부산대학교 도서관

Background: When assessing the benefit of an intervention, the study population (SP) may consist of a relevant target population (ZP) and a non-relevant population (nZP). We consider the situation that a significant treatment effect is observed only in SP but not in ZP leading to the question if and how the effect in SP may be used for conclusions about the effect in ZP.Methods: We assessed three test procedures: the first increases the level of significance α for ZP (elevation rule, ER). The second procedure involves a permutation-based test for a qualitative interaction between ZP and nZP (extension rule, EWR). The third one is a modification of the EWR, which takes the relation between ZP and nZP into account. In a simulation study, we compared the empirical type 1 error and power for all three test procedures. Results: EWR unacceptably exceeds the significance level for some simulated parameter constellations (median 5.8%, maximum 15.9%). The modified version of EWR has a lower empirical type 1 error (median 5.5%, maximum 10.2%). But EWR has no advantages with respect to the empirical power and type 1 error compared to ER with an increased significance level of α=15%.Conclusion: ER, with an increased significance level of α=15%, is the appropriate procedure with respect to the empirical power, when accepting a slightly increased type 1 error (median 6.1%, maximum 10.9% over all simulated scenarios).