Abstract:
|
Missing data, occurring in either outcome variables or covariates or both, are a common occurrence in cluster randomized trials (CRTs). It is important to handle missing data properly in the analysis to ensure that valid inferences are obtained. It has been widely recognized that imputation strategies which take into account the intra-cluster correlation should be used to reflect the homogeneity within clusters to handle the missing continuous or binary outcomes in the context of CRTs. However, little investigation has been done for handling the missing covariates in CRTs though missing covariates can lead to biased results as well. Based on empirical and simulated data, the performance of listwise deletion method, standard multiple imputation (MI) assuming data are independent, and multilevel MI accounting for the homogeneity within clusters was compared. Bias in the treatment effect estimate and the corresponding standard error, and the coverage probability were used as the evaluation criteria for assessing the performance of these strategies. Guidelines are provided to facilitate the use of appropriate method to handle the missing continuous covariates in practice.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.