two different sets of symptoms, with little or no clinical overlap)

In this project, we aim to develop a novel quantitative, big-data approach to enable precision diagnosis and treatment in this challenging application domain, with the ultimate goal providing data-driven tools that help clinicians significantly focus patient diagnosis and improve mental health outcomes.