Hypoglycemia remains a major concern for individuals with type 1 diabetes mellitus (T1DM) during physical activity. A study published in Diabetologia developed a machine learning-based tool to estimate hypoglycemia risk at the start of exercise and to provide rapid, accessible risk assessment to support safer participation in exercise.

Data from four diverse studies were combined, comprising 16,430 exercise sessions from 834 participants aged 12 to 80 years who used various insulin delivery methods. The Extreme Gradient Boosting algorithm was applied to develop two predictive models: a comprehensive model and a simplified model designed for practical use.

The comprehensive model incorporated 406 variables and achieved a mean receiver operating characteristic area under the curve (ROC AUC) of 0.89. The simplified model, based solely on starting glucose level, exercise duration, and glucose trend arrows, achieved a comparable ROC AUC of 0.87 and performed consistently across different exercise types and insulin delivery methods.

The simplified model was translated into a user-facing traffic-light heatmap tool that displays hypoglycemia risk based on the three variables. This tool provides an accessible method for estimating hypoglycemia risk immediately before exercise and may support safer exercise participation, reduce hypoglycemic episodes, and encourage greater engagement in physical activity among individuals with T1DM.