RAPiDS-funded project uses AI to create prediction models for postoperative sepsis
A patient is admitted exhibiting the symptoms of fast and shallow breathing, lightheadedness, shivers, sweats, and a change in mental status. These symptoms are general enough to pass as numerous maladies. But one thing sets them apart and narrows the diagnosis: The patient just had surgery. This could be postoperative sepsis.
Sepsis is a serious condition when the body responds improperly to an infection by targeting the body itself, harming the patient’s organs. If left untreated, this can progress to septic shock, where the rate of survival is greatly reduced.
“Nationally, postoperative sepsis has a much higher mortality rate for many reasons,” said Nicole Iovine, M.D., Ph.D., the chief hospital epidemiologist at UF Health Shands Hospital and clinical professor in the UF College of Medicine. “These reasons include a patient’s pre-existing medical conditions before surgery and the surgical procedure itself that they underwent, among many other variables.”
Nicole Iovine, M.D., Ph.D.
As an epidemiologist, Iovine mitigates the occurrence of infections and prepares organizational responses for disease outbreaks like COVID-19. But after becoming the chair of a sepsis committee in 2015, she and her colleagues also prioritized reducing rates of sepsis, including in the postoperative period. Now, the team is leveraging the power of artificial intelligence in their mission with assistance from the College of Medicine’s Quality and Patient Safety initiative, or QPSi.
“We had done many things, but we hadn’t yet really moved the needle in improving these rates,” Iovine said. “That’s why we wanted to take an AI approach to the prediction of postoperative sepsis with the aim of early intervention.”
QPSi’s approach to perioperative sepsis complements a long history of successful sepsis research programs across UF Health and the College of Medicine, where renowned scientists and clinicians from the UF Sepsis and Critical Illness Research Center, the Intelligent Clinical Care Center and more are working together to tackle this complex health care challenge.
Although sepsis rates in general have seen many improvements within UF Health Shands Hospital, higher rates of postoperative sepsis among patients were being reported, impacting performance in national quality indicators. This is when Iovine’s idea for a project that uses AI to decrease mortality rates by predicting postoperative sepsis came to fruition.
After her project was awarded a grant through the QPSi AI/QI Incubator’s RAPiDS program, Iovine began working with the AI/QI Incubator team to create an AI-powered prediction model.
“By offering informed pattern recognition based on millions of past patient records, our AI models have the potential to augment clinical judgment and may allow for earlier intervention in cases of postoperative sepsis, with the goal of improving patient outcomes,” said Roy Williams, M.P.H, biostatistics analyst III with QPSi’s AI/QI Incubator.
The model looks at a multitude of variables such as demographics, economic status, pre-existing conditions, lab values, and the type of surgery the patient underwent to predict if a patient is at higher risk of developing postoperative sepsis. This would then lead to creating a plan of action focusing on early intervention to decrease the chances of postoperative sepsis for the patient.
“Predicting sepsis is a complex challenge that demands extensive training data to identify subtle patterns,” said Paul Nickerson, QPSi’s AI/QI Modeling and Technical Operations manager. “Thanks to the robust data collection efforts of the Integrated Data Repository and the computational power of the HiperGator supercomputer, we can rapidly iterate and refine our AI models. These resources and collaborations are essential for advancing our ability to predict and ultimately improve patient outcomes.”
After about one year, several modeling innovations have been developed and are comparable to the state-of-the-art performance as published in the research. Now, efforts are being focused to further train the current model on specific populations’ real-time data, starting with cardiothoracic surgeries with the ultimate goal of implementation with real patients. This groundbreaking achievement would positively impact patient care not only within the hospital, but also throughout the state of Florida and beyond.
“QPSi offers such powerful tools that can directly impact the quality of care we deliver to our patients,” Iovine said. “I sincerely hope QPSi will continue to provide these services because there are many clinical problems, big and small, that could be potentially addressed by AI.”
Click on the video below to learn more about sepsis:
About the author:
Damarys Santacoloma, M.S.
COMMUNICATIONS MANAGER, QUALITY AND PATIENT SAFETY INITIATIVE
Damarys Santacoloma, M.S., graduated from Florida International University with a B.A. in philosophy and an M.S. in marketing. Before joining the University of Florida’s College of Medicine as the communications manager for the Quality and Patient Safety Initiative, or QPSi, she worked in a variety of fields, including the nonprofit sector, the startup space and in higher education, previously serving as the communications manager for UF Online. Her skills range from email marketing and customer relationship management, or CRM, to digital and print content creation and website maintenance. Now in her current role, Damarys assists in creating a communications strategy for the QPSi with an omnichannel approach in order to increase the initiative’s reputation and brand awareness.
