In medicine, many investigators use observational studies as a starting point to uncover causal effects of risk factors on health outcomes. Mendelian Randomization (MR) is one popular approach to draw better causal conclusions from observational data. In a nutshell, MR is a "big data" approach based on methods from economics. Specifically, MR adapts instrumental variables (IV) to large observational genome-wide association studies (GWAS) to determine causal mechanisms. Specialized MR methods and computer software have been developed leading to the widespread use of the approach, but to date little is understood about the statistical properties of these methods. For example, when do MR methods work well (and when do they not)? Other fundamental questions about the underlying causal mechanisms are also poorly understood. This project will investigate the statistical theory and methods behind MR so that investigators can properly use MR to draw causal conclusions from observational data, leading to better medical and health decisions.
The MR approach requires finding genetic variants in GWAS, called instruments, that satisfy the following assumptions: (A1) Instruments are associated with the risk factor, (A2) Instruments have no direct effect on the outcome, and (A3) Instruments are unconfounded by variables affecting both the outcome and the risk factor. Specific aims that will be investigated include (1) laying the theoretical foundations for MR including identification and statistical sufficiency, (2) studying the statistical properties, including efficiency, power, and robustness, of popular MR methods, such as the median method or the MR-Egger method, (3) proposing new MR methods that are robust to common problems associated with MR data, most notably violations of the three core assumptions (A1)-(A3), (4) developing numerical and visual diagnostic tools for assumptions underlying MR, and (5) producing easy-to-use software suites.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
