OpenAI is expanding ChatGPT into healthcare by integrating ChatGPT Health with Epic’s electronic health record system, giving clinicians access to patient information and AI-assisted analysis across medical records covering more than 325 million patients.
The integration will allow clinicians to import information from Epic and use ChatGPT to analyze and summarize appointment notes, laboratory results, medications, specialist documentation, and other elements of a patient’s medical history.
In some healthcare systems, OpenAI said ChatGPT will be embedded directly into existing electronic health record workflows, allowing clinicians to conduct pre-visit reviews and build clinical timelines without leaving a patient chart.
Register for the next Tekedia Mini-MBA.
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
The company said the integration is strictly read-only. ChatGPT can retrieve and analyze information from the medical record, but the AI does not write information back into the patient’s chart.
The move marks a significant expansion of OpenAI’s healthcare strategy. Rather than limiting ChatGPT to a consumer-facing tool for answering health questions, the company is positioning its models as an interface for accessing and synthesizing information already held within clinical systems.
For physicians, one potential benefit is reducing the time required to review fragmented patient histories before appointments. A clinician could use the system to summarize previous consultations, track changes in laboratory results or medications, and assemble information from different specialists into a single timeline.
OpenAI is also introducing a Healthcare Public Data plugin that can retrieve information from medical and government databases, including ClinicalTrials.gov, the Centers for Medicare & Medicaid Services’ coverage information, RxNorm, DailyMed and PubMed.
The company said the plugin can help healthcare professionals synthesize information involving clinical-trial eligibility, medication identifiers, coverage policies and provider records.
OpenAI is further allowing organizations that have a Business Associate Agreement with the company to use ChatGPT Work, Codex, apps and connectors for healthcare-related workflows designed to meet applicable compliance requirements.
The expansion comes as OpenAI reports rapidly increasing use of ChatGPT for health-related questions. The company said users are now submitting about 300 million health queries each week, up from 230 million when it began testing its dedicated health hub in January.
The consumer health service is available to U.S.-based users aged 18 and older across ChatGPT plans. Users can connect information from services including Apple Health, Function and MyFitnessPal, as well as medical records from systems and providers such as Epic, Oracle Health, One Medical and Function Health.
OpenAI said testing showed that about 70% of health-related conversations occurred outside the dedicated health hub. The company is therefore expanding access to connected health information across general ChatGPT conversations, allowing users, for example, to ask questions about food while drawing on information about their allergies or other connected health data.
Safety Remains A Major Concern
The expansion also puts greater scrutiny on the reliability of AI in medical settings.
OpenAI said it collected more than 4,300 physician assessments across 27 clinical use cases, including pre-visit reviews, clinical timelines, medication reviews and patient handoff summaries. It said 99.1% of the responses were judged safe.
That figure nevertheless leaves room for potentially consequential errors. In healthcare, a small number of unsafe recommendations can have serious consequences, particularly when users interpret an AI response as medical advice.
OpenAI has maintained that ChatGPT is not intended to replace clinicians or independently diagnose or treat patients. The company’s growing integration with medical records, however, brings its technology closer to decisions and workflows where errors can carry substantially greater risks than in ordinary consumer applications.
The timing also places the rollout under heightened scrutiny. OpenAI is facing lawsuits over alleged harmful health advice, including a case filed by a Florida pastor who claims ChatGPT gave him a recommendation that nearly proved fatal. In another lawsuit filed in May, family members of a user alleged that ChatGPT provided incorrect advice involving medication dosage.
The Epic integration therefore represents both a technological and commercial test for OpenAI. Access to large-scale clinical data could make ChatGPT substantially more useful to healthcare professionals, while the read-only architecture gives hospitals a degree of separation between AI-generated analysis and the underlying medical record.
Overall, OpenAI is seeking to move ChatGPT from a standalone conversational application into an AI layer that can sit on top of the systems professionals already use. Healthcare, where clinicians routinely spend significant time navigating large volumes of fragmented information, could become one of the most consequential applications of that approach.
At the same time, the closer ChatGPT gets to clinical workflows, the higher the standard for accuracy, privacy, security, and human oversight will become.