An explainable machine learning model for predicting one-year osteoporosis risk: development and validation in a
Jinyi Wu1,2, Dan Yu3, Dan Huang4
1Department of Public Health, Wuhan Fourth Hospital, Wuhan, 430000, China.
BMC Musculoskeletal Disorders
|July 2, 2026
Summary
This study developed an interpretable machine learning model to predict osteoporosis risk using routine biochemical indicators. The model offers an accessible and efficient alternative for early osteoporosis screening.
Area of Science:
- Biomedical Informatics
- Machine Learning in Healthcare
- Osteoporosis Research
Background:
- Osteoporosis poses a significant public health challenge, necessitating efficient risk prediction tools.
- Current methods like FRAX and radiomics have limitations in accessibility and resource requirements.
- Developing accurate, interpretable, and clinically practical predictive models is crucial for early intervention.
Purpose of the Study:
- To develop and validate an interpretable machine learning model for predicting the 1-year risk of osteoporosis.
- To utilize readily available biochemical indicators as input features.
- To provide an accessible and efficient alternative to existing osteoporosis risk assessment tools.
Main Methods:
- A prospective cohort study utilizing electronic medical records from a Chinese hospital (Jan 2020-Jan 2025).
- Inclusion criteria: patients with at least two DXA scans and blood tests.
- Feature selection via LASSO and univariate regression, followed by XGBoost model development and validation using AUC, sensitivity, and specificity. SHAP analysis for interpretability.
Main Results:
- A final model using 17 routine clinical features was developed.
- The XGBoost model achieved high predictive performance: AUC of 0.966 (derivation set) and 0.969 (external validation set) for spine osteoporosis.
- The model demonstrated robust predictive power for hip osteoporosis and provided interpretable insights via SHAP analysis.
Conclusions:
- A highly accurate, interpretable, and clinically practical machine learning model for 1-year osteoporosis prediction was successfully developed.
- The model, using only 17 routine features, surpasses many previous models and rivals radiomics-based approaches in performance while being more accessible.
- An online tool facilitates clinical application, offering an efficient strategy for early osteoporosis screening and risk assessment.
Related Concept Videos
01:29Bone Disorders
Aging and its effect on bone remodeling is the most common cause of bone disorders. In young and healthy people, bone deposition and resorption happen at an equal rate to maintain optimal bone health.
Bone deposition is also affected by the levels of sex hormones like estrogen and testosterone that promote osteoblast activity and bone matrix synthesis. When the level of these hormones decreases due to aging, it causes a reduction in bone deposition. As a result, bone resorption by osteoclasts...
Bone deposition is also affected by the levels of sex hormones like estrogen and testosterone that promote osteoblast activity and bone matrix synthesis. When the level of these hormones decreases due to aging, it causes a reduction in bone deposition. As a result, bone resorption by osteoclasts...
01:31Osteoclasts in Bone Remodeling
Osteoclasts are cells responsible for bone resorption and remodeling. They originate from hematopoietic progenitor cells present in the bone marrow. Numerous progenitor cells fuse to form multinucleated cells, each with 10-20 nuclei. A single osteoclast has a diameter of 150 to 200 µM. These cells have ruffled borders that break down the underlying bone tissue and release minerals such as calcium into the blood in bone resorption. Osteoclasts cling to bones with their ruffled edges during bone...
01:20Study Designs in Epidemiology
Epidemiological study designs are fundamental tools for investigating the distribution, determinants, and control of health conditions in populations. They help researchers understand the relationships between exposures and outcomes, and they broadly fall into two categories: "observational" and "experimental" studies.
Observational studies are those where the researcher does not intervene but rather observes natural variations. They include cross-sectional, cohort, and case-control studies.
Observational studies are those where the researcher does not intervene but rather observes natural variations. They include cross-sectional, cohort, and case-control studies.