Nurse-driven early-warning AI was linked to a 35.6% lower risk of death in a trial; a grant to adapt it for children was canceled

June 2025 — An AI early-warning tool built on nurses' charting habits was linked to fewer hospital deaths in a large trial. Here's what it found, what it cost, and what went wrong.

Cost and outcome (as of June 11, 2025)

Trial: 1 year, 74 clinical units (37 intervention, 37 usual care) across 2 health systems, 60,893 encounters. (Nature Medicine)
Deaths: 35.6% decreased instantaneous risk of death (adjusted hazard ratio 0.64). (Nature Medicine)
Other results: shorter stays (the paper reported 11.2%; an April 2026 correction revised the adjusted rate ratio for length of stay to 0.96, see Later developments); 7.5% lower risk of sepsis; 24.9% higher risk of unanticipated ICU transfer. (Nature Medicine)
Pediatric extension: $499,984 from the American Nurses Foundation, 2025–2027. (ANF)
Trial cost: not disclosed.

How it works

When nurses worry about a patient, "they change how they measure the patient," CU Anschutz explained. CONCERN looks for those documentation patterns, rather than the measurements alone, and alerts the care team.

"Nurses are particularly skilled and experienced in detecting when something is wrong with patients under their care," said Sarah Rossetti, PhD, RN, of Columbia, who led the study, in Columbia's release. "When we can combine that expertise with AI, we can produce real-time, actionable insights that save lives."

She added: "The CONCERN Early Warning System would not work without the decisions and expert opinions of nurses' data inputs." Columbia said the tool detected deterioration "nearly two days earlier than traditional methods." CU Anschutz put it at "up to 42 hours earlier."

What the trial found

The one-year trial's coprimary outcomes were in-hospital death and length of stay, according to the paper (Nature Medicine). The paper reports a 35.6% decrease in the instantaneous risk of death among encounters on intervention units. That is a hazard measure, not a 35.6% cut in the number of deaths. CU Anschutz described the result as "a 12% increase in the number of patients who left the hospital alive."

The trial also found 24.9% more unanticipated transfers to intensive care. More ICU use carries costs, and the trial did not report them.

David Albers, an associate professor of biomedical informatics at the University of Colorado School of Medicine and the study's AI methodologist, said the patients CONCERN targets "don't appear super sick, so you have to work (the data) to be able to get to them in time." He added: "With sepsis, every hour past the time you become septic, you lose percentage points on your likelihood of survival."

The canceled grant

On June 11, CU Anschutz reported: "While the RO1 (a National Institutes of Health grant) that was focused on adapting CONCERN to high-resource pediatric hospitals was recently canceled in the beginning of its third year, the American Nurses Foundation this spring awarded the team a grant to extend CONCERN to lower-resourced pediatric settings." The team is adapting CONCERN at Children's Hospital Colorado, it said.

The ANF lists that award as "Optimization of Pediatric CONCERN EWS to non-AMC settings." Neither release said why the NIH grant was canceled.

The research group's plans reach beyond children's hospitals. Columbia's Department of Biomedical Informatics describes an aim to spread the model and its decision support, using a CONCERN toolkit, to the remaining hospitals within Mass General Brigham, Vanderbilt University Medical Center and Washington University Medical Campus. It does not give a timeline.

What went wrong

A funding gap. The pediatric NIH grant ended early; a smaller foundation grant covers lower-resourced settings.
More ICU transfers. A trade-off the paper reports but does not cost.
Financial interest. Columbia has filed a patent application, and some authors are named inventors, the paper says.
What we don't know yet: which NIH grant was canceled, why, and how much funding was lost.

Our follow-up question for Columbia

Which NIH award was canceled, and what work on high-resource pediatric hospitals has stopped?

Later developments (added Oct. 2026)

NIH RePORTER shows the adult trial's grant, R01NR016941, received a fiscal 2026 award of $793,262 in February 2026, with the project running through Jan. 31, 2027. That grant was not the one canceled. (NIH RePORTER)
On April 13, 2026, Nature Medicine published an author correction. It said a modeling error affected the length-of-stay and readmission results. The adjusted incidence rate ratio for length of stay changed from 0.91 (0.90–0.93) to 0.96 (0.93–0.99), with P = 0.015, and unadjusted mean stays changed to 6.7 days with CONCERN versus 7.4 days with usual care. The correction did not change the mortality result, and the journal said statistical significance was unchanged for all primary and secondary outcomes. The 11.2% figure in our Cost box is from the original paper. (Nature Medicine)

Sources

Nature Medicine, Sarah C. Rossetti et al., “Real-time surveillance system for patient deterioration: a pragmatic cluster-randomized controlled trial,” April 2, 2025. nature.com
Columbia University Irving Medical Center, “Nurses and AI Collaborate to Save Lives,” April 2, 2025. cuimc.columbia.edu
CU Anschutz, Chris Casey, “Nurses Power AI Model to Improve Care of Hospitalized Patients,” June 11, 2025. news.cuanschutz.edu
American Nurses Foundation, “Nurse-Led Research and Innovation,” accessed Oct. 25, 2026. nursingworld.org
NIH RePORTER, “Communicating Narrative Concerns Entered by RNs (CONCERN),” project 5R01NR016941-10, accessed Oct. 25, 2026. reporter.nih.gov
Columbia DBMI, “CONCERN Study,” accessed Oct. 25, 2026. dbmi.columbia.edu
Nature Medicine, “Author Correction: Real-time surveillance system for patient deterioration: a pragmatic cluster-randomized controlled trial,” April 13, 2026. nature.com

Disclosure: [not in copy] Editor placeholder