American Heart Association Issues New Framework for Responsible AI Implementation in Health Care

Hundreds of health care artificial intelligence tools have been cleared by the U.S. Food and Drug Administration, yet only a fraction are rigorously evaluated for clinical impact, fairness or bias. Furthermore, FDA review only covers a small portion of health AI tools being developed and used in health care. The American Heart Association today released a new science advisory urging health systems to adopt clear and simple rules for using AI in patient care. Published in the Association's flagship journal, Circulation, the advisory introduces a pragmatic, risk-based framework for evaluating and monitoring artificial intelligence tools in cardiovascular and stroke care.

The advisory builds on prior published AI frameworks to identify critical gaps in current practices and includes key principles to help health systems build effective AI governance for selecting, validating, implementing, and overseeing AI tools. The four guiding principles for health systems in deploying clinical AI proposed are strategic alignment, ethical evaluation, usefulness and effectiveness, and financial performance. These principles will help ensure AI tools deliver measurable clinical benefit while safeguarding individuals from known and as yet unknown harms.

AI is transforming health care faster than traditional evaluation frameworks can keep up, according to Sneha S. Jain, M.D., M.B.A., volunteer vice chair for the American Heart Association AI Science Advisory writing group. The goal is to help health systems adopt AI responsibly, guided by pragmatic, risk-tiered evidence generation that ensures innovation truly improves care. The advisory highlights that while AI tools can enhance diagnostic accuracy and efficiency, many are deployed without rigorous local validation or bias assessment.

A recent survey found that only 61% of hospitals using predictive AI tools validated them on local data prior to deployment, and fewer than half tested for bias. This variability is most pronounced among smaller, rural and non-academic institutions, raising concerns about consistent care delivery across a variety of patient populations and safety. The American Heart Association's extensive network of nearly 3,000 hospitals participating in the Get With The Guidelines quality improvement programs, including more than 500 rural and critical access facilities, positions it as a trusted leader in advancing responsible AI governance.

The Association has committed over $12 million in research funding in 2025 to test novel health care AI delivery strategies for safety and efficacy. The science advisory writing group outlines that monitoring of AI tools cannot end after deployment. Performance of AI tools may drift as clinical practice changes or patient populations differ. They emphasize that health systems should integrate AI governance into existing quality assurance programs and define clear thresholds for retraining or retiring tools if performance declines.

Responsible AI use is not optional, it's essential, according to Lee H. Schwamm, M.D., FAHA, volunteer member of the American Heart Association committee on AI and Technology Innovation. This guidance provides practical steps for health systems to evaluate and monitor AI tools, ensuring they improve patient outcomes and support equitable, high-quality care. The advisory represents a significant step toward establishing standardized approaches to AI implementation in health care settings across the United States.