Confidence Intervals for Proportion Difference Based on Paired and Unpaired Binary Data

• Activity Number: 581
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Yue Shentu*+ and Yabing Mai and Jiajun Liu
• Companies: Merck & Co., Inc. and Merck & Co., Inc. and Merck & Co., Inc.
• Keywords: paired binary data ; confidence interval ; missing data ; proportion difference ; method of variance estimates recovery

Abstract

In clinical trials, paired binary data often arise from crossover study design or pre-test/post-test comparisons. Missing data due to drop-outs and other reasons may lead to incomplete paired binary data for a subgroup of subjects. Interval estimation for the proportion difference can be problematic in these situations. In this article we propose an extension of the method of variance of estimates recovery (MOVER) to construct confidence intervals (CIs) for the correlated proportion difference based on paired and unpaired data. Two sets of CI estimators, one based on paired data, the other based on pooled paired and unpaired data, are utilized in the double-MOVER procedure to construct the asymptotic CI. Extensive simulations show that the double-MOVER estimator performs well under various degrees of missingness and correlations, even with small to moderate sample sizes.