Case‐only trees and random forests for exploring genotype‐specific treatment effects in randomized clinical trials with dichotomous end points

摘要: Discovering gene-treatment interactions in clinical trials is of rising interest the era precision medicine. Nonparametric statistical learning methods such as trees and random forests are useful tools for building prediction rules. In this article, we introduce to recently proposed case-only approach discovering estimating marker-specific treatment effects a dichotomous trial endpoints. The motivational example case-control genetic association study Prostate Cancer Prevention Trial (PCPT), which tested hypothesis whether finasteride can prevent prostate cancer. We compare novel interaction tree method previously proposed. Because modeling simplicity - directly targeting at efficiency approach, yield more accurate heterogeneous better measure variable importance, relative uses data from both cases controls. Application PCPT yielded discovery genotypes that may influence prevention effect finasteride.