Model-Based Analysis of Heaped Longitudinal Cigarette Count Data in Smoking Cessation Trials

• Activity Number: 279
• Type: Invited
• Date/Time: Tuesday, August 2, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Daniel Heitjan*+ and Sandra D. Griffith and Yimei Li and Hao Wang and E. Paul Wileyto
• Companies: University of Pennsylvania and University of Pennsylvania and Children's Hospital of Philadelphia and The Johns Hopkins University and University of Pennsylvania
• Keywords: smoking cessation ; longitudinal ; software ; imputation ; count data ; calibration

Abstract

In smoking cessation trials it is common to collect daily cigarette counts, although primary analyses use coarser outcomes such as smoking status at end of treatment. Proper modeling of the daily counts could give more efficient and detailed analyses, allowing the estimation of treatment effects on mean cigarette consumption in non-quitters and the time dependence of treatment effects. Unfortunately, cigarette count data are often heaped, in the sense of being reported rounded to multiples of five, ten, or twenty. Heaping can substantially bias analyses of mean count, an intractable problem unless one has a valid model to predict heaped from true counts. We are in possession of a dataset where cigarette counts were measured by both electronic diaries (not subject to heaping) and conventional recall (strongly heaped). We will use these data to create a model to impute accurate cigarette counts from the recall data in a clinical trial. We will fit zero-inflated longitudinal Poisson models to the imputed accurate data, implementing the procedure in a mixture of R and SAS software.