AI model helps predict patient decline, drive collaborative care

Editor's Note

An AI prediction model that uses near-real-time data to generate a patient risk score shows the promise of AI for helping physicians and nurses coordinate on patient care, according to findings published March 25 in JAMA Internal Medicine.

Performed by researchers at Stanford Medicine, the study examined an AI-driven prediction model that captures data from patients’ vital signs, electronic health records and lab results to generate a risk score. When the AI predicts that a patient is about to decline, it alerts relevant physicians and nurses. 

Specifically, the AI model predicts ICU transfers and other indicators of health decline. Once the alerts come in, the care team is mobilized to take action around adjusting care, with heavy involvement from nurses. 

Stanford Hospital runs the model on nearly 10,000 patients. Researchers found that incorporating the AI assessment led to a significant improvement in clinical outcomes, including a 10.4% decrease in deterioration events for patients considered to be almost high risk, including transfers to the ICU or rapid response events.

Researchers write that the AI model has improved physician and nurse conversations around patient care and improved clinical outcomes. However, they emphasize that further study is required to understand why a certain percentage of flagged patients (about 20%) still experience deterioration, how to further improve the model’s accuracy, and how to prevent alert fatigue.