Improving Mendelian Randomization Analysis by Instrument Borrowing from Auxiliary Outcome Traits

Abstract

Mendelian randomization (MR) is a widely used approach for inferring causal effects of exposures on outcomes using genetic variants as instrumental variables; however, existing methods remain vulnerable to bias and/or loss of power in the presence of invalid instruments. We hypothesize that closely related outcome traits are likely to have a large overlap in underlying valid instruments in relation to a given exposure and that instrument borrowing (IB) across such traits can therefore yield more robust MR inference. To operationalize this idea, we first introduce a novel coheterogeneity statistic and its asymptotic theory under different paradigms which can be used to identify secondary outcome traits that are likely to share valid instruments with the primary trait. We then propose extensions of two popular methods, MR-Mode and MR-PRESSO, that improve MR inference for the primary trait, taking advantage of a secondary trait. Extensive simulation studies demonstrate that the proposed coheterogeneity statistic has expected finite-sample properties and IB-based methods consistently outperform their standard counterparts, with gains increasing as the degree of overlap in valid instruments across outcome traits grows. Applications to established positive and negative control hypotheses suggest that IB-based methods offer improved control of type I error and increased power. We further illustrate the practical utility of these methods by revisiting the long-debated hypothesis on the causal effect of vitamin D on cardiometabolic traits.