ABSTRACT
There has been increasing interest in discovering targeted therapy in modern drug development using genetic markers such as Single Nucleotide Polymorphisms (SNPs). Testing SNP’s predictiveness of treatment efficacy, measured by a clinical outcome, is fundamentally different from association detection for a quantitative trait such as the genome-wide association study. In targeted therapy development, clinical effect size matters, and a critical step is to identify which genetic subgroup(s) of patients should be the target of a drug. We first discuss several key statistical issues with common practice for testing SNP’s effect on treatment efficacy. Then we provide a new approach by formulating the hypotheses directly as a treatment efficacy measure through contrasts. Within an SNP, we control familywise error rate strongly by providing simultaneous confidence intervals for assessing different genetic effects (such as dominant, recessive, and additive) on the clinical response. Across all the tested SNPs, we control the per family error rate through the expected number of SNPs that have false coverage. Realistic simulations on Alzheimer’s disease are provided to demonstrate the validity of the proposed approach. Finally, we provide recommendations on how to use this confident effect method to identify and infer subgroups with differential treatment efficacy in practice.
