A machine learning model for hypoglycemia risk prediction in hospitalized patients with diabetes: development and validation

Le résumé fourni par la source

BACKGROUND: Hypoglycemia is a major obstacle for optimal glucose management in patients with diabetes. Existing prediction models primarily target non-hospitalized settings or use static variables, limiting applicability for hospitalized patients. We aimed to developed and validated a machine learning model predicting inpatient hypoglycemia using electronic medical records data, integrating dynamic clinical variables to improve accuracy and clinical utility. METHODS AND FINDINGS: We conducted a retrospective study of 37,966 inpatients with diabetes mellitus at Nanfang Hospital (2021-2022). After applying inclusion and exclusion criteria, 2,845 patients were included. Data preprocessing focused on analyzing potential predictors, including demographic characteristics, medication use, comorbidities, and laboratory parameters. Using a stepwise forward variable selection method based on XGBoost, we identified 10 optimal predictors. The cohort was split into training and testing sets at an 8:2 ratio. Predictive performance was assessed via AUC. Ten ML algorithms were evaluated, with CatBoost demonstrating the best performance (AUC = 0.85, PPV = 0.75, NPV = 0.89). CONCLUSIONS: Our ML-based predictive model for inpatient hypoglycemia shows robust performance and integrates readily available clinical parameters, offering significant potential for early risk identification and preventive intervention. Future research should focus on multicenter validation and seamless integration into clinical workflows to support dynamic risk assessment.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.