Comparison of Proportions of Composite Outcomes with Missing Values in their Components

• Activity Number: 233
• Type: Contributed
• Date/Time: Monday, August 3, 2009 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Xianbin Li*+ and Brian S. Caffo
• Companies: FDA and Johns Hopkins Bloomberg School of Public Health
• Keywords: Composite outcomes ; Missing values ; MLE ; Sample size ; Statistical power ; two-sample comparison

Abstract

Composite outcomes are commonly used in medical diagnoses, epidemiological research, and clinical trials. When missing values in their components exist, the method of handling missing data can have unintended consequences, even if the components are missing completely at random. Maximum likelihood estimators (MLEs) have been developed to estimate the proportion of success in composite outcomes with components missing completely at random. The currently available statistical method for comparative studies is the comparison of the logarithm of proportion ratio of composite outcomes. In this paper, we use the MLEs to compare two groups in the difference in proportion, relative risk, and logarithm of relative risk. Sample size and statistical power formulas have been developed. Simulation studies will be used to evaluate the performance of the developed methods.