Liu receives Harvey J. Greenberg Research Award for sepsis prediction research

Zeyu Liu's research on early sepsis prediction using integrated machine learning has been recognized by the Institute for Operations Research and the Management Sciences Computing Society. 

Zeyu Liu, assistant professor of industrial and management systems engineering at the Benjamin M. Statler College of Engineering and Mineral Resources, received the Harvey J. Greenberg Research Award from the Institute for Operations Research and the Management Sciences Computing Society.  

Story by Brittany Furbee, Communications Specialist
Photos by Paige Nesbit, Director of Marketing

Benjamin M. Statler College of Engineering and Mineral Resources

The Harvey J. Greenberg Research Award recognizes outstanding contributions that exhibit the promise of making a significant impact through computation in the areas of operations research, management science and analytics. Liu was selected for the award due to his research on early sepsis prediction using integrated machine learning.  

Sepsis is triggered by the body's extreme response to an infection and can be life-threatening. According to Liu, existing sepsis prediction algorithms suffer from high false-alarm rates.  

“Sepsis is among the deadliest conditions for ICU patients,” said Liu. “I am working on bringing the best models from both operations research and machine learning fields to improve what we can do to help patients survive.” 

Liu’s proposed sepsis prediction model reduces false-alarm rates and improves prediction accuracy. His method is calibrated and tested using high-frequency physiological data collected from bedside monitors in the ICU. Using this data, he combines machine learning models with a mathematical model called partially observable Markov decision process.  

The POMDP utilizes trained learning to gather information from mistakes produced from the machine learning model and then learns from these errors which increases sepsis prediction accuracy over time. Additionally, the model includes a unique wait-and-see decision function, meaning the model cannot tell if a patient has sepsis or not and more information is needed to generate an accurate assessment. The model's ability to relay this information to researchers is the key to reducing the rate of false alarm sepsis diagnoses. 

“Patients in the ICU and medical staff will benefit the most from this research,” said Liu. “We are able to provide more accurate alarm systems for ICU patients than other machine learning based methods. We are also able to reduce a great number of false alarms, which prevents the medical staff from being overwhelmed by them. It is a promising way to reduce alarm fatigue in the ICU.”  

His findings are outlined in the paper, “A Machine Learning–Enabled Partially Observable Markov Decision Process Framework for Early Sepsis Prediction,” which was recently published in the INFORMS Journal on Computing. 

Liu was recognized for his achievement during the 2022 INFORMS Annual Meeting that was held on October 16-19, in Indianapolis, Indiana.  

---