The authors reply

摘要

We thank Liublinska and Rubin (1) for their interest in our article on missing outcome data in randomized trials and observational research (2). In medical research, different methods are being used to handle missing data. In our article, we illustrated that in some settings, multiple imputation and complete case analysis with covariate adjustment (CCA-CA) yield the same unbiased results. We evaluated settings in which multiple imputation and CCA-CA utilized exactly the same amount of information on covariates or predictors of missingness. It therefore does not come as a surprise that results were the same for the 2 methods, as Liublinska and Rubin rightfully point out. Furthermore, we focused on situations in which only outcome data were missing (i.e., no missing data on covariates or treatment status), and these missing outcome data were also missing at random, meaning that missingness depended on observed variables only. We showed that in cases conditional on these observed variables (e.g., after covariate adjustment), missingness is completely at random and a CCA-CA thus yields unbiased estimates, similar to those from multiple imputation.