Patients whose blood glucose levels spike during surgery are at heightened risk for poor overall outcomes. A new AI tool has proven effective at predicting, prior to surgery, which patients will have the problem while under the knife.
Such informed anticipation can help surgical teams plan ahead for optimal resource allocation and targeted glucose management in the OR for these patients.
A study documenting the tool’s development, testing and suggested applications is running Methods of Information in Medicine.
Senior author Bala Nair, PhD, of the University of Washington in Seattle and colleagues built and validated several separate prediction tools using a dataset of perioperative records from more then 6,500 patients.
Comparing the tools against one another, the team found all those using machine learning were more accurate than a conventional linear regression model.
The best of the machine-learning algorithms, an extreme gradient boosting model, had the smallest median prediction error and the narrowest interquartile error range.
The researchers implemented this model as a web application called “Hyper-G” and demonstrated its usefulness at the point of care.
“Machine learning models are able to accurately predict peak glucose levels during surgery,” Nair et al. concluded. “Implementation of such a model as a web-based application can facilitate optimal decision-making and advance planning of glucose management strategies.”