Mixed Models for Repeated Zero-Inflated Counts from Smoking Cessation Data

• Activity Number: 360
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): E. Paul Wileyto*+ and Yimei Li and Daniel F. Heitjan
• Companies: University of Pennsylvania and University of Pennsylvania and University of Pennsylvania
• Keywords: Latent Mixture ; zero-inflated poisson ; zero-inflated negative binomial ; Mixed Models ; Ramdom Effects

Abstract

Raw Time-Line Follow-Back data from smoking cessation trials consist of daily counts of cigarettes smoked. Because complete abstinence is the goal, these counts tend to have an over-abundance of zeros, and are best represented as a two-component latent mixture. We used SAS NLMIXED to fit zero-inflated Poisson and negative binomial count models to data from a smoking cessation trial of the drug Bupropion, with a zero-inflated component represented by logistic regression. The model included a variance term for the negative binomial part, another for the logistic part, and it included a covariance term describing the relationship between the two variance components. The estimated covariance term was positive, and implied that those who were more likely to be abstinent at any given time were also smoking fewer cigarettes when they did smoke.