Speaker: Yumi Kondo
Title: A Flexible Mixed Effect Negative Binomial Regression Model for Detecting Abrupt Increases in MRI Lesion Counts in Multiple Sclerosis Patients
Abstract: An increase of contrast enhancing lesions (CELs) on repeated magnetic resonance imaging has been used as an indicator for potential adverse events in multiple sclerosis (MS) clinical trials. The objective of this work is to enhance a previously proposed method (Zhao et al. (2013)) for identifying unusual increases in new CEL activities for individual patients. The procedure signals such patients by estimating the probability of observing lesion counts as large as those observed on the recent scans of a patient conditional on the patient's lesion counts on previous scans. This conditional probability is computed based on a mixed effect negative binomial regression model. As the values of the conditional probability can vary substantially depending on the choice of the random effect distribution, we relax this restrictive parametric assumption to model the random effects with an infinite mixture of beta distributions, using the Dirichlet process, which effectively allows any form of distribution. As our inference is in the Bayesian framework, we discuss how to develop an informative prior based on previous clinical trials. This could be particularly helpful at the early stages of trials when little data is available. Our simulation study shows that our procedure estimates the conditional probability index more accurately than previous parametric alternatives and an informative prior for the mixed effects improves the accuracy of the conditional probability estimates. This enhanced method is illustrated with CEL data from 10 previous MS clinical trials.