Abstract

Meta-analysis is used to combine results from multiple studies to produce a more definitive answer to a question of interest than possible with a single study. Individual patient data meta-analysis combines the subject-level data from multiple studies in a statistical analysis. To do so it is necessary to obtain, process, and harmonize data from the various studies. Having the subject-level data allows one to address a broader range of questions than would be possible with only aggregate data based on published results. When one or more studies cannot or will not provide individual level data, it is proposed that synthetic data be created to represent the missing information. The novel approach of multiplying imputing missing individual data based on aggregate statistics via hot deck imputation and statistical modeling is presented. Methods are applied in a study of antibiotic treatment of bacterial vaginosis to prevent preterm delivery and studied through simulation.