Two-year retrospective evaluation found the Duke Deterioration Index outperformed a widely used vendor deterioration index, with the strongest results when paired with NEWS 2.

Vega Health, a health AI company that partners with health systems to integrate, scale, and monitor AI solutions, today released a Retrospective Evaluation of Adult Inpatient Deterioration Models, a local evaluation of deterioration models on two years of data from a multistate health system. The report is part of Vega's work to publicly benchmark health AI models, bringing a new level of transparency to an industry where health systems often choose AI tools without seeing how they perform on their own patients.

It is also the first evaluation of the Duke Health Deterioration Index outside of the Duke Health System.

Across 118,063 patient encounters at five hospitals, the Duke Deterioration Index outperformed a widely used vendor deterioration index at every possible alert threshold. At the same alert rate, the Duke Deterioration Index caught 55% more deterioration events than the vendor model. With a four-hour snooze on repeat alerts, the Duke Deterioration Index needed three-quarters fewer alerts to catch the same share of events: about six false alarms for every true positive, compared with 28 for the vendor model.

Key findings include:

• More accurate risk prediction. The Duke Deterioration Index was significantly more accurate than both the vendor deterioration index and NEWS 2 at identifying which patients would deteriorate, and it outperformed both across all possible alert thresholds.
• More events caught with the same number of alerts. Set to alert as often as the vendor deterioration index or NEWS 2, the Duke Deterioration Index flagged substantially more patients who went on to deteriorate.
• Fewer alerts for nurses. To catch the same share of deterioration events as the other models, the Duke Deterioration Index needed far fewer alerts. A four-hour snooze on repeat alerts for the same patient reduced alert burden further without missing additional events.

The strongest result came from pairing the Duke Deterioration Index with NEWS 2, a rules-based score nurses use at the bedside, offering new evidence that health systems don't have to choose between the two. Patients flagged high-risk by both tools were nearly three times as likely to deteriorate as those flagged by NEWS 2 alone. Roughly a third of deterioration events never triggered a medium- or high-level NEWS 2 alert, and the Duke model would have flagged more than half of them. The health system is now considering combining the Duke Deterioration Index with a rules-based score for implementation.

"Bedside nurses and rapid response nurses need different things from a deterioration signal. NEWS 2 tells a bedside nurse what's changing right now,” said Chris Provan, AI Evaluation Lead, Vega Health. “The Duke Deterioration Index helps a rapid response nurse decide which patients to check on first. Seeing how the two performed together on this system's data gave the team a practical path forward."

"Health systems usually choose AI tools based on vendor claims or a validation study run at another institution,” said Mark Sendak, CEO and co-founder, Vega Health. “Testing those tools against a system's own patients before go-live changes the conversation, and in this case it changed which configuration the clinical team wanted to pursue."

The evaluation was made possible by the Vega Health Platform, healthcare's AI enablement layer, which pulls historical and real-time data from the EHR and standardizes it so health systems can see how their data feeds and AI models are performing. Through its marketplace, Vega licenses models from developers including Duke Health and the Parkland Center for Clinical Innovation (PCCI), giving health systems more than one validated option to evaluate for the same clinical problem.

"External model validation used to be a two-year, grant-funded undertaking that required walking partners through every step," said Suresh Balu, Executive Director, Duke Institute for Health Innovation. "Vega Health delivered a full external evaluation of the Duke Deterioration Index in two weeks. This unprecedented speed not only validates Duke’s current work, but directly informs how we design future models for broad real-world deployment.”

Duke Health had no access to the health system's data and no role in the analysis, including which data and comparison models were used.

Results will likely vary for other health systems based on their patient populations and workflows, which is why site-specific benchmarking matters. The full report, including subgroup performance, is available at https://www.vegahealth.com/index/retrospective-evaluation-of-adult-inpatient-deterioration-models/.

About Vega Health

Located in the heart of Durham, North Carolina, Vega Health is the trusted partner for health systems working to identify, implement, monitor, and scale AI solutions aligned with their goals. Built by a group of experienced healthcare and technology operators, Vega Health helps healthcare organizations cut through the AI noise and start generating value from validated AI solutions that put patient care at the center of operations. The company’s proprietary platform, curated marketplace, and practical experience are essential infrastructure for health systems looking to generate measurable impact from healthcare AI. To learn more, visit www.vegahealth.com.

View source version on businesswire.com: https://www.businesswire.com/news/home/20260928225606/en/

At the same alert rate, the Duke Deterioration Index caught 55% more deterioration events than a widely used vendor model.