Using cluster ensembles to identify psychiatric patient subgroups

Abstract

Identification of patient subgroups is an important process for supporting clinical care in many medical specialties. In psychiatry, patient stratification is mainly done using a psychiatric diagnosis following the Diagnostic and Statistical Manual of Mental Disorders (DSM). Diagnostic categories in the DSM are however heterogeneous, and many symptoms cut across several diagnoses, leading to criticism of this approach. Data-driven approaches using clustering algorithms have recently been proposed, but have suffered from subjectivity in choosing a number of clusters and a clustering algorithm. We therefore propose to apply cluster ensemble techniques to the problem of identifying subgroups of psychiatric patients, which have previously been shown to overcome drawbacks of individual clustering algorithms. We first introduce a process guide for modelling and evaluating cluster ensembles in the form of a Meta Algorithmic Model. Then, we apply cluster ensembles to a novel cross-diagnostic dataset from the Psychiatry Department of the University Medical Center Utrecht in the Netherlands. We finally describe the clusters that are identified, and their relations to several clinically relevant variables.