|
Abstract:
|
Clinical trials sometimes contain two or more binary endpoints, and a simultaneous estimation of multiple response rates is required. In practice, some objectives are inevaluable to one or more responses due to uncontrollable factors, such as insufficient cell quantity in samples for certain laboratory test. To estimate a response rate, the standard method is to compute the proportion of responders among all subjects who are evaluable for the corresponding response. Under the assumption of non-informative missing, we develop an estimation method based on the self-consistent principle. Our simulation studies show that our method is more efficient than the standard method when the binary endpoints are correlated and the number of missing observation is substantial. To compare our method to the standard method in large sample cases, the relative asymptotic efficiency is developed.
|
