Osteoporosis Pre-Screening Using Ensemble Machine Learning in Postmenopausal Korean Women

Area of Science:

• Gerontology and Geriatric Medicine
• Biomedical Informatics and Data Science
• Public Health and Epidemiology

Background:

• Osteoporosis is a degenerative disease closely linked to postmenopausal aging, making early diagnosis critical for effective management.
• Existing diagnostic methods may not capture the full spectrum of risk factors contributing to osteoporosis in postmenopausal women.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting osteoporosis risk in postmenopausal women.
• To identify key features influencing osteoporosis risk using feature selection techniques.
• To compare the performance of different machine learning algorithms (Random Forest, AdaBoost, Gradient Boosting) in predicting osteoporosis.

Main Methods:

• Utilized data from 1431 postmenopausal women (aged 40-69) from the Korea National Health and Nutrition Examination Surveys.
• Selected 20 relevant features affecting osteoporosis using feature importance and recursive feature elimination.
• Trained three models (A, B, C) using Random Forest, AdaBoost, and Gradient Boosting algorithms, incorporating checkup features, survey features, or both.

Main Results:

• Model C, combining both checkup and survey features, demonstrated superior performance across all algorithms.
• AdaBoost achieved the highest accuracy (0.849) and Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.921.
• The ensemble models, particularly AdaBoost, outperformed recent studies with an AUROC 0.1-0.2 higher.

Conclusions:

• Machine learning, especially ensemble methods like AdaBoost, shows significant potential for the early diagnosis of osteoporosis in postmenopausal women.
• The developed model, integrating comprehensive health data, offers a promising practical medical tool for risk assessment.
• Further refinement of the model can enhance its utility in clinical settings for proactive osteoporosis management.