February 12, 2020

'Hypothesis generator': How these 2 hospitals use AI to reduce sepsis

Daily Briefing

    Sepsis kills one in five people annually worldwide, and it can be very difficult to detect. That's why major health systems like Geisinger and the Cleveland Clinic are turning to machine learning to detect and treat sepsis early, John McCormick reports for the Wall Street Journal.

    New infographic: The 5 question AI vendor evaluation guide

    Using AI to fight sepsis

    The machine learning systems being used at Geisinger and Cleveland Clinic, rely on the de-identified data of thousands of sepsis patients from the health systems' EHRs, McCormick reports. That data include information on patients' demographics, medical conditions, medical histories, and lifestyle choices. The developers also pulled data from the patients' hospital stays, including the patients' blood pressure and white blood cell levels.

    The developers inputted the data into the machines, which then found patterns that indicated when a patient was at a high risk of developing sepsis. The developers then created risk scores for different patient types and conditions, and programmed the systems to alert providers if a patient is at a high risk for sepsis, McCormick reports.

    Chirag Choudhary, a specialist in pulmonary and critical care at the Cleveland Clinic, said, "With the risk-prediction algorithm, the hope is to identify people six to 12 hours before clinical determination happens."

    The Clinic's system, which was developed in partnership with Jvion, includes data on which treatments are most effective for different types of patients, according to Shantanu Nigam, Jvion's CEO. That allows the system to also suggest treatment options when a sepsis alert is triggered.

    Geisinger's system, which was developed in partnership with IBM, currently is in its research stage but is showing positive results, according to Seth Dobrin, IBM Cloud and Cognitive Software's chief data officer.

    IBM and Geisinger tested the system on the EHRs of 10,600 sepsis patients, 25% of whom died while in the hospital. During the test, the system was able to accurately predict, within tenths of a percentage point, which patients would become septic, Dobrin said.

    The 'opportunities' that come with AI

    Donna Wolk, division director for molecular and microbial diagnostics and development at Geisinger, said the health system's sepsis program highlights the enticing prospects AI presents in health care.

    "Machine learning is a hypothesis generator; it gives you opportunities to explore further, dig deeper and intervene on behalf of our patients," Wolk said. "That's what's so exciting about this work."

    Christopher Murray, a professor at the University of Washington in Seattle and lead author on recent studies on sepsis deaths, said while using machine learning software to identify sepsis still isn't widespread, "it's a very worthwhile direction to go" (McCormick, Wall Street Journal, 1/28).

    Have a Question?

    x

    Ask our experts a question on any topic in health care by visiting our member portal, AskAdvisory.

    X
    Cookies help us improve your website experience. By using our website, you agree to our use of cookies.