Northwell Health’s Feinstein Institutes for Medical Research has secured a $3.4M National Institutes of Health grant to build machine learning tools that catch delirium in older hospital patients, a condition clinicians miss more than 75 percent of the time even where screening programs exist.
Delirium is an acute disturbance of mental ability, not ordinary confusion. Falls and longer stays follow it, along with memory trouble that outlasts the admission, and it counts as one risk factor for Alzheimer’s disease and related dementias.
Three pieces of the build
Liron Sinvani, director of the Geriatric Hospitalist Service at Northwell, leads the project with Alex Makhnevich of the Zucker School of Medicine and Theodoros Zanos, a professor in the Division of Health AI.
The work splits three ways. One model identifies patients arriving at the emergency department already delirious. A second predicts which admitted patients will develop it, a group that can reach 40 percent of admissions. Third, the team will build the models out with input far from the lab, joining physicians, nursing staff and hospital administrators who need the outputs to fit existing routines.
Sinvani described delirium as a medical emergency clinicians routinely miss, one whose consequences for patients do not fade quickly. In her view the award lets the team reverse that pattern with machine learning, catching the condition early or forecasting it outright and giving older adults better protection for their brain health.
The data behind the models
The foundation is preliminary data drawn from what the institutes call the biggest inpatient wearable dataset assembled in the US, a Division of Health AI effort backed by Scott and Debbie Rechler.
Zanos earlier led a study, reported in Nature Communications, in which an AI-driven wearable system forecast patient deterioration as much as 17 hours ahead of clinical recognition.
