Detecting and mitigating bias for inclusive and trustworthy clinical research: A scientific statement from the American Heart Association

Authors: ZHONG, JUDY - New York University School Of Medicine; AL-ZAITI, SALAH - University Of Rochester; BENNETT, DERRICK - University Of Oxford; DO, SYNHO - Massachusetts General Hospital; GAUDINO, MARIO - Weill Medical College Cornell University; GICHOYA, JUDY - Emory University, School Of Medicine; MUSAAD, SALMA - Children'S Nutrition Research Center (CNRC); NARAYAN, SANJIV - Stanford University; SAJOBI, TOLULOPE - University Of Calgary; SHEN, YU - Md Anderson Cancer Center; ARMOUNDAS, ANTONIS - Massachusetts General Hospital

Submitted to: Circulation: Genomic and Precision Medicine
Publication Type: Review Article
Publication Acceptance Date: 3/2/2026
Publication Date: 5/4/2026
Citation: Zhong, J., Al-Zaiti, S., Bennett, D.A., Do, S., Gaudino, M.F., Gichoya, J.W., Musaad, S.M., Narayan, S.M., Sajobi, T., Shen, Y., Armoundas, A.A. 2026. Detecting and mitigating bias for inclusive and trustworthy clinical research: A scientific statement from the American Heart Association. Circulation: Genomic and Precision Medicine. 19. Article e000101. https://doi.org/10.1161/HCG.0000000000000101.
DOI: https://doi.org/10.1161/HCG.0000000000000101

Interpretive Summary:

Technical Abstract: Bias in clinical research affects not only the internal validity of studies but also the equitable distribution of health benefits derived from studies. Among the most impactful forms are selection bias, attrition bias, and algorithmic bias, each of which is capable of distorting participant representation, treatment effect estimates, and model performance across important subgroups. This scientific statement provides a reference for cardiovascular researchers and clinicians, integrating detection, correction, and prevention strategies to address these biases. Selection bias may be mitigated through approaches such as inverse probability weighting and adjustment for sociodemographic imbalances; attrition bias can be addressed using intention-to-treat analyses and multiple imputation for missing data that are missing at random; algorithmic bias requires fairness-aware modeling, diverse training data sets, and explainable artificial intelligence techniques. These 3 forms of bias are not exhaustive, but their careful management is essential to achieving scientific rigor, fairness, and real-world applicability, and requires multidisciplinary collaboration to embed equity and validity throughout the research lifecycle.