Parametric Models for MRI Count Data in Multiple Sclerosis Clinical Trials: A Robustness Study

• Activity Number: 471
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Mallikarjuna Rettiganti*+ and Haikady Nagaraja
• Companies: University of Arkansas for Medical Sciences and The Ohio State University
• Keywords: multiple sclerosis ; negative binomial ; Poisson-inverse gaussian ; likelihood ratio, score, wald ; robustness ; type I error

Abstract

Magnetic resonance imaging (MRI) lesion count data in multiple sclerosis (MS) clinical trials tend to be over dispersed with respect to the Poisson distribution. Mixed-Poisson models such as negative binomial (NB) distribution and the Poisson-Inverse Gaussian (P-IG) distribution have been shown to fit well to such count data. Previous studies have shown that when distributional assumptions are satisfied, parametric tests have higher power to detect a significant treatment effect than nonparametric tests. However, robustness of these tests to violations of distributional assumptions has not been investigated before. We conducted a detailed robustness study using simulations to evaluate the performance of asymptotic and exact parametric tests such as the likelihood ratio test, Rao's score test and several Wald tests (WT) to these violations. We considered two scenarios: i) True distribution is NB; assumption is P-IG and ii) True distribution is P-IG; assumption is NB. Overall, for case i), one of the exact WTs has the highest power while maintaining below nominal Type I error rates. For case ii), the exact WT has the least Type I error among others and had reasonably high power.