Development and evaluation of a machine learning prediction model for short-term mortality in patients with diabetes or hyperglycemia at emergency department admission

Comprehensive Summary

This study by Wändell and colleagues developed a machine learning (ML) model to predict 30-day mortality among patients with preexisting diabetes admitted to emergency departments. Using data from the Region Skåne catchment area in Sweden - including 632,744 patient visits between 2017 and 2018 - the researchers analyzed clinical variables such as laboratory tests, physiological parameters, surgical procedures, and medication records. The model employed stochastic gradient boosting (SGB) to estimate each patient’s risk of death within 30 days of triage. Results indicated that for male patients, laboratory tests reflecting infection and ketoacidosis were the most predictive variables. For female patients, laboratory results were also key predictors, though the model showed lower positive predictive values (PPV), suggesting a tendency toward false positives. Overall, the SGB model performed well in identifying patients at low mortality risk but tended to overestimate mortality among moderate-risk patients. The study demonstrates the potential of ML in risk stratification for diabetic patients in emergency settings but highlights the need for further refinement and validation before clinical deployment.

Outcomes and Implications

This study underscores how machine learning can support emergency care by rapidly identifying diabetic patients at low risk of short-term mortality. Given the growing global burden of diabetes and the high rates of hospitalization associated with it, such predictive tools could help clinicians allocate resources more efficiently and prioritize care for those most in need. While the current model tends to overpredict risk for moderate-severity cases, its strong accuracy for low-risk patients makes it a valuable decision-support aid. Future refinements could improve calibration across risk levels and expand integration into triage systems, ultimately supporting faster, more data-driven emergency care for patients with chronic disease.