Name: John Lawrence
Institution: Food & Drug Administration
Street: 12119 amber ridge circle
Phone: 301 594-5375
Fax: 301 594-5494
Authors: John Lawrence
Title: Changing the test statistic after an interim analysis
Abstract: Often, there are a wide variety of test statistics that can be used to test a specific set of hypotheses. For example, there are many different statistics that can be used to test for equality of the distributions of a time-to-event endpoint, e.g. the logrank or the Gehan-Wilcoxon statistic. After an interim look at the data, the investigator may decide that a different test statistic would be more powerful. But, changing the test statistic can inflate the type I error rate. In this talk, I will discuss a strategy for changing the test statistic that maintains the correct type I error rate. A simulation study demonstrates that this strategy can result in a procedure that has higher power than the test without adaptation.