AI-Powered Tool Uses Nurses' Notes to Predict Patient Deterioration, Reducing Mortality and Hospital Stays

Researchers have developed an AI-powered tool called the CONCERN Early Warning System (CONCERN EWS) that can analyze nurses' shift notes to detect early signs of patient deterioration. This innovative approach has shown significant promise in improving patient outcomes and reducing hospital stays. In initial clinical trials involving over 60,000 patients from 2020 to 2022, CONCERN EWS helped lower the risk of patient death by more than 35% and reduced the average hospital stay by over half a day. Additionally, the system decreased the risk of sepsis among hospitalized patients by 7.5%. If these results can be replicated on a larger scale, hospital systems will have a powerful tool to enhance patient safety and control healthcare costs. The study, published in April in Nature Medicine, explains how the machine-learning algorithm operates. Lead researchers from Columbia University and the University of Pennsylvania designed the system to prioritize the nuanced observations made by nurses during frequent patient interactions. These insights often catch subtle changes in a patient’s condition that might otherwise go unnoticed. CONCERN EWS works by analyzing the natural language in nurses' electronic health record (EHR) entries, but its primary innovation lies in interpreting the metadata associated with these notes. For example, a nurse might note that a patient's skin color has changed, they appear lethargic, or seem cognitively off. Based on these observations, the nurse might decide to check on the patient more frequently or delay certain medications. CONCERN EWS identifies patterns in these decisions, such as when assessments are performed more often or at unusual times, to predict potential health issues. Sarah Rossetti, an associate professor of biomedical informatics and nursing at Columbia University and lead author of the study, highlighted that the ML model helped reduce hospital stays by an average of 11%. By interpreting nurses' detailed and timely observations, the system can alert care teams to signs of trouble up to 42 hours earlier than conventional methods. This early detection provides crucial time for interventions before a patient's condition worsens. To develop the algorithm, the research team utilized NVIDIA RTX A2000 12GB Graphics Cards. The system was tested in four hospitals across two hospital systems in Massachusetts and New York, where it proved effective in identifying at-risk patients well before standard protocols would have recognized the signs. In May, the success of CONCERN EWS earned the research team one of three prestigious "Reimagining Nursing Initiative" grants awarded annually by the American Nurses Foundation. With a total grant pool of $1.5 million, the team plans to collaborate with Children’s Hospital Colorado to develop and evaluate a pediatric version of the system in community hospitals. Kenrick Cato, a professor of informatics at the University of Pennsylvania and co-lead of the project, explained that the next phase will focus on expanding the tool's capabilities and ensuring it can benefit a broader range of patients. The team is optimistic that their approach will continue to demonstrate its value in enhancing patient care and reducing hospitalization durations. Additional news coverage and videos about CONCERN EWS provide further insights into the technology and its potential impact on healthcare.