Artificial intelligence is entering a new phase in health care—not primarily as a tool for discovering experimental drugs or generating medical images, but as a system designed to help more patients receive treatments that already work. A new Viewpoint published by JAMA argues that the most immediate public-health opportunity for AI may lie in identifying people who are eligible for proven interventions, reaching them before opportunities for care are missed, and coordinating the complex steps required to deliver treatment. The authors describe this approach as a public-health agenda for artificial intelligence, one centered less on technological novelty than on closing the gap between what medicine knows and what patients actually receive.
That gap remains one of the largest and least visible problems in modern health care. Clinical guidelines may recommend screening, vaccination, preventive medications, surgery, rehabilitation, or long-term disease management, yet eligible patients frequently go untreated. Some are never identified by health systems. Others receive a recommendation but cannot obtain an appointment, do not understand the next step, lose contact with a clinic, or face financial, geographic, linguistic, or social barriers. In this context, AI could function as an infrastructure for finding missed opportunities across large populations. By analyzing information already stored in electronic health records, insurance claims, laboratory databases, pharmacy records, and public-health registries, algorithms could help determine who is likely to benefit from an intervention and which patients require immediate outreach.
The technical foundation for such systems is population-level risk stratification. Machine-learning models can examine thousands of variables simultaneously, including diagnoses, test results, medication histories, patterns of missed appointments, hospital admissions, demographic characteristics, and changes in clinical status. Rather than waiting for a clinician to notice that a patient meets a guideline, an algorithm can continuously compare patient data with evidence-based eligibility criteria. For example, a system might identify people overdue for cancer screening, patients with uncontrolled hypertension who may benefit from medication adjustment, or individuals with a chronic disease who have not received recommended follow-up. The purpose is not to replace clinical judgment. It is to create a reliable signal that helps care teams focus attention where the probability of meaningful benefit is highest.
Identification, however, is only the first step. The authors emphasize that AI must be connected to outreach and care coordination if it is to improve health rather than merely produce more alerts. A model that flags thousands of patients but sends those cases into an already overloaded inbox may increase administrative burden without improving outcomes. Effective systems would need to rank cases by urgency, estimate the most appropriate communication channel, and support workflows that connect patients with nurses, physicians, pharmacists, community health workers, or social-service organizations. In some circumstances, an automated message might be sufficient. In others, a patient may require a phone call, transportation assistance, language support, financial counseling, or an appointment with a specialist.
This distinction is crucial because health-care delivery is not a simple information problem. A patient may be technically eligible for a treatment but unable to access it because of cost, limited transportation, unstable housing, caregiving responsibilities, or distrust created by previous experiences with the medical system. AI can help identify patterns associated with these barriers, but algorithms cannot solve them independently. The proposed public-health model therefore treats artificial intelligence as part of a coordinated service network. Its value would depend on whether institutions can respond to the needs that algorithms reveal, not merely on whether the algorithms achieve high predictive accuracy.
The Viewpoint also raises an important technical and ethical challenge: a model may be statistically accurate while still worsening inequities. Health data reflect the structure of the health system that produced them. If certain communities have historically received less care, their records may contain fewer diagnoses, fewer referrals, and fewer opportunities to demonstrate that an intervention worked. An algorithm trained on those data could interpret missing information as low risk, reproducing the very disparities it was intended to reduce. Bias can also enter through the choice of outcome, the definition of eligibility, the design of training datasets, and differences in how hospitals document care. For that reason, AI-driven outreach would require continuous evaluation across racial, ethnic, socioeconomic, geographic, and disability groups.
Privacy and governance are equally central. Population-health algorithms may combine highly sensitive medical and social information to generate predictions about disease, adherence, or future care needs. Patients and communities must be able to understand how their information is being used, what decisions an algorithm influences, and how to challenge an incorrect classification. Strong safeguards would be needed to limit unauthorized access, prevent data from being repurposed for discrimination, and ensure that automated recommendations remain subject to human oversight. The authors’ argument places responsibility not only on technology companies and hospitals, but also on government agencies that establish standards for data quality, transparency, accountability, and clinical safety.
The proposed agenda would also require partnerships that extend beyond traditional medical institutions. Health departments, insurers, technology developers, primary-care practices, hospitals, pharmacies, community organizations, and social-service agencies often possess different pieces of the information needed to reach a patient successfully. Coordinating those systems is technically difficult because data may be stored in incompatible formats, governed by different privacy rules, or updated at different speeds. Public investment could help create interoperable data infrastructure and shared evaluation standards, while public programs could support implementation in communities that commercial markets have historically underserved. Without such investment, AI may be deployed most rapidly where resources are already abundant, leaving the patients with the greatest unmet needs behind.
The authors ultimately frame artificial intelligence as a tool for improving the delivery of established medicine rather than as a substitute for medical expertise or public policy. The central test is practical: can an AI system help a health organization recognize that a patient needs a proven intervention, make contact at the right time, remove barriers to care, and verify that the intervention was actually received? Answering that question will require more than promising demonstrations or impressive benchmark scores. It will require prospective studies, monitoring for unintended consequences, transparent reporting, and sustained funding for the people and institutions responsible for acting on algorithmic recommendations. If those conditions are met, AI could become an invisible but powerful layer of public-health infrastructure—one that helps transform medical knowledge into treatment delivered, patients reached, and preventable illness avoided.
Subject of Research: The use of artificial intelligence to identify patients eligible for proven health interventions and improve outreach, care coordination, and health-care delivery.
Web References: 2025 JAMA Summit Report on Artificial Intelligence; JAMA+ AI
References: JAMA Viewpoint, DOI: 10.1001/jama.2026.16748; Corresponding author: Adam L. Beckman, MD, MBA, Health and Opportunity Leadership Institute, City College of New York.
Keywords: Artificial intelligence, public health, health-care delivery, disease intervention, government, patient identification, outreach, care coordination.
