The Growing Use of Artificial Intelligence in Health Care and Implications for Disparities

Introduction

Artificial intelligence (AI) is more and more being built-in into health care, together with however not restricted to prognosis and therapy plans, drug improvement, prediction of well being dangers and outcomes, well being monitoring, and medical imaging. AI can additionally automate points of well being care together with information processing and administrative duties, reimbursement choices, affected person interactions, and clinical decision-making. Additionally, people are more and more utilizing AI for well being info and recommendation.

While there was an increase in funding for and use of AI in well being care lately, public opinion on AI’s position in offering correct well being info stays combined. Further, there are considerations that AI could result in job losses and scale back customized human-based interactions. Moreover, AI can exacerbate well being disparities if the underlying information on which fashions are constructed are biased and/or not inclusive. Alternatively, some recommend that AI could assist mitigate disparities whether it is rigorously designed. This transient examines the implications of the rising use of AI for disparities in well being and well being care and discusses components that may assist scale back AI-related bias in well being care.

Growing Use of AI in Health Care

AI instruments have gotten more and more integrated into numerous points of the well being care system. For instance, hospitals report utilizing AI or predictive fashions as each administrative instruments to carry out duties reminiscent of patient scheduling, billing, and medical coding, and as clinician-facing instruments to foretell well being risks and outcomes amongst sufferers. A 2025 survey carried out throughout 16 states discovered that eight in ten (84%) well being insurers report utilizing AI or machine studying for fraud detection, utilization administration, and prior authorization, amongst different makes use of. Health techniques additionally report utilizing AI to “limit claim denials and streamline prior authorization processes.”

The public is also more and more utilizing AI for well being info and recommendation though many have restricted belief within the reliability of AI instruments. According to OpenAI information from 2026, greater than 40 million folks globally flip to ChatGPT each day for well being info. The information additionally present that AI chatbots have gotten an vital supply of knowledge for medical health insurance and billing recommendation, with customers asking between 1.6 and 1.9 million questions per week concerning plan comparisons, claims, billing, and protection. Further, a 2026 KFF survey finds that a couple of third (32%) of adults say they use AI chatbots for well being info or recommendation (Figure 1). However, two-thirds (67%) of adults general say they belief AI instruments or chatbots “not too much” or “not at all” to supply dependable well being info, and about three in 4 (77%) say the identical concerning details about psychological well being and emotional well-being. While charges of use for and belief in AI for bodily well being recommendation are comparable throughout racial and ethnic teams, Black and Hispanic adults are extra possible than their White counterparts to report utilizing AI for psychological well being recommendation and Black adults (29%) are considerably extra possible than White adults (20%) to say they belief AI instruments or chatbots to supply dependable details about psychological well being and emotional well-being “a great deal” or “a fair amount.”

Impact of AI on Disparities in Health and Health Care

As the usage of AI in well being care grows, analysis means that AI fashions can exacerbate racial and ethnic well being disparities. A 2024 systematic review of 30 research over a ten-year time interval (from 2013 to 2023) that assessed cases of racial bias perpetuated by AI and machine studying algorithms in well being care discovered a major affiliation between AI utilization and an exacerbation of racial disparities in well being and well being care outcomes. These disparities included longer ready occasions for appointments, decrease charges of success in predicting psychological well being outcomes, and underdiagnosis of well being circumstances, notably for Black and Hispanic folks in comparison with different teams. For instance:

• One examine discovered {that a} machine studying algorithm used for creating affected person appointment schedules led to Black sufferers experiencing 33% longer wait occasions than different sufferers. This was because of the mannequin utilizing socioeconomic indicators reminiscent of employment standing, zip code, insurance coverage sort, and previous no-show charges, that are correlated with race, to create appointment schedules.
• Another study discovered {that a} extensively used algorithm to information well being care choices assigned Black sufferers the identical degree of threat as White sufferers though Black sufferers have been sicker. The algorithm used well being care prices as an imperfect proxy for sickness, since much less cash is spent on Black sufferers who’ve an equal degree of want on account of inequities in entry to care. The authors recommend that addressing this disparity would considerably improve the share of Black sufferers receiving extra care.
• In diagnostics, AI fashions could underperform on sufferers with darker pores and skin as a result of training datasets usually tend to accumulate information from lighter skinned sufferers.

In the systematic evaluate, the authors recognized 4 main and interrelated causes for AI-perpetuated disparities together with: biased underlying datasets, historic and systemic biases that may be encoded into AI when it’s skilled on these information, algorithmic design bias, and biased utility and/or deployment of AI.

These AI-related racial and ethnic disparities additionally prolong into psychological well being prognosis and therapy suggestions. For instance, language-based AI fashions underperformed on predicting despair severity for Black sufferers as in comparison with White sufferers because the two teams use various kinds of language to precise despair signs and AI is commonly primarily skilled on language utilized by White sufferers given that there’s extra information out there on White sufferers since they make up a bigger share of the inhabitants. However, researchers discovered that even fashions skilled completely on the depression-related social media language utilized by Black people carried out poorly at predicting despair severity within the group whereas fashions skilled with the identical social media information on White people carried out properly at predicting that group’s despair severity. The authors recommend that this could possibly be on account of different components past language, reminiscent of paralinguistic options like speech price or tone, serving as higher predictors for despair severity amongst Black people. A separate examine discovered that a number of AI fashions made inferior treatment suggestions for Black psychological well being sufferers when the affected person’s race was explicitly or implicitly talked about, possible on account of biases embedded within the information on which these fashions are skilled. An AI mannequin used for suicide prediction additionally carried out worse for Black sufferers, with researchers discovering that it efficiently detected 62% of suicides amongst White sufferers however solely 10% amongst Black sufferers.

Research has discovered that the usage of race in medical algorithms might also influence the reliability of AI instruments for sure teams since they’re usually skilled on these algorithms. AI fashions are sometimes skilled on medical algorithms used to foretell diagnoses and coverings, which in some instances have traditionally used race as an element and resulted in worse outcomes for some teams. One of probably the most well-known examples of this apply is the usage of separate measures of kidney function (i.e., estimated glomerular filtration charges, eGFRs) for Black sufferers in comparison with non-Black sufferers, which resulted in lots of Black sufferers not receiving a kidney transplant. Another study discovered that eradicating the usage of race from spirometry, a take a look at used to measure lung operate, would improve the variety of Black individuals who would qualify for lung illness prognosis and incapacity funds. Further, a 2019 study discovered that an algorithm used to foretell the chance of safely having a Vaginal Birth after Cesarean Delivery (VBAC) incorrectly predicted a decrease chance of success for VBAC for Black and Hispanic ladies than White ladies, which led to docs performing extra cesarian deliveries on Black and Hispanic ladies than White ladies. A rising variety of organizations and well being care establishments have not too long ago moved to take away race from these algorithms. However, to the extent AI is skilled on algorithms or outcomes from algorithms that use race as an element, AI might perpetuate these racial biases.

Research additionally exhibits that AI fashions could promote racial and ethnic well being misinformation, resulting in misdiagnosis or delayed care. A examine of a number of AI chatbots discovered cases of the instruments selling “race-based medicine” and false claims about race reminiscent of distinction in pores and skin thickness between Black and White sufferers. Further, all AI chatbots included within the examine incorrectly acknowledged that Black males’s and ladies’s regular lung operate tends to be decrease than their White counterparts’, reflecting its coaching on the underlying race-biased algorithm to calculate lung operate.

If rigorously designed, AI has the potential to assist address disparities. For instance, AI-driven choice assist instruments can be utilized to determine and proper real-time clinician bias, notably throughout high-stress durations when “cognitive load” usually results in disparities in documentation and prognosis. By automating administrative duties reminiscent of scheduling and billing, AI might assist scale back workers burnout at safety-net hospitals, which disproportionately deal with underserved teams. AI may also be used to determine the social determinants that drive well being inequities via the evaluation of enormous quantities of inhabitants information, which may then assist information interventions to handle disparities. AI can even assist determine disparities in well being outcomes that may in any other case go unrecognized. For instance, in a current study, researchers used machine studying to determine extra deaths on account of COVID-19 that have been unrecognized in official mortality reviews and located that these unrecognized deaths occurred disproportionately amongst folks of coloration, these with decrease instructional attainment, and people with decrease family incomes, amongst different components.

Careful design and inclusive information assortment; a various workforce; and a concentrate on moral issues, transparency, and a collaborative strategy are components which will assist mitigate AI biases in well being care. Identification and mitigation of biases throughout AI fashions’ improvement, in addition to steady monitoring and inclusion of extra consultant information over time, can assist to handle AI-related bias in well being care. Further, having a diverse and consultant information science workforce and coaching AI builders to acknowledge biases in algorithm improvement additionally play an vital position in growing equitable AI fashions. Developing and imposing moral standards for AI in well being care that inform how AI fashions and algorithms will likely be designed to assist scale back bias and discrimination and establishing accountability within the creation and use of these algorithms might also assist to scale back algorithmic bias. Further, collaborating with a variety of stakeholders, reminiscent of well being care employees, policymakers, community members, and ethicists when growing AI instruments can provide a broader and extra nuanced understanding of the influence of AI on well being disparities.

Researchers and different specialists have elevated their concentrate on the creation of frameworks and coalitions to assist information equitable use of AI in well being care. In 2023, the Coalition for Health AI launched steerage for the implementation of AI instruments that facilities fairness, equity, and ethics. The steerage consists of suggestions on growing a typical set of rules to information the event and use of AI instruments and a coalition or advisory board to assist guarantee fairness and facilitate trustworthiness in health-related AI. In early 2024, specialists in well being, drugs, know-how, and coverage issued a call for “ongoing dialogue and ethical commitment from all stakeholders” to make sure that AI in well being care is inclusive following a sequence of discussions on the 2023 Responsible AI for Social and Ethical Healthcare (RAISE) worldwide symposium. In 2024, the Council of Medical Specialty Societies and the Doris Duke Foundation created the Encoding Equity alliance, whose goals are to determine the wrong use of race in medical algorithms and tips, design “accurate and equitable decision tools”, and accumulate and disseminate proof on the usage of AI in well being care to advertise well being fairness.

While there was growing exercise on the state-level to manage AI in well being care, the Trump administration has prioritized deregulation of AI, diminished or eradicated fairness necessities for AI in well being care, and is difficult state laws that impose strict anti-bias necessities.  President Trump issued Executive Order (EO) 4148 in January 2025 that rescinded quite a lot of Biden administration EOs, together with these associated to equitable use of AI in well being care. He changed these EOs with EO 14179, which shifts focus away from “equity” mandates and “algorithmic fairness” and in the direction of “minimally burdensome” necessities to encourage innovation. While quite a few states have not too long ago launched or enacted laws associated to AI in well being care, the Trump administration is difficult state legal guidelines that impose strict bias audits or transparency necessities for AI by way of EO 14365 issued in December 2025. Under the EO, the Department of Justice created an AI Litigation Task Force in January 2026 to problem states with AI legal guidelines discovered to be inconsistent with federal coverage. The EO additionally directs the Secretary of Commerce to limit federal grant cash, particularly the Broadband Equity Access and Deployment (BEAD) Program funds, in states with “onerous” AI legal guidelines. For instance, Colorado handed the “Consumer Protections for Artificial Intelligence” legislation in 2024, which amongst different issues, requires well being care suppliers and well being insurers to take steps to forestall algorithmic discrimination. However, implementation of the legislation has been postponed on account of legal challenges.