Missing Data Analysis: A Case Study of a Randomized Controlled Trial

Abstract

Missing data is a pervasive problem in the analysis of many clinical trials. In order for the analysis of a study to produce unbiased estimators, the missing data problem must be addressed. First, the missing data pattern must be established; second, the missingness mechanism must be determined; and third, the most appropriate imputation method for imputing the missing values must be found. The purpose of this paper is to explore the imputation methods best suited for the missing data from the Diet and Exercise for Elevated Risk Trial (DEER) in a secondary analysis of the data. The missingness pattern in the data set is arbitrary and the missingness mechanism is MAR. A simulation study suggests that the two best methods for imputation are subject-specific mean imputation and multiple imputation. I conclude that mean imputation is the best method for handling missing data in the DEER data set.