Multiple Imputation of Heaped Longitudinal Cigarette Count Data

• Activity Number: 614
• Type: Invited
• Date/Time: Thursday, August 8, 2013 : 8:30 AM to 10:20 AM
• Sponsor: ENAR
• Author(s): Sandra D. Griffith*+ and Saul Shiffman and Yimei Li and Daniel F. Heitjan
• Companies: Cleveland Clinic and University of Pittsburgh and The Childrens' Hospital of Philadelphia and University of Pennsylvania
• Keywords: imputation ; missing data ; smoking cessation ; clinical trials ; heaping ; measurement error

Abstract

Although smoking cessation trials commonly collect daily cigarette counts, they seldom make use of these data, instead designating smoking status at the end of treatment as the primary analysis variable. The daily counts hold the promise of much more efficient, detailed and informative analyses, but are beset by serious quality issues - specifically, measurement error and heaping. These effects can induce both bias and loss of efficiency. We describe an attempt to create a correct and efficient analysis of the heaped daily count data from a smoking cessation trial. First, we use data from another trial, in which both precise and heaped counts were collected, to construct a model for predicting true counts from heaped counts and baseline factors. Then we apply the model to the clinical trial data to create imputed precise daily counts. Finally, we analyze the imputed data in a complex longitudinal model that accommodates random subject effects and zero inflation.