Related Experiment Videos
Multimodel Interpretable Machine Learning for Osteoporosis Prediction in Diabetes Using a Large Critical Care
Dongdong Cheng1, Wei Zhang1, Jingyu Chen1
1Orthopaedics Department, Nantong First People's Hospital.
Journal of Visualized Experiments : Jove
|July 20, 2026
Summary
Machine learning models can predict osteoporosis risk in diabetic patients using routine data. An interpretable ExtraTrees classifier, enhanced by SHAP, shows promise for early risk stratification.
Area of Science:
- Biomedical Informatics
- Machine Learning in Healthcare
- Bone Metabolism
Background:
- Diabetes mellitus increases skeletal fragility and fracture risk.
- Early osteoporosis (OP) identification in diabetic individuals is challenging.
- Machine learning (ML) offers potential for improved diagnostic tools.
Purpose of the Study:
- Develop and compare ML models for OP prediction in diabetic patients.
- Evaluate clinical utility and enhance transparency of predictive models using SHAP.
- Identify key predictors for OP risk in this population.
Main Methods:
- Trained and validated multiple ML algorithms (tree ensembles, gradient boosting, linear models) on MIMIC-IV data.
- Evaluated model performance using AUC, precision-recall, calibration curves, and DCA.
- Utilized SHAP for global and local feature interpretation.
Main Results:
- Ensemble tree-based methods outperformed other models.
- ExtraTrees classifier achieved highest performance (AUC=0.862, Avg. Precision=0.866).
- SHAP identified gender, age, and routine labs as key predictors; model showed good calibration and net clinical benefit.
Conclusions:
- An interpretable ensemble tree-based model effectively predicts OP risk in diabetic patients using routine clinical data.
- SHAP integration enhances clinical transparency.
- The model shows potential as a decision-support tool for early OP risk stratification.