Adjusting for medication status in genome-wide association studies

Genome-wide association studies (GWAS) have revolutionized our understanding of how genetic variations influence health outcomes. However, these studies face a significant challenge when participants are on medications that alter the traits under investigation, such as cholesterol or blood pressure levels. Failure to properly account for medication use can lead to misleading conclusions about genetic effects, either exaggerating or diminishing their true impact.

A study published by journals.plos.org in late 2026 introduces an innovative approach named the Genetic Empirical Medication Reduction Adjustment (GEMRA) method. This technique aims to refine the accuracy of GWAS by empirically adjusting phenotypic measurements of individuals taking medication, thereby approximating what their trait values might have been without treatment.

Traditional approaches for handling medication status in GWAS have included excluding medicated individuals, treating medication use as a covariate, or ignoring medication effects altogether. Each of these methods presents drawbacks: excluding medicated participants reduces sample size and statistical power, while ignoring medication can introduce bias due to heterogeneous phenotype values. Including medication as a covariate may also fail to fully address indirect effects arising from complex causal pathways.

GEMRA differentiates itself by employing a heuristic search for an optimal adjustment, applying either a relative percentage reduction or an absolute additive correction to observed phenotypic values in medicated individuals. This method was rigorously evaluated through simulations and real-world data analyses utilizing multi-population cohorts from the UK Biobank and the Uganda Genome Resource.

The researchers focused on key traits affected by medication: low-density lipoprotein cholesterol (LDLc), influenced by statins, and systolic and diastolic blood pressure (SBP and DBP), affected by antihypertensive drugs. Their findings indicate that modeling medication effects as proportional reductions rather than fixed amounts better captures the true medication impact across diverse populations. Specifically, statins were found to reduce LDLc by approximately 40%, while antihypertensives lowered SBP and DBP by about 10% each.

Simulations comparing GEMRA to conventional methods demonstrated that GEMRA yields genetic effect estimates with notably less bias and higher statistical power. Unlike approaches that restrict analyses based on medication status or include medication use merely as a covariate, GEMRA effectively minimizes false positives and improves detection of true genetic associations.

An important insight from the study is the risk of collider bias—an analytical artifact that can arise when conditioning on medication status. Such bias can falsely associate genetic variants linked to other health factors with the trait of interest, complicating interpretation. GEMRA’s adjustment strategy helps circumvent this problem by restoring phenotypic values closer to their unmedicated counterparts, maintaining the integrity of causal inference.

Moreover, the study underscores the value of cross-population analyses. By applying GEMRA to datasets from both the UK and Uganda, the researchers revealed that medication effects are broadly consistent but that improper modeling can create artificial disparities between populations. This highlights the necessity of tailored adjustments sensitive to population-specific medication use patterns and phenotypic distributions.

The broader implications of this work extend beyond cholesterol and blood pressure traits. As medications frequently influence phenotypes studied in genetic research, GEMRA presents a versatile framework that could be applied to other traits and diseases, enhancing the reliability of GWAS findings in diverse global cohorts.

While GEMRA shows promise, the authors caution about the importance of accurately estimating medication effect sizes. Incorrect adjustment levels can still introduce bias, although most adjustments improve upon uncorrected analyses. Additionally, variability in individual responses to medication introduces some uncertainty, which GEMRA partially accounts for by increasing estimate precision.

In summary, the GEMRA method represents a significant advance in genetic epidemiology, offering a more robust way to incorporate medicated individuals into genome-wide analyses. By mitigating bias and preserving statistical power, this approach supports more precise identification of genetic factors influencing health, ultimately contributing to better-informed strategies in personalized medicine and public health.