In a remarkable 36% reduction in the probability of hospital death, an inventive use of artificial intelligence that essentially watches patients via the eyes of the nurses providing care was made possible. That being said, the Trump administration quickly canceled funds to expand the research from adults to sick youngsters.
According to a study published in Nature Medicine, during a one-year trial of an AI system functioning in real-time at four hospitals in two prestigious organizations—NewYork-Presbyterian and Mass General Brigham—the length of stay decreased by 11.2% and the risk of hospital death for a patient under the AI system was 35.6% lower than “usual care.” In other words, according to the study, 31,000 people on average returned home nearly 15 hours earlier, and 84 people who could have died otherwise did not.
Sarah Rossetti, the study’s co-principal investigator, stated in an interview that their goal is to implement this method in every hospital in the nation and beyond. Rossetti is an Associate Professor of Biomedical Informatics and Nursing at Columbia University and a seasoned critical care nurse.
Rossetti’s academic affiliation was the reason behind the abrupt termination of the study’s National Institute of Nursing Research grant, which was intended to support further research and extension to include pediatric patients. Due to Columbia’s reported lackluster response to antisemitic protests on campus, the Trump administration withdrew $400 million in financing from the university, which resulted in the grant’s cancellation. Even though Columbia finally complied with the administration’s demands and Kenrick Cato, Rossetti’s co-PI, is a clinical informatics researcher at the University of Pennsylvania, there is no indication as to whether or when the funding would be reinstated.
The Front Lines Provided an Idea
The AI system’s concept dates back more than two decades, according to Rossetti, when she was a hospital nurse caring for severely ill patients and recording everything from vital signs to her own observations on lengthy paper flow sheets. She saw that more frequent visits to the patient’s room and more thorough descriptions were made by experienced nurses who were concerned. Depending on which of those two buckets you’re in, Rossetti pointed out that there are quite distinct approaches to care. You can be critically stable or critically unstable.
Rossetti and Cato, who had previously worked as informatics researchers, brought together a group of people from several universities to develop an artificial intelligence system that would turn what nurses saw “into a new data point that everyone could see and focus on.”
The end result was a system called the Communicating Narrative Concerns Entered by RNs (CONCERN) Early Warning System (EWS), which has a catchy acronym and a long, complicated complete name. CONCERN employs over 1,200 “ensemble-based models” that consider everything from the number of days the patient has been hospitalized, whether it is a weekday or a weekend, the time of year, the time of day or night, the nurse shift, and many other variables, in addition to what nurses record in the electronic health record. Patients at risk of deterioration were recognized by the resulting prediction model two days before the typical Modified Early Warning Score, which is based on lab data and vital signs.
A crucial factor in a real decrease in the death rate was that CONCERN gave nurses a tangible data point to replace their more ethereal “I’m worried” instinct, which occasionally manifests, occasionally is suppressed because nurses are reluctant to bother doctors, and occasionally, according to research, even when it is communicated quickly, is not promptly followed through on. When designing CONCERN, the researchers collaborated closely with physicians and nurses to integrate it with the current clinical workflow and explain how various inputs affected the predictions. As a result, when the system’s green light changed to an urgent red or warning yellow, the trust was strengthened and a quicker reaction was possible.
There was a 7.5% lower incidence of sepsis, a difficult-to-detect infection that can end in catastrophic organ failure if treatment is delayed, but more patients were unexpectedly transferred to the intensive care unit as a result.
One community hospital and one academic medical center from each of the two health systems were among the participating hospitals. A less sophisticated version of CONCERN, however, was independently verified by another research team using retrospective data at more than 200 hospitals, Rossetti informed. That shows consistent nurse documentation behavior and the system should work at all kinds of institutions in all kinds of situations, she said.
The team is working to make up the $500,000 grant that was intended to allow them to test the system at a children’s hospital, expanding the research to two additional adult hospitals under separate funding from the American Nurses Foundation, and preparing a limited number of academic research licenses that come with a suite of tools that enable a simple linking of any EHR to the CONCERN system using the FHIR interoperability standard, despite the remarkable decrease in patient deaths that has been shown thus far.
In the meantime, Columbia has filed for a patent that would allow for the kind of income flow required to aggressively implement the system across the country, particularly in places with little resources.
Although the researchers believe they have decreased costs while improving care safety, they have not yet established if saving lives results in financial savings. Today, the teaem is looking for money to work with Columbia’s business school to research cost effectiveness. Reduced duration of stay combined with shorter intensive care unit stays may or may not save overall health care expenses and increase hospital funding.
According to Rossetti, the two health systems are thinking about ways to extend CONCERN’s use to all of their hospitals, while the study’s participating institutions are still using it.
