Artificial Intelligence Learns Treatment Strategies for Hypotension

Congratulations to Dr. Tezcan Ozrazgat-Baslanti, Dr. Crystal Johnson-Mann, and Dr. Tyler Loftus for their project “OR-DRD-AI2020: The Artificial Intelligence learns optimal treatment strategies for hypotension in surgery” being selected as an awardee from UF Research, Artificial Intelligence Research Catalyst Fund. 

Photos of Dr. Tezcan Ozrazgat-Baslanti, Dr. Crystal Johnson-Mann, and Dr. Tyler Loftus

The Complication: Hypotension

In the United States, where the average American can expect to undergo seven surgical operations during a lifetime, each year 1.5 million patients develop a medical complication. Intraoperative hypotension (low arterial blood pressure during surgery) is a well-recognized modifiable risk factor for complications after surgery. The pathophysiology of postoperative complications is influenced by patients’ preexisting health, the type of surgery, the management of intraoperative events and especially those events relating to hypotension. 

Current practices of administering intravenous fluids and vasopressors during surgery as a treatment of hypotension are suboptimal. Patients and physicians make essential decisions on which diagnostic and therapeutic interventions should be performed or deferred. They make these decisions under time constraints and uncertainty regarding patients’ diagnoses and predicted response to treatment, which may lead to cognitive and judgment errors.  

The Role of Reinforcement Learning

Reinforcement learning has the potential to help deliver high-value surgical care through sound judgment and optimal decision-making. Reinforcement learning is a subfield of machine learning that identifies a sequence of actions to increase the probability of achieving a predetermined goal. The algorithm mimics human trial-and-error learning processes to calculate optimum recommendation policies.  

This project involves the use of detailed electronic health records from adult hospitalized patients undergoing surgery to determine optimal treatments for hypotension. We aim to develop and validate a deep reinforcement learning model that provides individualized and clinically interpretable treatment decisions. The model will produce data-driven decisions to help balance blood pressure during surgery and decrease postoperative complication risks in surgical patients. 

The Goal: Improving Patient Outcomes

Precise data-driven artificial intelligence tools can assist in identifying the best treatment strategies in the highly dynamic environment. Using these tools to reduce the risks associated with hypotension, clinicians can make more timely decisions to guide treatment and improve patient outcomes.