Abstract:
|
In clinical trials, stratification is often used to maintain the balance of treatment allocation with respect to the important prognostic factors. Based on FDA ICH E9 guidance, the stratification factors used during the randomization should be adjusted in the statistical analysis model. There are some occasions that patients are inadvertently stratified into the wrong stratum during the randomization process. For example, in diabetic study, patients are often stratified by their baseline HbA1c values, e.g., HbA1c < 9% and HbA1c >= 9%, and then randomized into the treatment groups. Patients can be mis-stratified when there are some wrong readings of the lab tests. When a subject is mis-stratified, it raises the question of whether we should use the stratum level selected in the randomization or the true stratum level to which the subject belonged in the statistical analysis model. The results from our simulation study demonstrate that adjusting for the strata levels selected in the randomization in the statistical model may give biased treatment effect estimators when the treatment allocation with respect to the stratitication factor is imbalanced.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.