ランダム化臨床試験におけるケースオンリー解析-治療効果予測マーカーの検出力評価-

抄録

The case-only approach is widely known as a resource-efficient approach to identify biomarkers that interact with a treatment. It measures stored baseline samples from trial participants who experience the outcome. The previously proposed case-only approach for binary outcomes is most commonly used in early-stage, retrospective profiling experiments for selecting biomarkers. In this article, we compare power and alpha error for assessment of interaction between a treatment and a baseline marker from the case-only approach to those from the full-cohort approach in which all the baseline samples are measured. We calculated power and alpha error based on a test for interaction in the case-only approach and then compared the power and alpha error of the test for interaction in the case-only approach and the full-cohort approach. We showed that the case-only approach can get 80% statistical power under the scenarios of the risk ratio=0.5, probability of outcome event=0.5-0.7, marker expression probability=0.3-0.7 and case sample size>300. We could get enough power in full-cohort approach in the RR=0.5-0.7, while the case-only approach gave less than 80% power. We revealed that the case-only approach can get 80% power and situations that the case-only approach or the full-cohort approach should be used.