Performance of machine learning-based prediction models for hypoglycemia in Chinese patients with diabetes: A systematic review and meta-analysis.
PLoS OneSep 22, 2026 (epub)
Clinical ResearchEndocrinology
Jinhua Yan, Yanping Song, Yangmei Du et al.
✦ AI-curated · Sources linked
In 30 seconds
This systematic review and meta-analysis evaluated machine learning models for predicting hypoglycemia in Chinese diabetes patients. Thirteen studies were included, revealing a pooled hypoglycemia prevalence of 25% and an overall area under the receiver operating characteristic curve (AUC) of 0.90, indicating strong predictive performance.
Key findings
- Pooled prevalence of hypoglycemia was 25% (95% CI: 17%-33%).
- Overall pooled AUC for prediction models was 0.90 (95% CI: 0.87-0.93).
- Subgroup AUCs included 0.89 for XGBoost and 0.81 for decision tree models.
- Common predictors included age, insulin use, and HbA1c levels.
Why it matters
Understanding the predictive performance of machine learning models for hypoglycemia can enhance early risk identification in diabetes management, potentially improving patient outcomes and safety.
What to keep in mind
Methodological limitations and insufficient validation were noted in many included studies.