Introduction

In the previous chapters, we repeatedly used the word 'causal' to describe the inferences obtained by Mendelian randomization. In this chapter, we clarify what is meant by the causal effect of an exposure on an outcome. We give a more detailed explanation of the theory of instrumental variables, and explain in biological terms various situations that may lead to violations of the instrumental variable assumptions and thus misleading causal inferences. We conclude by discussing the difference between testing for the presence of a causal relationship and estimating a causal effect, and the additional assumptions necessary for causal effect estimation.

Conclusions

The instrumental variable assumptions make assessment of causation in an observational setting possible without complete knowledge of all the confounders of the exposure–outcome association. Genetic variants have good theoretical and empirical plausibility for use as instrumental variables in general, but the instrumental variable assumptions may be violated for a number of reasons.

Relevant paper to chapter:

Section 3.2.6 (Assessing the IV assumptions). S. Burgess. Reply to Glymour, Tchetgen Tchetgen and Robins: Statistical testing vs. scientific judgment in assessing instrumental variable assumptions for Mendelian randomization (letter). Am. J. Epidemiol. 2012; 176(5):456-457.