Mendelian randomization is an approach to causal inference that uses genetic variants as instrumental variables to estimate causal effects. Although often applied with a single exposure and outcome Mendelian randomization has been extended to consider models including multiple traits. The estimates obtained are usually interpreted as ‘lifetime’ effects of the exposure on the outcome. In this talk I will consider this definition and what it means in applications of Mendelian randomization in two settings; when the exposure(s) are time varying and when there are multiple potential exposure traits which have a highly correlated genetic structure.