Stat Colloquium: Dr. Ruizhe Chen
Johns Hopkins University
TITLE: Multivariate Zero-Inflated Generalized Poisson Data Generation Methods for Simulating Counts of Adverse Events
ABSTRACT: Counts of maximum grade adverse events collected in clinical trials are important measurements of treatments' toxicity and tolerability. Studying the frequencies and correlations of adverse event counts by types, treatment cycles, and grades can provide further insights into the toxicity profiles of the underlying treatments. A prerequisite to establish such statistical inferential methods is the ability to properly generate multivariate count data with designated event rates and correlation structures. In this talk, we present methods for simulating multivariate count data that follow zero-inflated generalized Poisson (ZIGP) distributions. The proposed methods are developed based on the NOrmal-To-Anything (NORTA) and Sample-Iterate (SI) data simulation frameworks. In particular, we have adapted the NORTA with correlation adjustment by polynomial regression approach to the case of ZIGP distributed marginals. Our simulation study results show great performance of the proposed approaches in simulating ZIGP distributed count data with desired rate, scale, proportion of zeros, and correlation matrices. We apply the proposed methods in simulating AE counts based on the NCCTG Study N9741, a randomized multicenter phase III colorectal cancer study. The presented method can also enjoy a broader applicability where we showcase a scenario in simulating counts of hospital visits based on a National Medical Expenditure Survey dataset.