Adjusting for Medication Use in GWAS and Its Impact on Mendelian Randomisation Analyses: An Example of Systolic Blood Pressure in UK Biobank

Medication use is common in large-scale population cohorts, and can modify phenotypic traits of interest. This can potentially bias effect estimates in genome-wide association studies (GWAS) and impact downstream analyses such as Mendelian randomisation (MR). The best approach to account for medication use in GWAS is unclear. In this study, we compared seven different methods of adjusting for antihypertensive use in a systolic blood pressure (SBP) GWAS of 407,960 White British individuals in the UK Biobank. We found that direct adjustments to measured SBP (adding constants, class-specific constants, censored normal regression) in general yielded a greater number of genome-wide significant variant associations and unmasked stronger GWAS effect estimates than unadjusted measures of SBP. Adjustment for class-specific constants showed the greatest difference relative to unadjusted GWAS. Restriction methods which limit the sample to either untreated individuals or age ranges with low levels of antihypertensive use had less power, due to reduced sample sizes. Effect estimates of treated individuals were deflated relative to untreated individuals, demonstrating the importance of medication adjustment. In MR analyses, we found no substantial differences in inverse-variance weighted (IVW) estimates when using differing exposure GWAS methods in estimation of the effect of SBP on coronary artery disease. Larger variations in IVW estimates were observed for the causal effect of body mass index on SBP across adjustment approaches. This suggests that bias may arise in MR analyses when the exposure included in the estimation affects the probability of treatment. Finally, we demonstrate that medication adjustment can reveal potentially novel genetic loci, offering additional insight into the biology of a trait.