'AI tool captures all-cause deteriorations, lowers risk of death by 35%'

 An AI tool that analyzes nurses’ notes for subtle clinical changes helped reduce patient risk of death by 35.6%, length of stay by 11.2% and sepsis risk by 7.5%, according to research published April 2 in Nature. 

In a yearlong, multisite study, researchers assessed the tool across 74 clinical units in two health systems. Among 60,893 hospital encounters, about half involved the early warning system and the other half did not. 

The system, dubbed COmmnuticating Narrative Concerns Entered by RNs (CONCERN), is a machine learning algorithm that uses real-time nursing surveillance notes and data patterns to detect all-cause deterioration risks. 

Other EWSs often “rely on late and noisy physiologic indicators of deterioration” such as lab results and vital signs, according to the researchers. In contrast, this tool leverages nurses’ “subtle, yet observable, clinical changes that may not be captured in physiological data or well displayed in EHRs,” including small changes in mental status from baseline or slower recovery of arterial blood pressure after turning a patient. 

Every hour, the CONCERN EWS signals mentions of concern in nurses’ narrative tones, as well as their increased surveillance. For example, the tool monitors for assessments performed at uncommon times. 

The model then assigns a deterioration risk score of low, increased or high. Scores are updated hourly and presented in the EHR. 

Across the two systems, 28 acute care units and nine ICUs used the CONCERN tool, while 25 ACUs and 12 ICUs performed usual care. 

In addition to the above reduced risk results, units that used the EWS also saw a 24.9% increase in in-hospital risk of unanticipated ICU transfers. 

“Our study demonstrates that nursing surveillance patterns are a valuable signal to predict deterioration of hospitalized patients,” the researchers wrote in conclusion. 

New York City-based Columbia University Irving Medical Center led the research, which also involved Somerville, Mass.-based Mass General Brigham, Nashville, Tenn.-based Vanderbilt University Medical Center, and St. Louis-based Washington University Medical Campus. 

Further research will evaluate the predictive model’s capabilities in other hospital units and inpatient populations, including pediatrics.

https://www.beckershospitalreview.com/patient-safety-outcomes/ai-tool-captures-all-cause-deteriorations-lowers-risk-of-death-by-35/