Robustness of GLP1R-Region Mendelian Randomization Estimates for Heart Failure

Background/Objectives: We assessed the robustness of GLP1R-region Mendelian randomization estimates for heart failure (HF) across instruments, outcome sources and uncertainty models. Methods: HbA1c associations from GCST90014006 were combined with Henry, HERMES and European-ancestry GBMI HF data. Eleven-, seven- and four-variant instruments were evaluated using correlated-variant generalized least squares (GLS), alternative linkage disequilibrium (LD), influence analysis, type 2 diabetes calibration and exposure-error sensitivity analyses. Effects represented a one-unit decrease in inverse-rank-normal-transformed HbA1c. Results: The principal Henry 11-variant multiplicative random-effects estimate was an approximate relative risk of 0.568 (95% confidence interval 0.299–1.080), with Q = 29.998 on 10 degrees of freedom. The fixed-effect interval was 0.392–0.823; a t-reference interval was 0.274–1.179. Exposure-error profile likelihood gave 0.511 (0.233–1.008), with residual heterogeneity supported by simulation. The four-variant GLS estimate was 0.909 (0.574–1.439). Eleven-variant odds ratios were 0.369 (0.206–0.659) in HERMES and 0.528 (0.317–0.880) in GBMI. Nine available variants gave a type 2 diabetes odds ratio of 0.204 (0.112–0.370). Conclusions: Protective point estimates varied with outcome source and instrument composition. Heterogeneity-aware uncertainty remained substantial after exposure-error sensitivity analysis. Glycaemic calibration and regional evidence require interpretation alongside variant-specific function when assessing pharmacological GLP-1R agonism.