News|Articles|November 3, 2026
Most AI health equity guidance is general and U.S.-based, Georgetown review finds
Author(s)Austin Littrell
Fact checked by:Keith A. Reynolds
Only 22 of the articles in a Georgetown scoping review defined health equity, a gap the lead author said could steer re
Advertisement
Only 22 of 166 peer-reviewed articles on artificial intelligence (AI) and health equity included an explicit definition of health equity, according to a scoping review published Oct. 8 in Health Affairs Scholar
Researchers at Georgetown University searched four databases in January 2025 for commentaries, editorials, literature reviews and other articles offering guidance on equitable AI use in health care settings, health systems and health services research
The 166 articles that met their criteria were published between 2018 and early 2025
“When health equity is vaguely defined, there is risk that resources will be primarily allocated to the most advantaged group, rather than the group who needs the resources the most,” lead author Caroline R. Efird, Ph.D., an assistant professor in the Department of Health Management and Policy in the School of Health at Georgetown University, said in a news release from Georgetown University Medical Center.
Most of the 22 articles that did define the term modeled it on the World Health Organization (WHO) definition, which describes health equity as the absence of unfair, avoidable or remediable differences among groups of people
Fifteen of the 22 came from U.S.-based author teams
A largely U.S. literature
First authors at U.S. institutions wrote 116 of the 166 articles, or 70%. The United Kingdom followed with 10 and Canada with nine. No first author was based at an institution in South America, and Africa and Australia each accounted for a single article
“A US-centric perspective on AI and health equity likely has blind spots,” Efird said in the release
Little written for specific conditions
Sixty-eight percent of the articles discussed AI and health equity in general terms without focusing on a particular disease or condition. Among those that did, cardiovascular conditions, chronic disease and dermatologic conditions including skin cancers each accounted for 5%
COVID-19 accounted for 3%, and aging, diabetes and radiology each accounted for 2%
Many clinical dermatology textbooks and public image databases have historically shown skin lesions mostly on lighter skin tones, the authors wrote, and an AI tool trained only on that material could struggle to detect how lesions appear on darker skin
“Specialty-specific expertise (e.g., radiology, dermatology, cardiology) is crucial for reducing the risk of worsening health disparities when AI is used in healthcare delivery,” Efird said. “For example, specialists may be better equipped than data scientists to recognize how diagnostic disparities could occur because of the utilization of training data that does not include diverse and representative patient populations.”
The authors called for radiologists, oncologists, gynecologists, neurologists and other specialists to work alongside programmers and health equity researchers through model design, development, implementation and evaluation
What the guidance asks of clinicians
Across the review, articles held that AI should support human judgment in diagnosis rather than replace it and that direct patient-provider interaction should accompany any use of AI in care
Several argued that practice cannot be reduced to algorithms that leave out the empathy, judgment and contextual interpretation a clinical encounter requires. If health care professionals are expected to use AI, the authors wrote, they must be trained on its basic principles and its limits
For health systems, the articles emphasized institutional AI policies built around preventing harm. In the United States, where health patterns differ by race and gender, the authors wrote that race-conscious and gender-conscious algorithms are encouraged
“Diverse, interdisciplinary teams are essential for AI algorithms that are being developed and implemented in healthcare settings,” Efird said. “The inclusion of data scientists, health equity-focused researchers, healthcare providers, specialists, and patients has the potential to reduce model bias.”
A shifting regulatory backdrop
The authors contrasted the European Union’s AI Act, passed in 2024, which requires AI software used for medical purposes to include risk-mitigation systems, high-quality data sets, clear user information and human oversight, with U.S. frameworks they said have prioritized deregulation since early 2025
They cited a January 2025 executive order from President Donald Trump
that revoked the Biden administration’s 2023 order on safe and trustworthy AI
Under a 2024 rule implementing Section 1557 of the Patient Protection and Affordable Care Act
, covered health care entities have an ongoing duty to make reasonable efforts to identify patient care decision support tools that use race, color, national origin, sex, age or disability as input variables, and to mitigate the risk of discrimination from each one
Those obligations took effect May 1, 2025, and the text was unchanged in the federal government’s electronic code as of Oct. 23. The review does not discuss the rule
Physicians still carry the largest share of legal risk when an AI tool contributes to patient harm, Deepika Srivastava, chief operating officer of The Doctors Company, said in a 2025 interview
, because malpractice law has not settled how liability is divided among physicians, health systems and developers
Recommendations and limits
The authors urged grant funders and academic journals to require researchers who use AI to define equity explicitly and to report results for subgroups and marginalized populations
They also recommended that health systems adopt equity-focused AI governance with continual oversight and that national regulators build frameworks based on the WHO definition into their evaluations
Sixty-four percent of the reviewed articles were commentaries, editorials or opinion pieces. Because the review synthesized recommendations rather than outcomes from studies or trials, it could not show whether any of the suggestions reduce health inequities
It included only English-language articles, and the authors acknowledged that more has been published since their search ended
Advertisement
Related to this article
View more
Advertisement
Advertisement
Trending on Medical Economics
12345
