Novel approaches for the modeling of disease progression in multiple sclerosis Project Summary The disease course of patients with multiple sclerosis (MS) is known to be highly variable, and this heterogeneity complicates prediction of disease progression. Since several potent treatments for MS have been recently developed, identification of factors associated with disease progression is extremely important so that optimal treatment decisions can be made. The expanded disability severity scale (EDSS) is the main clinical measure of disease course, and changes on this scale are used to measure disease progression in MS. Since the EDSS is an ordinal scale rather than a continuous variable, traditional statistical models for longitudinal data are not appropriate for modeling disease progression. An alternative approach is to model transitions between EDSS states over time using a Markov model. This approach requires a number of important modeling decisions, including selection of an appropriate link function and the appropriate number of states in the model. In addition, the poor interrater reliability and presence of relapse can impact the estimated probabilities of transition among the EDSS states from simple transition models. Therefore, extensions to the traditional Markov model will also be investigated. We will jointly apply Bayes factors and Bayesian variable selection to our proposed transition models to select the appropriate model framework. Then, under a selected model framework, we will apply our previously published variable selection approach to identify covariates associated with specific transitions. Such an approach will allow for more accurate prediction of patient- specific disease progression profiles.
