(OR03-06-25) Evaluating Approaches That Refine Mendelian Randomization Genetic Instruments to Improve Causal Associations Between Dietary Intake and Human Health

Objectives: Mendelian randomization (MR) uses genetic instruments (GI) to identify causality between exposures, like diet, and health outcomes. Almost all MR of dietary intake use a non-curated list of genome-wide significant (GWS) variants as the GI, and therefore, causal estimates are likely biased by variants that act indirectly on diet. We tested whether GWS-GI could detect established relationships between diet and health and whether systematically filtering the GI could better identify the direct effect of diet on health.

Methods: We utilized six literature-supported relationships between diet and biomarkers (i.e., positive controls). Positive controls included oily fish on triglycerides (TG) and cardiovascular disease (CVD), alcohol on alanine aminotransferase (ALT) and liver cirrhosis, and bread type on LDL cholesterol and CVD. GIs were obtained from publicly available GWAS. To address the misspecification of variants, we applied Steiger filtering and two phenome-wide association study-based approaches (PheWAS; one developed herein). We used two-sample MR to evaluate causality, including the inverse variance weighted (IVW) and the weighted median (WM) estimators.

Results: The WM estimator using the GWS-GIs identified the expected causal effects for oily fish on TG (ß=-0.12 [95% CI -0.16 to -0.07]) and bread type on LDL (ß=0.12 [95% CI 0.08 to 0.17]) but a false positive of bread type on height (ß =-0.12, [95% CI -0.27 to -0.09]). For all GIs and tests, an unexpected inverse relationship between alcohol and ALT was observed (GI-GWS IVW: ß=-0.02 [95% CI -0.035 to -0.008]), which became positive when accounting for BMI via multivariable MR (ß=0.034 [95% CI 0.025 to 0.044]). Filtering the GI improved the ability to detect causality for diet-biomarker but not diet-disease relationships. PheWAS-informed filtering removed pleiotropic variants that tended to be associated with other health-related biomarkers or socioeconomic factors.

Conclusions: MR studies of diet’s causal role in health have flooded the literature; however, our inconsistent associations with positive and negative controls using multiple tests and filtering methods signal a need for caution. More thoughtful curation of the GI is critical to reduce confounding due to health and environmental factors when evaluating diet’s causal effect on health.

Funding Sources: NIH NIDDK and NLM