Machine-Learning Can Help Anesthesiologists Foresee Complications

Pubblicato da Meba il

During surgeries, anesthesiologists must monitor the vital signs of patients and administer the proper doses of anesthesia at the right times. While managing these responsibilities in a high-pressure situation, it can be difficult to anticipate surgical complications. One issue that can arise is hypoxemia, a condition in which the blood oxygen levels of the patient become too low. Hypoxemia has been associated with serious consequences such as cardiac arrest, cerebral ischemia, and post-operative infections. Although anesthesiologists can monitor blood oxygen saturation in real-time, there are currently no reliable ways of predicting hypoxemic episodes perioperatively.

To address this issue, researchers at the University of Washington have developed a machine-learning system which they have called “Prescience”. Before the surgery begins, the system uses patient data, such as age and weight, to provide an estimate of the risk of an individual having a hypoxemic episode during the operation. Additionally, the system is able to predict hypoxemia at any point during the procedure by using real-time information from the patient’s vital signs. In their paper published in Nature Biomedical Engineering, the authors demonstrated that anesthesiologists were able to predict hypoxemic episodes 16 percent more accurately when they had access to Prescience compared to when they did not.

In addition to its predictive ability, Prescience is also able to provide explanations for its predictions, so anesthesiologists can better understand why a patient is at risk. “One of the things the anesthesiologists said was: ‘We are not really satisfied with just a prediction. We want to know why’,” reported Su-In Lee, senior author on the paper. After acquiring a dataset of 50,000 surgeries from the University of Washington Medical Center and Harborview Medical Center, Prescience found that the body mass index of the patient was one important preoperative feature helping to predict whether or not a patient would experience hypoxemia during surgery. During the operation, Prescience found that minute-to-minute blood oxygen levels were the most important predictive feature.

The authors plan to continue working with anesthesiologists to improve the system’s interface, as well as developing versions of Prescience which can predict other dangerous conditions.

Study in Nature Biomedical EngineeringExplainable machine-learning predictions for the prevention of hypoxaemia during surgery…