A machine-learning-based osteoporosis screening tool integrating the Shapley Additive exPlanation (SHAP) method:...

Area of Science:

• Gerontology and Public Health
• Biomedical Informatics
• Medical Diagnostics

Background:

• Current osteoporosis screening tools lack accuracy and convenience.
• There is a critical need for improved osteoporosis detection methods.
• Existing tools function as the initial defense against osteoporosis but are suboptimal.

Purpose of the Study:

• To develop a machine-learning-based tool for osteoporosis screening.
• To create a clinically applicable and interpretable osteoporosis screening solution.
• To enhance the accuracy and accessibility of osteoporosis detection.

Main Methods:

• Utilized data from 9405 American participants aged 50 and older.
• Selected 13 accessible indicators, employed Lasso for feature selection, and compared six models.
• Employed SHAP for interpretability and conducted external validation on an independent dataset.

Main Results:

• Gradient Boosting Machine (GB) demonstrated superior performance, with key factors including weight, age, and height.
• External validation showed the GB tool achieved an AUC of 0.828, outperforming the Osteoporosis Self-Assessment Tool (OST) AUC of 0.781.
• The machine learning model exhibited high sensitivity, specificity, and F1-score, indicating robust predictive capability.

Conclusions:

• A validated, clinically applicable osteoporosis screening tool leveraging machine learning was successfully developed.
• The new tool offers improved accuracy and accessibility for osteoporosis screening.
• This advancement represents a significant step forward in early osteoporosis detection.