Machine Learning Model Accurately Identifies High-Risk Surgical Patients

Researchers and physicians at the University of Pittsburgh (Pittsburgh, PA, USA) and UPMC (Pittsburgh, PA, USA) developed this model by training the algorithm on the medical records of over 1.25 million surgical patients. The focus of the model was on mortality and the occurrence of major cerebral or cardiac events, such as stroke or heart attack, following surgery. The model was then validated using the records of another 200,000 surgical patients from UPMC. After validation, the model was implemented across 20 UPMC hospitals. Each morning, the program reviews the electronic medical records of patients scheduled for surgery and flags those identified as high risk. This alert enables clinical teams to improve care coordination and introduce prehabilitation measures before surgery, such as healthier lifestyle choices or a referral to the UPMC Center for Perioperative Care, thus lowering the risk of complications. Clinicians can also activate the model on demand at any time.

Additionally, the research team compared their model with the industry standard, the American College of Surgeon’s National Surgical Quality Improvement Program (ACS NSQIP), to further gauge its effectiveness. The ACS NSQIP, used at hospitals nationwide, requires manual input of patient information by clinicians and is unable to make predictions if data is missing. The researchers found their model to be more effective at identifying high-risk patients than the ACS NSQIP. As the model continues to be fine-tuned and developed, the researchers plan to train the program to predict the likelihood of other complications, such as sepsis and respiratory issues, that often result in prolonged hospital stays after surgery.

“We designed our model with the health care worker in mind,” said Aman Mahajan, M.D., Ph.D., M.B.A., chair of Anesthesiology and Perioperative Medicine, Pitt School of Medicine, and director of UPMC Perioperative and Surgical Services. “Since our model is completely automated and can make educated predictions even if some data are missing, it adds almost no additional burden to clinicians while providing them a reliable and useful tool.”

Related Links:
University of Pittsburgh 
UPMC