Inferences on Zero-Inflated Count Data

• Activity Number: 626
• Type: Contributed
• Date/Time: Thursday, August 4, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Biopharmaceutical Section
• Author(s): Madhuja Mallick*+ and Liang Chen
• Companies: Merck Research Laboratories and Merck Research Laboratories

Abstract

In many areas of research, experimental data are collected and analyzed as counts that are referred to as the number of times an event occurs. For example, in clinical trials, endpoints based on count data are most common in trials designed to study chronic diseases where the counts are done within an interval in time, usually resulting in a number 0 or 1 (e.g., survival). Some common examples of count data include the number of heart attacks, the number of hospitalizations, number of cigarette smoked etc. Poisson regression is a general framework for analyzing count data. However, count data with excessive zeros relative to a Poisson distribution could result in an over-dispersion of the data. Common approaches for addressing over-dispersion include using negative binomial, zero-inflated Poisson, and zero-inflated negative binomial models. This presentation will discuss the inferences of different candidate models for handling zero-inflated count data using either simulated or clinical trial data.