On Errors in Stratified Randomization

Stratified randomization is widely used in clinical trials to achieve balance of the treatment assignment with regard to important prognostic factors. Sometimes the prognostic or stratification factor is concerned with patient medical and medication history, for which accurate information may not be available at the time of randomization. Therefore, the value of the stratification factor provided at randomization may later be identified to be incorrect. When such errors occur, patients may be allocated into strata they do not belong to. The intended treatment balance regarding the stratification factors may be disrupted. We evaluated the impact of such errors in stratification on the intended balance. We further investigated the analysis options in presence of stratification errors under both randomization and population models. Some simulations were performed to evaluate the treatment balance and properties of analysis approaches. We also provided experience with some real clinical trials.