Skip to main content
Update Location

My Location

Update your location to show providers, locations, and services closest to you.

Enter a zip code
Select a campus/region

UF researchers create algorithm that accurately predicts surgical complications

Azra on a computer with someone else

University of Florida researchers have developed a computer algorithm that is more accurate than a doctor when it comes to predicting surgical complications.

An artificial intelligence system successfully tested at UF is a dual breakthrough: It can reduce risks to patients by forecasting surgical complications while also helping surgeons optimize their operating-room strategies.

“We recognized in a lot of patients that there is a window of opportunity before, during and after surgery to prevent complications from happening in the first place,” said Azra Bihorac, M.D., a professor of medicine and surgery in the UF College of Medicine, part of UF Health.

The findings by Bihorac and her collaborators in the UF colleges of Engineering and Medicine were published recently in the journals Annals of Surgery and Surgery. Surgery and its risks are pervasive: The average American can expect to have seven operations during his or her life and about 1.5 million U.S. patients develop a medical complication from a surgery each year.

The algorithm works by collecting and analyzing real-time patient data from electronic health records. It then calculates the risk of eight major postoperative complications, including acute kidney injury, sepsis, and cardiovascular and neurologic issues. It also estimates the probability of death up to two years after surgery.

Putting the system into real-time use for UF Health patients requires new technology that is now being acquired, Bihorac said.

An algorithm improves on the current methods for predicting surgical complications, which Bihorac said are neither automated nor particularly sophisticated.

“Now, it really relies on your judgment as a physician. More and more, it’s been shown that human judgment isn’t always the best predictor,” she said.

A machine-learning algorithm that produces more accurate risk assessments helps both patients and surgeons, Bihorac said. Evidence of multiple higher risks may prompt a doctor-patient conversation

about whether surgery is truly appropriate. Knowing that a patient is at high risk of complications allows surgeons to be more strategic in the operating room, such as using individualized blood pressure management or avoiding medications that can damage the kidneys.

The algorithm’s real-time monitoring can also be beneficial, Bihorac said. If a patient who enters surgery as a low-risk case has unexpected bleeding, the algorithm notes the elevated risk and alerts doctors to consider sending them to the intensive care unit after the procedure. It can also help smooth the patient’s transition from operating room to hospital bed.

“We can personalize this for every step of the treatment process. The care team may change during the patient’s stay but the handoff can be seamless,” she said.

To test the algorithm’s accuracy, the researchers compared its preoperative risk assessments with those done by 20 physicians. Each physician evaluated eight to 10 cases and estimated the risk of six common surgical complications.

The algorithm predicted the risk of surgical complications more accurately than its human counterparts, the researchers found. Compared with the algorithm, the physicians underestimated the risk of an intensive care unit stay and acute kidney injury and overestimated the risk of death, sepsis and cardiovascular complications. More broadly, physicians were more likely than the algorithm to underestimate complications that actually occurred and overestimate ones that did not happen.

“We showed that the algorithm performed better than the surgeons for a majority of the complications. It was more accurate and misclassified fewer patients,” Bihorac said. “An algorithm is not going to replace physicians, but it’s going to make use of patient data that is far too large and complex for a physician to analyze.”

In the second study, researchers tested the algorithm’s accuracy at calculating the risk of eight major surgical complications. Using a database of 51,457 surgical patients at UF Health Shands Hospital between 2000 and 2010, the algorithm predicted complications with accuracy ranging from 70 percent for wound complications to 86 percent for sepsis.

Correctly forecasting postsurgical complications does more than just help patients, the researchers noted. These complications can lead to adverse patient outcomes, so accurately predicting and preventing complications can save money, they said.

Next, the researchers want to further test the algorithm in different patient populations and expand its use beyond preoperative analysis. The researchers are also seeking a grant to develop a simplified version of the risk algorithm that could be used by patients.

A group of 22 engineering and medical researchers helped to develop and test the algorithm. Significant contributions and support for the project were provided by Xiaolin Li, Ph.D., a professor in the department of electrical and computer engineering and co-principal investigator on the grant, UF Health Chief Data Officer Gigi Lipori and the UF Health information technology team, Bihorac said. Funding was provided by the National Institutes of Health and UF’s Clinical and Translational Science Institute.

About the researcher: Azra Bihorac, M.D., is the R. Glenn Davis professor of medicine, surgery and anesthesiology in the UF College of Medicine’s department of medicine, division of nephrology, hypertension & renal transplantation.

About the author

Doug Bennett
Science Writer, Editor

For the media

Media contact

Matt Walker
Media Relations Coordinator (352) 265-8395