Clinical AI does not operate without bias. Every model reflects assumptions about the data it uses, the outcomes it prioritizes and the patients it serves. The real question is whether health care organizations understand those assumptions, test them across populations and care settings, and govern them in ways that improve care without widening existing gaps.
In this piece in MedCity News, Shakira J. Grant, MD, a member of the standards committee that helped develop URAC’s Health Care AI Accreditation, explains why removing demographic variables does not automatically make an AI model fair. Health care delivery already varies by access, geography, resources, referral patterns and follow-up care. AI tools that ignore those differences may miss important context or history or amplify blind spots.
Dr. Grant argues for a more intentional approach to clinical AI design and oversight. She believes that models should account for real-world differences in how patients access and receive care, test whether they perform reliably across populations and include diverse community and clinical perspectives throughout development, deployment and monitoring.
Her perspective reinforces a central principle behind URAC’s Health Care AI Accreditation: responsible AI requires lifecycle oversight. Health care organizations need clear governance, documentation of intended use and limitations, transparency, subgroup performance testing and ongoing monitoring as patient needs, data and workflows change.
- Read Dr. Grant’s perspective in MedCity News here. (subscription may be required)
- Learn more about URAC’s Health Care AI Accreditation here
