PrOsteoporosis: predicting osteoporosis risk using NHANES data and machine learning approach

Area of Science:

• Gerontology
• Biomedical Informatics
• Public Health

Background:

• Osteoporosis is a common condition in the elderly, often diagnosed late due to bone mineral density (BMD) testing limitations.
• Early detection of osteoporosis is crucial for timely intervention and management to prevent fractures and improve quality of life.

Purpose of the Study:

• To develop and validate a machine learning (ML) approach for osteoporosis identification.
• To integrate demographic, laboratory, and questionnaire data for a more effective screening tool.

Main Methods:

• Utilized data from the National Health and Nutrition Examination Survey (NHANES) with 8766 participants and 223 variables.
• Employed feature selection techniques (Minimum Redundancy Maximum Relevance, SelectKBest) and applied four ML algorithms (RF, NN, LightGBM, XGBoost).
• Balanced data using SMOTE and evaluated model performance using F1 score and AUC.

Main Results:

• The LightGBM model achieved the highest performance with an F1 score of 0.912 and AUC of 0.972 on the test set.
• Key predictors identified for osteoporosis were height, age, and sex.
• The models demonstrated strong potential in differentiating individuals with and without osteoporosis.

Conclusions:

• Machine learning models, particularly LightGBM, show promise for early osteoporosis detection and personalized prevention.
• Height, age, and sex are significant predictors, offering valuable insights for clinical risk assessment.
• This ML approach provides a practical and effective alternative or supplement to traditional BMD testing for osteoporosis screening.