M.L. Millenson. (2025). AI Uses What Nurses See And Death Risk Plunges 36% – But Grant Pulled. Forbes Magazine.

This article highlights a significant innovation in healthcare. The artificial intelligence system, Communicating Narrative Concerns Entered by RNs (CONCERN) Early Warning System (EWS), analyses nurses’ observations. The system predicts patient deterioration. A study across four hospitals revealed a remarkable 36% reduction in in-hospital deaths with the use of CONCERN. There was also an 11.2% decrease in length of stay. The AI’s ability to translate nurses’ insights into actionable data allowed for earlier interventions. This capability reduced the risk of conditions like sepsis.

It significantly impacts patient mortality rates in hospital settings. It leverages the patterns in nurses’ observations. It also uses the nuances within their documentation in the electronic health record (EHR).

Here’s a breakdown of how it works:

• Analysing Nursing Surveillance Patterns: Many traditional early warning systems primarily rely on physiological data. They focus on vital signs and lab results.
• CONCERN EWS focuses on nurses’ surveillance documentation patterns. This includes analysing the metadata of their EHR activities.
• Increased frequency of assessments: For example, how often a nurse checks a patient’s respiratory rate.
• Assessments at uncommon times: Such as checking vital signs in the middle of the night for a non-ICU patient.
• Nursing medication administration interventions: Like not administering a scheduled medication because the patient is unstable.
• Incorporating Narrative Concerns: The system also uses natural language processing to identify mentions of concern within nurses’ narrative notes. This captures the qualitative aspects of a nurse’s assessment that might not be present in structured data.
• Identifying Early Deterioration: Patterns of increased surveillance and documented concern indicate a nurse’s clinical intuition. This intuition suggests that a patient’s condition is worsening. CONCERN EWS processes this data through its machine learning algorithm. It can identify patients at risk of deterioration up to 42 hours earlier than models relying on physiological indicators.
• Providing Actionable Insights: The system calculates a deterioration risk score. This score can be green, yellow, or red. It is displayed to the care team in the EHR. This provides a concrete data point. It can prompt action. This happens even when a nurse’s concern might be more of a “feeling”. The system also offers a detailed breakdown of the risk score. It includes contributing factors. Trends over time are also shown. This enhances transparency and explainability.
• Facilitating Timely Interventions: The earlier prediction of deterioration gives clinicians greater lead time to perform necessary interventions. This can include closer monitoring and changes in treatment. The study results showed an increased rate of unanticipated intensive care unit (ICU) transfers. The study found a 24.9% increase in the instantaneous risk of unanticipated ICU transfer in the intervention group.
• Improving Patient Outcomes: By enabling earlier interventions, CONCERN EWS was shown to have a significant positive impact on patient outcomes. The study demonstrated a 35.6% decrease in the instantaneous risk of in-hospital mortality for patients monitored by the AI system compared to usual care. There was also a 7.5% decreased instantaneous risk of sepsis in the intervention group.

CONCERN EWS acts like a “second pair of eyes.” It leverages the expert knowledge and real-time observations of nurses. This helps identify subtle signs of deterioration that traditional systems might miss. The care team detects issues early. Their timely actions are key factors in the significant reduction of patient mortality rates observed in the study.

Research Articles

Rossetti, S.C., Dykes, P.C., Knaplund, C. et al. Real-time surveillance system for patient deterioration: a pragmatic cluster-randomized controlled trial. Nat Med (2025).

Rossetti, S. et al. Healthcare process modeling to phenotype clinician behaviors for exploiting the signal gain of clinical expertise (HPM–ExpertSignals): development and evaluation of a conceptual framework0. J. Am. Med. Inform. Assoc. 28, 1242–1251 (2021).

Additional Reading on Nursing Concern or Worry

Byrne, A. L., Massey, D., Flenady, T., Connor, J., Chua, W. L., & Lagadec, D. L. (2025). When Nurses Worry: A Concept Analysis of Intuition in Clinical Deterioration. Journal of Advanced Nursing.