A0995
Title: SMART-MC: Characterizing the dynamics of multiple sclerosis therapy transitions using a covariate-based Markov model
Authors: Priyam Das - Virginia Commonwealth University (United States) [presenting]
Abstract: Treatment switching is common in managing multiple sclerosis (MS), where patients transition across disease-modifying therapies (DMTs) due to variable responses, disease progression, patient characteristics, or side effects. To study how covariates influence treatment transitions, a Markovian model sparse matrix estimation with covariate-based transitions in Markov chain modeling (SMART-MC) is adopted, in which transition probabilities depend on patient-level covariates. Modeling real-world transitions poses challenges, such as parameter identifiability and sparsity. Identifiability is addressed by constraining each transition-specific covariate vector to have unit L2 norm. Sparsely observed transitions are estimated as constants, and empirically unobserved transitions are set to zero, reducing complexity while preserving interpretability. To optimize the nonconvex likelihood, a scalable, parallel global optimization algorithm is developed, validated through benchmarks, and supported by theoretical guarantees. The analysis reveals distinct DMT switching patterns across MS subgroups defined by age, race, and clinical characteristics.
