Assessing if randomized treatment group should be included in the imputation model when imputing missing outcome data in randomized superiority clinical trials

Regulatory agencies overseeing the development of new medical devices, pharmaceuticals, and biologics, such as the United States Food and Drug Administration (FDA), primarily require the primary efficacy analysis in a drug or device clinical trial to be performed on the Intent-to-Treat (ITT) sample, usually defined as all randomized patients. Since the ITT sample often includes patients who prematurely withdrew from the study before collection of the primary endpoint, imputation of missing primary endpoint values is required for patients in the ITT sample who did not have the primary endpoint measured. Currently, there are no specific regulatory guidelines on the exact missing data imputation method to use when analyzing clinical trial data.;In this study, we examine methods of imputing missing primary outcome data in a 1:1 superiority randomized clinical trial where the outcome is scheduled to be measured once post-baseline. To impute missing primary outcome data, we concentrate on multiple imputation for continuous outcomes using Markov Chain Monte Carlo and linear regression imputation methods, and on multiple imputation for dichotomous outcomes using logistic regression. The primary objective of this study is to assess whether randomized treatment group should be included as a covariate in the imputation model for the primary outcome in a clinical trial. Using simulations and statistical theory, we found that randomized treatment group, when included as a covariate in the imputation model, maintains the alpha level and power of the subsequent statistical test comparing randomized treatment on the outcome mean (continuous outcome) and outcome proportions (dichotomous outcome); otherwise alpha may be lower than the nominal alpha and the power of the test may be insufficient.;Therefore, we recommend that, prior to analyzing treatment effect on a continuous or dichotomous primary outcome scheduled to be measured at one time point post-baseline in a 1:1 superiority randomized clinical trial, randomized treatment group be included as a covariate in multiple-imputation models used to impute the missing primary outcome data.