![time point measure time point measure](https://imageproxy.ifunny.co/crop:x-20,resize:640x,quality:90x75/images/68ee011ba1e2938a4b97f313b3ddec572ecb210972d49cb2989a320ab98c82f5_1.jpg)
![time point measure time point measure](https://image.slidesharecdn.com/pointoftimeandperiodoftime-141031120252-conversion-gate01/95/point-of-time-and-period-of-time-3-638.jpg)
24 More recently, it has been questioned how appropriate MR is when applied to exposures that vary over time. 1, 23 Many single-nucleotide polymorphism (SNP)-phenotype associations have a consistent direction of effect and similar effect sizes throughout the lifecourse, though this pattern is not uniform. It has long been recognized that MR estimates relate to exposures that generally act over a considerable period of time, often since birth. 18, 22 MR applied to one measure of an exposure that varies over time Time-varying genetic associations have been reported for a range of phenotypes, 15–21 thus consistent effect sizes may not be estimated when applying MR to exposures measured at different time points across the lifecourse, regardless of sampling variation and measurement error. height in adulthood 12), longitudinal within-individual phenotypic variability (BMI 13), monotonic change (myopia 14) or, likely, a mixture of these. Within-individual variation may be largely a function of measurement error (e.g. Many exposures of interest vary over time, 10, 11 being subject to both between-individual and within-individual variation. MR studies have largely leveraged information from a single measurement of the exposure and outcome, often due to limited availability of repeatedly measured data. 5, 6 In order to estimate an average treatment effect, extra assumptions are needed: first, that a difference in average exposure between populations with and without a risk allele would result in the same difference in outcome as if an environmental factor increased average exposure in a population by the same amount (gene–environment equivalence) and, second, that the structural model relating the exposure and outcome is linear and additive with a homogeneous effect of the exposure on the outcome. These are (i) relevance: that genotype is associated with the exposure of interest (ii) independence: that there is no common cause of genotype and outcome (iii) exclusion: that genotype does not affect the outcome through any path other than the exposure. Three assumptions are required for MR analyses to test the null hypothesis that an exposure X does not cause an outcome Y for any individuals. 2 Reverse and residual confounding is reduced because the formation of the genotype occurs prior to the phenotypic development and is generally not related to environmental factors. 1 MR is generally implemented within an instrumental variables (IV) framework that exploits the randomization inherent in the allocation of genotypes at conception and gamete cell formation, using this random variation in alleles to instrument differences in observed exposures between individuals. Mendelian randomization (MR) is a powerful tool through which the causal effects of modifiable exposures (risk factors) can be estimated from observational data under assumptions that in some circumstances may be more plausible than the unmeasured confounding and no measurement error assumptions required by conventional methods. MR results from time-varying exposures should be interpreted with respect to the underlying liability for an exposure. MR estimates the causal effect of changing the exposure liability that gives rise to an exposure value at a given time. MR does not estimate direct or total causal effect of changing an exposure value at a given time. Many exposures of interest vary over time, yet Mendelian randomization (MR) is commonly applied to one measurement of an exposure.