Risk Prediction of Low Bone Density in Elderly Patients with Supervised Machine Learning Algorithms

Area of Science:

• Gerontology and Public Health
• Biostatistics and Machine Learning
• Medical Informatics

Background:

• Low bone mineral density (BMD) is a significant age-related condition associated with increased fracture risk and mortality.
• Machine learning (ML) offers a powerful tool for early prediction of low BMD by analyzing clinical, biochemical, and demographic data.

Purpose of the Study:

• To assess the predictive capabilities of eleven ML models for identifying low BMD.
• To pinpoint the most significant risk factors contributing to low BMD using the top-performing ML model.

Main Methods:

• A cross-sectional study utilizing data from the National Health and Nutrition Examination Survey (2005-2020).
• 12,108 participants aged ≥ 50 years with BMD data were analyzed using 57 features.
• Supervised ML algorithms were trained and evaluated using metrics like accuracy and AUC, with SHAP analysis for predictor importance.

Main Results:

• The Extra Trees classifier demonstrated superior performance with 76.7% accuracy and an AUC of 0.85.
• Fourteen key predictors of low BMD were identified, including sex, age, BMI, race, serum uric acid, and diabetes status.
• Established and novel risk factors were highlighted through feature importance analysis.

Conclusions:

• Tree-based ML models, especially Extra Trees, are effective for predicting low BMD.
• The study identified crucial risk factors, supporting personalized screening for osteoporosis and osteopenia.
• ML effectively captures complex multifactorial relationships in population health data for BMD prediction.