Evaluation and Application of a Test for Treatment-Biomarker Interaction Effects Using Probabilistic Indices
MetadataShow full item record
A predictive biomarker is a patient characteristic which associates with different effects of the treatment on the patient. In clinical trials, it is often of interest to assess whether the effects of the treatment differ depending on the level of the biomarker. A common method to assess the treatment-biomarker interaction effect is to use Cox proportional hazard model. However, this method might not be appropriate when the proportional hazard assumption is not satisfied. Jiang, Chen and Tu (2016) developed a non-parametric test for the treatment-biomarker interaction based on Efron's estimator of probabilistic indices, which is free of any assumptions. This report is mainly a review of their proposed method and will focus on determining its test size on simulated data. Various simulation studies demonstrated that the method based on Efron's estimator performs well when the censoring rates are low. That is, the produced test sizes were closer to the nominal test size. Other findings from the simulation studies will lead to discussions on why the method performs badly when the censoring rate is high, and how to dichotomize the biomarker when it is continuous based on some cut point. The usefulness of these ideas in practice are presented by applying them to sets of clinical trial data.