AI Healthcare
Physicians adopt clinical AI quickly — and still verify most outputs
By Arjun
·
3 Aug 11:48 AM
·
Via MedicalXpress
Surveys circulating in mid-2026 report roughly four in five U.S. physicians using AI in clinical workflows, while a large majority say they routinely validate outputs against bias and hallucination risk. Hospitals drafting ChatGPT-class policies are encoding that trust-but-verify habit rather than handing over autonomy. It matters because raw adoption metrics alone overstate how much clinical judgment has actually shifted to models. Caveat: self-reported survey behavior can diverge from what later chart audits and malpractice reviews reveal in practice.
Read the original →
Our summary is original writing; the full story belongs to MedicalXpress.
Sign in or create a free account to keep this story in a bookmark folder.
Israel’s Sheba Medical Center, through its ARC innovation arm, is deploying OpenAI’s ChatGPT for Healthcare / OpenAI for Healthcare across clinical and operational workflows — the company’s first large international hospital partnership. Clinicians get cited answers drawn from literature and Sheba’s own protocols under role-based access, while patient data stays isolated and final decisions remain with staff. Sheba already runs imaging and ER documentation AI; this adds a generative layer plus early research-model access. It matters as a blueprint for AI-powered hospitals abroad. Caveat: governance, liability and local validation will decide whether the template travels.
A NEJM AI study across RWJBarnabas Health found Epic’s Deterioration Index, paired with automatic rapid-response alerts, cut high-risk inpatient mortality from 23.1% to 18.6% — about an 18% drop in risk-adjusted odds — while raising response-team activations. Leaders stress earlier critical-care eyes, not autonomous treatment. Follow-on work targets patients whose risk scores are rising fast. It matters because an EHR-native tool already in wide use may translate beyond one health system. Caveat: single-network results still need broader replication before national standard-of-care claims.
OpenAI for Healthcare and ChatGPT Health — encrypted clinical assistants launched earlier for a small set of U.S. hospitals — are now entering Sheba’s network as the first major non-U.S. deployment. The pitch is evidence-grounded decision support and lighter admin load, not autonomous diagnosis. Sheba will feed approved pathways into the workspace and return real-world clinical feedback to OpenAI. It matters because specialized medical model lines are becoming a competitive front in regulated care. Caveat: early hospital partnerships are not the same as proven outcome gains across health systems.
Operational detail from the RWJBarnabas deployment shows the Epic Deterioration Index continuously reading vitals, labs and nursing assessments, recalculating risk every fifteen minutes and paging rapid-response teams at top tiers. Nearly half of intervention patients generated alerts; not every alert triggered a full team activation. It matters because workflow integration — timing, paging, staffing — explains outcome changes as much as the model itself. Caveat: hospitals without comparable response capacity may not see the same mortality shift from software alone.