Application of machine learning algorithms to identify people with low bone density.

Area of Science:

• Biomedical Informatics
• Gerontology
• Public Health

Background:

• Osteoporosis prevalence is increasing globally, posing significant health and economic challenges.
• Early detection of osteoporosis is difficult due to its subtle onset and the infeasibility of widespread screening.
• There is a critical need for effective methods to identify individuals at high risk for osteoporosis.

Purpose of the Study:

• To develop and validate a machine learning algorithm for identifying low bone density.
• To utilize readily available demographic and blood biochemical data for risk prediction.
• To improve early detection and management strategies for osteoporosis.

Main Methods:

• Utilized NHANES 2017-2020 data for participants over 50 years old with complete femoral neck bone mineral density (BMD) data.
• Developed six machine learning models (LR, SVM, GBM, NB, ANN, RF) using Lasso regression for variable selection.
• Validated model generalizability using NHANES 2013-2014 data and assessed performance using AUC, accuracy, and calibration curves.

Main Results:

• The logistic regression (LR) model demonstrated the best discrimination (AUC 0.785) and calibration in the test set.
• Key predictors identified by the LR model included age, BMI, gender, creatine phosphokinase, total cholesterol, and alkaline phosphatase.
• The LR model showed good predictive power in the external validation dataset and superior clinical utility.

Conclusions:

• Machine learning models, particularly logistic regression, can effectively classify low bone density using accessible biomarkers.
• This approach can significantly aid clinical decision-making for osteoporosis prevention and management.
• The study highlights the potential of leveraging routinely collected data for proactive bone health assessment.