Machine learning model integrating oral microbiota and clinical features for predicting osteoporosis and bone loss in high-altitude populations

Osteoporosis and bone loss (OP&BL) are major public health challenges, especially in high-altitude environments with chronic hypoxia. Current diagnostic methods, based on low-altitude populations, are impractical for large-scale screening in resource-limited, high-altitude settings. This study developed a machine learning-based predictive model for OP&BL by integrating oral microbiota data with clinical and questionnaire variables. We analyzed data from 560 Tibetan adults residing at high altitudes. Bone health status (OP&BL vs. normal) was determined by dual-energy X-ray absorptiometry. Oral microbiota profiles were characterized via 16 S rRNA sequencing. After feature selection using elastic net regression, five machine learning models, namely Logistic Regression (LR), Naïve Bayes (NB), Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGB), were trained (60%, 337/560) and validated (40%, 223/560). Feature selection identified nine predictors: Age, Gender, BMI, oral microbial genera Abiotrophia, Frequency of spicy food consumption (H23), Tooth brushing frequency (J5), Frequency of sweet-drink consumption (J3b), Current marital status (Separated/Divorced, A5_3), and frequency of numbing food consumption (H27). The LR model demonstrated good and stable performance with an AUC of 0.885 (95% CI: 0.823–0.937) on the test set, along with good calibration and the highest net clinical benefit. SHAP analysis indicated that oral factors Abiotrophia and Tooth brushing frequency together accounted for nearly 10% of the model’s total predictive contribution. We developed a machine learning model integrating oral microbiota and clinical data for predicting OP&BL in people living above 3500 m. This model could offer a promising non-invasive tool for early screening in resource-limited settings and highlights the potential role of oral factors in high-altitude bone health. This study develops a machine learning model tailored for high-altitude populations, addressing the limitations of current diagnostic methods based on low-altitude data. By integrating oral microbiota data with clinical features, we propose a novel, non-invasive prediction model for osteoporosis and bone loss, specifically designed for high-altitude environments. We evaluated multiple machine learning algorithms, including Logistic Regression (LR), Naïve Bayes (NB), Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGB). The LR model demonstrated superior, generalizable performance with an AUC of 0.885 (95% CI: 0.823–0.937) on the test set, along with excellent calibration and clinical utility. Key predictors included the oral microbiota genus Abiotrophia and Tooth brushing frequency, highlighting the potential combined relevance of oral hygiene and microbiome-host interactions for bone health.