Mendelian randomization, lipids and coronary artery disease: trade-offs between study designs and assumptions.

Abstract

Mendelian randomization (MR) studies have been described as naturally occurring randomized trials. However, MR studies often deviate from appropriate trial design principles, and many use two-sample designs with strong assumptions not required in randomized trials. Using data from the Million Veteran Program, we empirically evaluated the impact of study design choices and two-sample approaches in an MR study of lipids and coronary artery disease. We designed an MR study of participants of European descent with no history of coronary artery disease and no contraindications to low-density lipoprotein cholesterol (LDL-C)-related therapies to estimate 10-year odds ratios of coronary artery disease per 39 mg/dL increase in LDL-C or 15.8 mg/dL increase in high-density lipoprotein cholesterol (HDL-C). We then sequentially modified the design to reflect common decisions made in MR studies. For LDL-C, one-sample estimates ranged from 1.50 (95% CI: 1.34, 1.68) to 2.23 (95% CI: 1.93, 2.59) and two-sample estimates from 1.13 (95% CI: 1.01, 1.26) to 1.30 (95% CI: 1.15, 1.46). For HDL-C, estimates ranged from 0.76 (95% CI: 0.68, 0.86) to 0.93 (95% CI: 0.65, 1.34) across one- vs two-sample analyses. Estimates were most sensitive to the inclusion of prevalent outcomes. These results indicate the magnitude of MR estimates can vary with study design. We recommend future MR studies assess the sensitivity of estimates to different design decisions.