Data-driven classification of bipolar I disorder from longitudinal course of mood

The Diagnostic and Statistical Manual of Mental Disorder (DSM) classification of bipolar disorder defines categories to reflect common understanding of mood symptoms rather than scientific evidence. This work aimed to determine whether bipolar I can be objectively classified from longitudinal mood data and whether resulting classes have clinical associations. Bayesian nonparametric hierarchical models with latent classes and patient-specific models of mood are fit to data from Longitudinal Interval Follow-up Evaluations (LIFE) of bipolar I patients (N= 209). Classes are tested for clinical associations. No classes are justified using the time course of DSM-IV mood states. Three classes are justified using the course of subsyndromal mood symptoms. Classes differed in attempted suicides (P= 0.017), disability status (P= 0.012) and chronicity of affective symptoms (P= 0.009). Thus, bipolar I disorder can be objectively classified from mood course, and individuals in the resulting classes share clinical features. Data-driven classification from mood course could be used to enrich sample populations for pharmacological and etiological studies.