Bayesian P-Splines to investigate the impact of covariates on Multiple Sclerosis clinical course

This paper aims at proposing suitable statistical tools to address heterogeneity in repeated measures, within a Multiple Sclerosis (MS) longitudinal study. Indeed, due to unobservable sources of heterogeneity, modelling the effect of covariates on MS severity evolves as a very difficult feature. Bayesian P-Splines are suggested for modelling non linear smooth effects of covariates within generalized additive models. Thus, based on a pooled MS data set, we show how extending bayesian P-splines (Lang and Brezger, 2001) to mixed effects models, represents an attractive statistical approach to investigate the role of prognostic factors in affecting individual change in disability.