Prediction of Individual Long-Term Outcomes from a Multivariate Cure-Mixture Frailty Model

• Activity Number: 219
• Type: Invited
• Date/Time: Monday, August 2, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Society for Research on Nicotine and Tobacco
• Author(s): Yimei Li and Daniel F. Heitjan*+ and E. Paul Wileyto
• Companies: University of Pennsylvania and University of Pennsylvania and University of Pennsylvania
• Keywords: prediction ; Bayesian inference ; cure model ; frailty model

Abstract

In smoking cessation trials, subjects usually receive treatment for several weeks, with additional information collected 6 or 12 month after that. An important question concerns predicting long-term cessation success based on short-term clinical observations. But several features need to be considered. First, subjects commonly transit several times between lapse and recovery, during which they exhibit both temporary and permanent quits, and both brief and long-term lapses. Second, although we have some reliable predictors of outcome, there is also substantial heterogeneity in the data. We therefore introduce a cure-mixture frailty model that describes the complex process of transitions between abstinence and smoking. Then based on this model, we propose a Bayesian approach to predict individual future outcomes. We will compare predictions from our model to a variety of ad hoc methods.