Vega Health has released a two-year evaluation of adult inpatient deterioration models, the first public test of the Duke Deterioration Index beyond the health system that built it.
The review drew on 118,063 patient encounters at five hospitals inside a multistate network. Against a common commercial deterioration index and the NEWS 2 early-warning score, Duke’s model came out ahead on accuracy. The edge held no matter where the alert threshold was set.
Alert burden told the sharper story. Tuned to fire as often as the commercial tool, Duke’s index surfaced 55% more patients who later declined. Nurses who snoozed repeat alerts for four hours saw the gap widen. Duke’s model raised roughly six false alarms for each real event. The rival product raised 28.
Vega, which helps health systems integrate, scale and monitor clinical AI, wants public benchmarking to become routine. Hospitals routinely pick tools without proof they work on their own patient populations, the company said.
Fewer false alarms matters at the bedside. Alert fatigue is a common reason deterioration tools get switched off, and a model that cries wolf rarely earns nurse trust. Vega licensed AI models from the Parkland Center for Clinical Innovation earlier this year.
