Rapid identification of severe hypoglycemia and hyperglycemia in hospitalized patients using laboratory-integrated machine learning

Dysglycemia is common in hospitalized patients and is associated with adverse clinical outcomes. Although central laboratory glucose testing is the diagnostic reference, clinically relevant delays may occur between sample processing and result reporting, particularly for critical values requiring urgent intervention. We investigated whether early photometric reaction signals generated during routine enzymatic glucose measurement could enable earlier recognition of severe dysglycemia.We developed a laboratory-based machine learning model to predict plasma glucose concentration using early analyzer-derived photometric absorbance features during the enzymatic reaction. Model development was performed, and performance was evaluated using an internal test split and an independent external validation cohort. Regression performance was assessed using mean absolute error, root mean squared error (RMSE), and coefficient of determination. Clinical risk classification was evaluated for severe dysglycemia defined as glucose <54 mg/dL or >250 mg/dL.In the internal test set, the model achieved high agreement between predicted and measured glucose concentrations (RMSE 7.1 mg/dL; R² 0.990). In external validation, prediction accuracy remained high (RMSE 45.1-52.0 mg/dL; R² 0.904-0.928 across evaluated models). For severe dysglycemia alerts (< 54 mg/dL or >250 mg/dL), multivariable linear regression provided the highest sensitivity (98.6%) with specificity of 90.9%..Analyzer-derived photometric reaction signals can be leveraged to predict glucose concentration and identify severe dysglycemia before routine test reporting is finalized. This laboratory-integrated approach may enhance patient safety by enabling earlier recognition of critical glucose abnormalities and supporting timely clinical decision-making. This approach may shorten the time to recognition of critical glucose abnormalities and facilitate earlier intervention.