Adjusted significance levels for subgroup analyses in clinical trials

Subgroup analyses in clinical trials are becoming increasingly important. In cancer research more and more targeted therapies are explored and probably only a portion of the whole population will benefit from them. Subgroups of interest can be analyzed in several ways, but a correction of the type I error probability is needed in order to appropriately draw conclusions. Often a conservative Bonferroni approach is taken where the total significance level is distributed (equally or unequally) over the analysis including all patients (overall analysis) and the subgroup analysis. However, more efficient methods are available that take into account the correlation that exists between the test statistics for the overall and the subgroup analysis. The latter approaches are very appealing but have not found their way into practice. The aim of this paper is to show that these methods are the same as the methods used when dealing with interim analyses, i.e., group sequential methods, and hence standard software can be used to calculate the appropriate significance levels. Further, we show that this correction can be applied even when the size of the subgroup is unknown until the end of the trial. Using a simulation study with survival data, we also show that the familywise error rate is well controlled, even with small sample sizes. We hope that this will promote the use of these methods in future cancer clinical trials.