A multicenter Japanese study used gradient-boosted decision tree models to predict short-term blood glucose levels in patients with hyperglycemic crisis after continuous intravenous insulin infusion. The direct prediction model achieved R² values of 0.853 and 0.802 in internal and external validation, respectively, but only 45.3% of predictions in external validation were within 50 mg/dL of the measured values. Models predicting reductions in blood glucose had limited generalizability across institutions. This retrospective study did not demonstrate improved clinical outcomes or establish a safe automated dosing strategy, and broader prospective validation is still needed.
Source:EMJ · Read Original
Leave a Reply