Causal inference in medicine via mendelian randomization

Abstract

The central aim of preventative care is to manage or avoid entirely life-threatening and costly disease endpoints. Success in this broad goal requires researchers and clinicians to correctly distinguish between biomarkers that cause disease from those that are simply correlated with outcome. The randomized controlled trial is a scientifically valid approach to assess causal relationships, but is time-consuming and expensive, and success is not a guaranteed endpoint. Recently, a statistical approach has been translated from the econometrics literature, a strategy which utilizes genetic information identified from human subjects as “instruments” to generate an assessment of causality between biomarker and disease. This methodology, dubbed Mendelian Randomization, is directly analogous to that of the controlled trial, circumventing the issues of confounding and reverse causation that precludes conventional epidemiological studies from making causal assessments. Owing to the growing dissection of genetically heritable traits in the literature, Mendelian Randomization has emerged as a high-value tool for efficient translation of genetics research to the bedside. In the following chapter, we present the framework of Mendelian Randomization and motivation for causal assessment, the analogy of Mendelian Randomization to the randomized controlled trial, discuss general considerations for study design and assumptions of the approach, and exemplify case studies from the literature of applications of MR to type 2 diabetes and other clinical endpoints.