Mendelian Randomization: A Perfect Causal Epidemiologic Approach to Simulate a Randomized Trial?

[Epidemiologic Inquiry 2006, 1: 16]

In epidemiology, we always seek to find the perfect approach to assess causation, with the aim to simulate the randomized controlled trial in observational research. One special example in which perfect randomization can be observed is the case of Mendelian randomization in genetic epidemiology.

Recently, there have been great interest in Mendelian randomization approaches for estimating causal effects of gene products. Briefly, because genetic polymorphisms are randomly assorted at conception, all covariates between individuals should theorectically be perfectly balanced between those with different polymorphisms--thus theorectically eliminating all confounding for any association between the genetic polymorphism and disease. Furthermore, if one assumes that the measured genetic polymorphisms directly affects the gene product (e.g. a biologic hormone), then one can also estimate the unconfounded association between the gene product and disease by implementing a instrumental variable analysis approach. (for more info, see Greenland, IJE 2000;29:722-729).

While such a method theorectically estimates an unconfounded Mendelian-randomized causal association between the gene product and disease, a recent article in AJE summarized the many limitations of such a method. Beside general problems in genetic studies such as population stratification and linkage disequilibrium, one subtle but important issue that Mendelian-randomization's analytic approach cannot account for is the problem of "canalization", or the post-genetic adaptation for the genetic effects by other uncontrolled factors. Examples of such a phenomenon include differential nutrient intakes to compensate for genetic deficiency, different lifestyle behaviors to adapt to obesity, or differential use of exogenous hormone agents to compensate for genetically-predisposed low levels of endogenous hormones. (FYI: causally, this can also be considered a type of time-dependent confounding for those familiar with structural DAGs).

Thus, while Mendelian randomization appears to be the perfect epidemiologic approach to directly estimate causal effects, it still has limitations and assumptions in its application, just as there are limitations of all study designs including randomized controlled trials.

For highlighting such important issues, the editors select as the Epidemiologic Inquiry Accolade: Investigation of the Week...

Limits to Causal Inference based on Mendelian Randomization: A Comparison with Randomized Controlled Trials
Nitsch et al.
Am. J. Epidemiol. 2006 163: 397-403.