Robust instrumental variable methods in Mendelian Randomization
Presenter: Simon Thompson
When: Thursday, July 14, 2016 Time: 9:00 AM - 10:30 AM
Room: Lecture Theatre (Level 1)
Session Synopsis:
Robust instrumental variable methods in Mendelian Randomization
Attempting to distinguish causal relationships from observational associations is a crucial scientific endeavour, and a pervading theme of many methods and applications in biostatistics and biometrics. Mendelian randomization is the use of genetic variants as instrumental variables to identify and estimate causal relationships between an exposure and an outcome based on observational data. Studies using Mendelian randomization are now being extensively employed in epidemiology to distinguish causal and observational associations, and as a way to prioritize potential targets for pharmaceutical intervention. As the approach has become more popular, a critical assessment of the instrumental variable assumptions on which it is based can get neglected. In this session we discuss to what extent the instrumental variable assumptions can be empirically assessed, explore some recent developments in instrumental variable analysis that provide some robustness against failure of the assumptions, and discuss their applicability in studies based on Mendelian randomization.
Robust instrumental variable methods in Mendelian Randomization
Presenter: Maria Glymour
When: Thursday, July 14, 2016 Time: 9:00 AM - 10:30 AM
Room: Lecture Theatre (Level 1)
Session Synopsis:
Instrumental variable assumptions, and their practical interrogation in Mendelian Randomization studies
Causal inferences based on instrumental variables (IV) analyses of observational data rest on strong assumptions. These assumptions are primarily structural and can be represented using Directed Acyclic Graphs (DAGs). I will discuss the structural assumptions, how to evaluate whether violations of those assumptions could plausibly account for results in a particular study, and strategies to evaluate the assumptions using the data at hand. In general, background assumptions about the causal structure and modifying or confounding covariates can suggest opportunities to falsify the assumptions for a valid IV. In many settings, although the assumptions cannot be proven, plausible violations are unlikely to account for the observed results of IV studies. I will also briefly discuss non-structural assumptions, such as monotonicity, which can be used to relate the IV estimate to the causal effect for a particular population subgroup.
Robust instrumental variable methods in Mendelian Randomization
Presenter: Stephen Burgess
When: Thursday, July 14, 2016 Time: 9:00 AM - 10:30 AM
Room: Lecture Theatre (Level 1)
Session Synopsis:
Weighted median estimators for robust inference in Mendelian randomization
Developments in genome-wide association studies and the increasing availability of summary genetic association data have made application of Mendelian randomization relatively straightforward. However, obtaining reliable results from a Mendelian randomization investigation remains problematic, as the conventional inverse-variance weighted method only gives consistent estimates if all of the genetic variants in the analysis are valid instrumental variables. A novel weighted median estimator for combining data on multiple genetic variants into a single causal estimate will be presented. This estimator is consistent even when up to 50% of the information comes from invalid instrumental variables. In a simulation analysis, it is shown to have better finite-sample Type 1 error rates than the inverse-variance weighted method, and is complementary to the recently proposed MR-Egger (Mendelian randomization-Egger) regression method. In analyses of the causal effects of low-density lipoprotein cholesterol and high-density lipoprotein cholesterol on coronary artery disease risk, the inverse-variance weighted method suggests a causal effect of both lipid fractions, whereas the weighted median and MR-Egger regression methods suggest a null effect of high-density lipoprotein cholesterol that corresponds with the experimental evidence. Both median-based and MR-Egger regression methods should be considered as sensitivity analyses for Mendelian randomization investigations with multiple genetic variants.
Robust instrumental variable methods in Mendelian Randomization
Presenter: Hyunseung Kang
When: Thursday, July 14, 2016 Time: 9:00 AM - 10:30 AM
Room: Lecture Theatre (Level 1)
Session Synopsis:
Robust Confidence Intervals for Causal Effects with Possibly Invalid Instruments
Instrumental variables have been widely used to estimate the causal effect of a treatment on an outcome. Existing confidence intervals for causal effects based on instrumental variables assume that all of the putative instrumental variables are valid; a valid instrumental variable is a variable that affects the outcome only by affecting the treatment and is not related to unmeasured confounders. However, in practice, some of the putative instrumental variables are likely to be invalid. The talk presents two methods of constructing confidence intervals in the presence of possibly valid instruments. The first method is simple and is an easy modification of widely used methods in the instrumental variables literature. The second method uses recent developments in de-biased estimators of l_1 penalized problems along with hard-thresholding to construct robust confidence intervals. Both confidence intervals have theoretical guarantees on having the correct coverage and are shown to outperforms traditional confidence intervals popular in instrumental variables literature when invalid instruments are present. We also demonstrate the two approaches on a real data set.
Robust instrumental variable methods in Mendelian Randomization
Presenter: Stijn Vansteelandt
When: Thursday, July 14, 2016 Time: 9:00 AM - 10:30 AM
Room: Lecture Theatre (Level 1)
Session Synopsis:
Mendelian randomisation analysis of time-to-event endpoints
Mendelian randomisation analysis is well established for continuous outcomes and to some extent for binary outcomes. The analysis of time-to-event endpoints is, however, largely based on ad hoc approaches. Formal approaches are lacking because of complications due to censoring, and because of differential survival which causes the instrumental variables assumptions on which a Mendelian randomisation analysis relies, to fail within the study population at risk at a given time (including the study population at the start of the study). In this talk, we develop the IV approach for regression analysis in a survival context, primarily under an additive hazards model. We will discuss several, relatively simple approaches that make different assumptions about the data-generating distribution. Formal conditions are given justifying each strategy, and the methods are illustrated in a novel application to a Mendelian randomization study to evaluate the effect of diabetes on mortality using data from the Health and Retirement Study.
