A machine learning model developed by researchers at the University of Pittsburgh and the University of Pittsburgh Medical Center (UPMC) is capable of flagging patients who are at a high risk of surgical complications. By analyzing electronic medical records, the software identifies patients who could benefit from personalized coordinated care or prehabilitation to improve their surgical outcomes.
To train the algorithm, the researchers utilized medical records from over 1.2 million surgical patients. The model was specifically focused on predicting whether patients would experience complications such as strokes, heart attacks, or other major cerebral or cardiac events after surgery. According to the researchers, their model outperformed the American College of Surgeon’s National Surgical Quality Improvement Program (ACS NSQIP), which relies on manual data entry.
Approximately 4.2 million people worldwide die each year due to surgical complications within 30 days of a procedure. Prior to the COVID-19 pandemic, surgical complications were the third leading cause of death in the United States. Additionally, these complications contribute to significant healthcare costs, with each case costing hospitals over $11,000 and totaling more than $31.3 billion nationally annually.
Among the 15% of patients who experience surgical complications, high-risk patients constitute about half. The researchers emphasize the importance of improving the health of these high-risk patients before surgery to lower mortality rates and reduce healthcare costs. However, identifying high-risk patients can be challenging for busy clinicians who need to integrate a vast amount of health data and perform additional testing and clinical assessments.
With their algorithm, the researchers aimed to create an easy-to-use model that provides an automated and accurate risk assessment to the healthcare team quickly, utilizing existing data in the electronic health record. After training the algorithm, the model was validated against 200,000 UPMC surgical patients and then implemented across 20 UPMC hospitals. Clinicians can run the model anytime on-demand, in addition to the daily automated reviews.
The researchers are continuously working on improving the software. Their future plans include incorporating the ability to predict the likelihood of sepsis, respiratory issues, and other post-surgery complications that result in extended hospitalization.
The development of this machine learning model represents a significant advancement in identifying high-risk patients who could benefit from targeted interventions before surgery. By leveraging this technology, healthcare providers can enhance patient outcomes, reduce mortality rates, and cut healthcare costs. Further improvements to the software will expand its predictive capabilities, enabling more comprehensive pre-surgical care planning and potential complications management.
