CMStatistics 2021: Start Registration
View Submission - CMStatistics
B0980
Title: Avoiding bias in Mendelian randomization when stratifying on a collider Authors:  Claudia Coscia - Spanish National Cancer Research Centre (CNIO) (Spain) [presenting]
Dipender Gill - Imperial College (United Kingdom)
Teresa Perez - Universidad Complutense de Madrid (Spain)
Nuria Malats - Spanish National Cancer Research Centre (Spain)
Stephen Burgess - University of Cambridge (United Kingdom)
Abstract: Mendelian randomization (MR) uses genetic variants as instrumental variables to investigate the causal effect of a risk factor on an outcome. A collider is a variable influenced by two or more variables. A naive calculation of MR estimates in strata of the population defined by colliders may generate collider bias. We propose an approach that allows stratifying the MR analysis avoiding bias. The approach constructs a new variable, the residual-collider, as the residual from regressing the collider on the instrument, and calculates causal estimates in strata defined by quantiles of the residual-collider. Stratum-specific collider and residual-collider estimates will have equivalent interpretations, but residual-collider estimates will not suffer from collider bias. We apply this approach in several simulation scenarios considering different characteristics of the collider variable and instrument strengths, and to investigate the causal effect of smoking on bladder cancer in strata of the population defined by bodyweight. The approach generated unbiased estimates in all simulation settings, and a trend in the stratum-specific MR estimates at different bodyweight levels that suggested stronger effects of smoking on bladder cancer among individuals with lower bodyweight. This approach can be used to perform MR studying heterogeneity among subgroups of the population while avoiding collider bias.