Machine learning prediction for menopause women with low bone mass: a multicenter and retrospective study

Area of Science:

• Gerontology
• Medical Informatics
• Bone Metabolism

Background:

• Postmenopausal osteoporosis poses a significant fracture risk.
• Early detection of low bone mass (LBM) is crucial for fracture prevention.
• Machine learning (ML) offers potential for improved LBM risk prediction.

Purpose of the Study:

• To develop and validate an ML-based prediction model for LBM risk in postmenopausal women.
• To enhance the accuracy of identifying individuals requiring osteoporosis screening and preventive treatment.

Main Methods:

• Retrospective cross-sectional study involving 3,738 (internal) and 1,008 (external) menopausal women.
• Variable selection using LASSO and elastic net methods.
• Development and evaluation of ML algorithms (including XGBoost) and logistic regression for LBM risk prediction.

Main Results:

• The optimal prediction model achieved an AUC of 0.918 (internal) and 0.910 (external) for LBM prediction.
• The XGBoost model demonstrated notable predictive performance.
• The final model incorporated 11 key variables for discrimination.

Conclusions:

• An effective ML-based prediction model can identify postmenopausal women at high risk for LBM.
• This tool supports primary care providers in guiding decisions for osteoporosis prevention.
• Early identification facilitates timely interventions to prevent fractures.