Truncated Cauchy Combination Test: a Robust and Powerful P-value Combination Method with Arbitrary Correlations

Abstract: Cauchy combination test has been widely used for combining correlated p-values, but it may fail to work under certain scenarios. We propose a truncated Cauchy combination test (TCCT) which focus on combining p-values with arbitrary correlations, and demonstrate that our proposed test solves the limitations of Cauchy combination test and always has higher power. We prove that the tail probability of our test statistic is asymptotically Cauchy distributed, so it is computationally effective to achieve the combined p-value using our proposed TCCT. We show by simulation that our proposed test has accurate type I error rates, and maintain high power when Cauchy combination test fails to work. We finally perform application studies to illustrate the usefulness of our proposed test on GWAS and microbiome sequencing data.