Construction and validation of a multi-dimensional health indicator-driven osteoporosis risk prediction model: a
Zhifeng Cai1,2,3, Xun Lu4, Hua Ni5
1Central Portion, Zhangjiagang Center for Disease Control and Prevention, Zhangjiagang, Jiangsu, 215600, China. cccczfeng1999@163.com.
Background:
Rising osteoporosis prevalence among elderly populations and limitations of current single-factor screening methods necessitate development of comprehensive multi-dimensional risk prediction models.
Methods:
A two-center cross-sectional study was conducted, enrolling 15,307 elderly participants in 2023. Among them, 11,957 participants from Zhangjiagang Center for Disease Control and Prevention were randomly divided into training set (8369 cases) and internal validation set (3588 cases) at a 7:3 ratio, while 3,350 participants from Zhangjiagang fifth People's Hospital served as the external validation set. Multi-dimensional health data including demographic information, physiological indicators, lifestyle factors, nutritional supplementation, and laboratory examinations were collected. Osteoporosis diagnosis was performed using ultrasound bone density testing (UBD T-score ≤-2.5). Univariate and multivariate logistic regression analyses were used to analyze risk factors, construct prediction models, and create nomograms. The predictive performance of five algorithms including logistic regression, random forest, extreme gradient boosting, support vector machine, and naive Bayes was compared. Model efficacy was evaluated using ROC curves, calibration curves, and decision curve analysis.
Results:
The prevalence of osteoporosis was 25.3% (3040/11957). Multivariate logistic regression analysis identified age and heart rate as independent risk factors, while BMI, education level, occupation type, exercise habits, daily milk consumption, hemoglobin, and triglycerides were protective factors. The logistic regression model demonstrated optimal and stable performance, with AUCs of 0.687 (95% CI: 0.674-0.700), 0.675 (95% CI: 0.655-0.696), and 0.679 (95% CI: 0.657-0.701) for the training set, internal validation set, and external validation set, respectively. Variable importance analysis showed that occupation type, daily exercise time, and hemoglobin demonstrated high and stable importance across all datasets.
Conclusion:
The osteoporosis risk prediction model based on multi-dimensional health indicators demonstrates good discriminative performance, calibration, and clinical utility, providing an effective tool for early screening and precision prevention of osteoporosis in elderly populations.
