Primary Submission Category: Dynamic Treatment Regimes
Individual Treatment Effects in Bipolar Disorder with Latent Mood Dynamics
Authors: Anna Pham, Amy Cochran, Melvin Mcinnis,
Presenting Author: Anna Pham*
Individual-level experimentation is central to personalized decision-making, particularly for chronic conditions such as bipolar disorder, where treatment responses vary widely across individuals. In an ideal “n-of-1” setting, we would learn an individual’s treatment effect by repeatedly turning an intervention on and off and observing how outcomes respond. In practice, this is complicated by the fact that outcomes evolve dynamically and depend on prior outcomes, past interventions, and unobserved psychological states, creating carryover and temporal dependence that confound simple comparisons. We study the identification and estimation of individual treatment effects on mood in people with bipolar disorder, using a simulation environment grounded in parameters estimated from real patient data via a Kalman filter. Interventions are applied repeatedly over time and influence outcomes through latent state dynamics. We examine how longitudinal design choices affect the ability to recover accurate individual treatment effects and characterize how assumptions about mood dynamics and carryover shape estimation bias and variance. Our results provide practical guidance for individualized causal analysis in bipolar disorder and more broadly inform methods for personalized causal inference in dynamic settings.