Diagram adapted from Relton & Davey Smith, Two-step epigenetic Mendelian randomization: a strategy for establishing the causal role of epigenetic processes in pathways to disease, International Journal of Epidemiology, 2012, 41, 161–176. Felix As a result the one-sample MR effect estimate will be an underestimate of the true causal effect 10, • Using two non-overlapping samples avoids this. Two Mendelian randomization studies have applied novel approaches to instrumental variable selection in methylation data, identifying bidirectional causal effects of CPT1A and triglycerides, as well as of RNMT and C6orf42, on high-density lipoprotein cholesterol response to fenofibrate. . DO When using summary GWAS data in what might be considered to be true two-sample MR, it is possible that the two samples overlap because of some cohort studies contributing to both GWAS (for example many adult cohort studies have contributed both to GWAS of adiposity measurements and also of disease outcomes such as CHD and type 2 diabetes). comment on the ‘strong’ assumptions of MR, but rarely do we see such statements about the equally strong, and untestable, assumptions of conventional multivariable regression analyses. These are directed acyclic graphs (DAGs), thus the absence of an arrow between any two variables (nodes) indicates we do not consider it plausible that there is a causal effect between those two. . Risk of Prostate Cancer Incidence among Atomic Bomb Survivors: 1958-2009. This may be due to the winner’s curse because in the stage 1 we rank all 172 estimates of the allele score on the outcomes, such that the highest ranked are more likely to be higher than their respective true values because of the random variation of these sample estimates about their true values. B A copy of the book "Mendelian randomization: Methods for using genetic variants in causal estimation" is included in the course fees for in-person courses (not for online courses). JJ Thompson S For example, UK Biobank will soon release GWAS data on all 500 000 participants and has already amassed large numbers of incident cases of cardiovascular disease and common cancers such as breast cancer. Nature Genetics, 47(9). . In Figure 1 a and b, the IV is randomization to receiving a statin or not (i.e this is an example of IV analyses in an RCT). What is more surprising is that they seem to have also used sex-combined results for determining effects of WHR adjusted for BMI, despite the fact that it is clear from the title of the original GWAS paper that sex differences were examined and found 19 ( Table 2 ). A further potential explanation for why most of the emphasized (based on statistical testing) MR results are seen for adult BMI, rather than any of the other adiposity risk factors, is that the genetic instrument for adult BMI is stronger than for the other traits. Similarly, an odds ratio of 1.27 (1.09, 1.49) for the effect of adult BMI on all lung cancers is declared as a positive result but the same conclusion is not made for an odds ratio of 1.33 (95%CI: 0.75, 2.36) for the MR effect of WHR on squamous lung cancer. RM Davey Smith . 4 Gamazon, E. et al. Randall . EP-I Horikoshi So far, MR studies in this area have focussed solely on Alzheimer’s dementia, with all three reporting no impact of diabetes [4–6]. Mendelian Randomization results from GeneAtlas. Paré et al. Describe any key additional analyses that would have been important to conduct, such as of sub-phenotypes or interactions, that were not possible because of the summary data. Davey Smith G NM The authors speculate that the protective effect of adult BMI on breast cancer (including postmenopausal) might represent a complex interplay between early life BMI and later weight gain. M #if you repeat this you will typically get a non-significant result, #let's repeat this 1,000 times and see how often we get a significant results, #proportion of tests that are significant, #let's plot the difference in the means for this estimates, the "effect size", "Estimated effect sizes (1000 simulations) for sample of size n=", #this is the effect size for all our estimates, #add a line to indicate where the true population mean lies, #let's look only at the tests that are significant at an alpha=0.05, #this is the effect sizes for the significant results. Burgess This issue is not discussed by Gao et al.