OpenAI's Epic Integration: A Productivity Leap With a Diagnostic Void
AI-generated image · Bay Street Wire
The move to import patient data into ChatGPT Health promises efficiency for clinicians, but safety concerns and a lack of diagnostic validation leave the tool in a precarious position.
As TechCrunch first reported, OpenAI has announced a significant expansion of its healthcare ambitions by integrating ChatGPT Health with Epic’s electronic health record (EHR) system. This integration allows clinicians to import data from a system that serves more than 325 million patients. In certain deployments, the AI will be embedded directly within EHR workflows, enabling medical professionals to access and summarize laboratory results, medications, specialist documentation, and appointment notes without leaving a patient's chart.
Clinicians can use the tool to identify changes in patient history, build clinical timelines, and conduct pre-visit reviews. To further bolster utility, OpenAI is introducing a Healthcare Public Data plug-in capable of pulling information from PubMed, RxNorm, DailyMed, CMS Coverage, and ClinicalTrials.gov to help staff synthesize data regarding provider records, coverage policy versions, medication identifiers, and trial eligibility.
However, a divide exists between administrative productivity and clinical safety. OpenAI claims that after collecting more than 4,300 responses from physicians across 27 clinical use cases—including handoff summaries and medication reviews—it found that 99.1% of responses were safe. As TechCrunch notes, even a small number of unsafe answers can result in harmful outcomes for patients.
This risk is not theoretical. TechCrunch reports that a Florida-based pastor recently sued the company alleging a near-fatal recommendation, and in May, OpenAI was sued by family members of a user who alleged the AI provided wrongful dosage advice. Despite these lawsuits, OpenAI maintains that its AI is not suitable for treatment or diagnosis, and the Epic integration is limited to read-only access.
***Opinion:*** *While the ability to summarize a patient's chart is a welcome productivity gain, labeling this as a 'health' tool without rigorous, peer-reviewed validation of its diagnostic accuracy is a dangerous gambit. Until OpenAI can prove its reliability through transparent clinical trials rather than internal safety percentages, this integration remains a high-risk administrative assistant masquerading as a medical one.*
For organizations with a Business Associate Agreement, OpenAI is also permitting the use of connectors, apps, Codex, and ChatGPT Work. This follows the general rollout of ChatGPT for health to U.S. consumers last month, which OpenAI says has seen users submit 300 million health-related queries per week.