BIONEWS
Hospitals spent two years building their own AI medical record tools; now they're consolidating on Epic, the dominant U.S. electronic health records platform.
Six pathologists at Stanford Health Care stained a lymph node biopsy 70 times trying to identify a cancer. The case had gone cold. Then a physician on the workup asked the hospital's internal chatbot a single question about skin lesions in the patient's history. The tool returned a sarcomatoid squamous cell carcinoma diagnosis logged years earlier at a different health system. That earlier cancer, STAT+ reported this week, explained the lymph node finding the pathologists had spent weeks trying to identify.
Penn Medicine has roughly 70 physicians piloting a homegrown tool called Chart Hero. Duke is building "Scout." Children's Hospital of Philadelphia is building "Chipper." All three are Epic-integrated. So is ChatEHR. Cliniques universitaires Saint-Luc in Brussels published a peer-reviewed account of its own homegrown system in PLOS Digital Health.
On the vendor side, the consolidation is further along. Microsoft Nuance's DAX Copilot is fully embedded in Epic, and 150-plus health systems have committed to deploy it as of 2024. Abridge has received recognition in ambient AI categories. The competitive set — DAX, Abridge, Nabla, DeepScribe, and Suki — has narrowed around Epic's integration layer for ambient documentation.
Chart bloat is the structural problem these tools answer. A typical inpatient chart at a US academic medical center now runs into the hundreds of pages of mixed-signal notes, copied-forward snippets, and billing-coded text. A clinician trying to find what changed since yesterday's visit reads for ten minutes before they can ask a useful question.
"System-wide rollout" still means a vanguard. Penn is at 70 pilot physicians across the institution. Four named homegrown systems is a thin tail. The diagnostic-miracle narrative is the most photogenic part of the deployment, and vendors will keep using it to sell the ambient-chart-review product underneath.
The next check is Penn: a system-wide license at pilot's end would mean the strategy is replicable; 70 physicians staying put would mean the diagnostic-miracle story is doing more work than the workflow one.