Skip to content Skip to main navigation Report an accessibility issue
Anahita Khojandi

Sepsis Detection Algorithm Wins First Place

An alarm at a patient’s bedside is supposed to spur immediate, life-saving action. But when pre-emptive alarms are constantly sounding off, they blend into the background.

“Sometimes as you walk down the hall of an intensive care unit (ICU), there are dozens of alarms coming at you,” said ISE Associate Professor Anahita Khojandi. “They may be so prevalent that your brain starts to tune them out.”

Medical workers experiencing this “alarm fatigue” may not recognize when a patient is truly in need of attention until it’s too late. For sepsis, when the body overreacts to an infection and starts destroying its own tissues, the mortality rate increases by 8% for every hour that treatment is delayed.

“It’s all about detecting it early, early, early,” Khojandi emphasized, “but you need to balance that with the issue of alarm fatigue.”

Current state-of-the-art, machine learning-based sepsis detection algorithms sound the alarm when a patient’s heart rate, blood pressure, and other readings mimic what has been seen in former patients who developed sepsis. The problem is that many of these readings fluctuate, momentarily reaching the algorithms’ danger threshold even in patients who never develop sepsis.

Now the ISE team, which includes Khojandi, Professor Xueping Li, and their former student Zeyu Liu, has produced a revolutionary program that monitors the patient’s physiological variables in real time and develops an informed, changeable “belief” about a patient’s true but “hidden” sepsis status.

“Our model begins with the belief that the patient does not have sepsis,” said Khojandi. “As the bedside monitors gather more data, that belief may change.”

By partnering with clinicians at the UT Health Science Center and Emory University, the team has been able to train and test the new model using patient bedside data of unparalleled detail and precision.

“Only a handful of centers in the US collect this kind of data,” Khojandi said. “I cannot emphasize enough the importance of working with clinicians and the importance of the data they have been collecting.”

The developed program reduces false sepsis alarms by up to 28%, decreasing alarm fatigue and restricting the much-needed care to patients who really need it. This remarkable improvement has earned them the First Place Harvey J. Greenberg Research Award from The Institute for Operations Research and the Management Sciences (INFORMS).

The award will officially be conferred at the 2022 INFORMS Annual Meeting this October.

Contact

Izzie Gall (865-974-7203, egall4@utk.edu)