Title
Power Analysis for Clinical Trials Using Negative Binomial Models with Application to Multiple Sclerosis Count Data
Speaker
Mallik Rettiganti, The Ohio State University
Abstract
A key variable of interest in the study of the disease progression of relapsing remitting multiple sclerosis (RRMS) is the magnetic resonance imaging (MRI) based brain lesion count. Univariate and bivariate negative binomial distribution (NB and BNB) can satisfactorily model such count data.
For RRMS Parallel Group (PG) and Baseline versus Treatment (BVT) trials, Nauta et al. (1994) and Sormani et al. (2001) propose the use of nonparametric tests and do power analyses to compare the mean MRI counts. For the PG trials, Aban et al. (2009) have proposed approximate likelihood-based tests such as the likelihood ratio test (LRT), Rao's score test (RST) and Wald tests that use asymptotic chi-square critical values.
These tests work well for large sample sizes but fail to maintain the desired significance level for small sample sizes. In this talk we describe exact parametric tests to compare the group means for the PG trials using the NB model. Using the BNB model we develop parametric tests to compare means of the baseline and treatment periods. Using these tests we perform power analyses and sample size estimation for the PG and BVT trials. We also carry out a robustness study for the PG trial. Generally, the LRT and one of the Wald tests with exact critical values perform well.
