Machine learning model for osteoporosis diagnosis based on bone turnover markers.

Area of Science:

• Biomedical diagnostics
• Computational biology
• Gerontology

Background:

• Osteoporosis diagnosis relies on bone mineral density, but bone turnover markers (BTMs) and demographic data offer complementary insights.
• Early identification of osteoporosis is crucial for timely intervention and fracture prevention.

Purpose of the Study:

• To evaluate the diagnostic performance of BTMs and demographic variables in identifying osteoporosis using machine learning.
• To compare the efficacy of various machine learning models in osteoporosis diagnosis.

Main Methods:

• A cross-sectional study included 280 participants (88 with osteoporosis, 192 controls).
• Serum BTMs and demographic data (age, sex) were collected.
• Six machine learning models (XGBoost, LGBM, CatBoost, random forest, SVM, KNN) were trained and evaluated using AUROC, F1-score, and accuracy.

Main Results:

• Light Gradient Boosting Machine (LGBM) achieved the highest Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.706 after optimization.
• LGBM's F1-score improved from 0.50 to 0.65 post-optimization.
• A combined model of LGBM, XGBoost, and CatBoost yielded an AUROC of 0.706, an F1-score of 0.65, and an accuracy of 0.73.

Conclusions:

• BTMs, age, and sex are significant predictors for diagnosing existing osteoporosis.
• Machine learning models utilizing these accessible clinical data show potential for effective osteoporosis assessment.
• This study supports the use of BTMs and demographic data as a valuable tool for early osteoporosis diagnosis and management.