Estimation of Treatment Difference Following a Sequential Clinical Trial
通过模拟比较了序贯试验后使用Whitehead序贯估计量与普通非序贯最大似然估计的效果，发现非序贯估计量在多数情况下与序贯估计量相当甚至更优。
Abstract To reduce the number of patients needed to reach a conclusion in a clinical trial, a sequential trial design may be used. In theory the distribution of responses is biased when a sequential trial is stopped. The usual estimation methods of treatment difference applied in nonsequential situations are, therefore, not necessarily applicable following a sequential test. Whitehead (1983) proposed a method for estimating the treatment effect following some types of two-sample sequential tests. By stochastic simulation we compared this method with an ordinary maximum likelihood estimator for nonsequential two-sample situations, following two different sequential tests. The results show that there is little difference between the methods, although Whitehead's method was developed especially for sequential plans. A correction of the bias of the maximum likelihood estimate proposed by Cox (1952), however, gives results that are much closer to the expected values. We also investigated the estimation of treatment difference following a two-sample sequential Wilcoxon test proposed by Skovlund and Walløe (1988). The (nonsequential) estimation methods applied following this test give good results over a range of distributions and for various real treatment differences. This article advocates a class of nonsequential estimators that has generally been considered inappropriate for sequential trials. Simulations demonstrate that these nonsequential estimators are as good as (and in some situations better than) the sequential estimator developed by Whitehead.