Data-driven identification of metabolic and cardiovascular biomarkers in high-altitude workers: a machine learning approach.

Background: Workers in high-altitude mining settings face increased cardiometabolic risk due to chronic exposure to low oxygen levels. Traditional fitness-for-work (FFW) assessments often evaluate biomarkers in isolation, missing relevant health patterns. Aim: To improve the risk stratification of the FFW status in high-altitude workers by identifying relevant biomarkers through ML models. Methods: A retrospective cohort of 420,966 preemployment examination records, corresponding to 89,149 workers between 2021 and 2024 was analyzed. Workers were classified as fit or unfit for work, in each of their medical examinations, according to national guidelines. Several supervised ML models were applied, including random forests (RF), support vector machines, k-nearest neighbors, and decision trees, to identify relevant predictors of FFW. Logistic regression was performed to assess statistical associations between biomarkers and fitness outcomes. Results: Among the 420,966 preemployment examination records, 48,783 were particularly assessed for fitness for high-altitude work. Among these, 8% were classified as unfit for high-altitude work. Significant predictors included body mass index (BMI), blood glucose, triglycerides, and systolic blood pressure. The Random Forest (RF) model outperformed SVM and KNN, achieving the highest predictive performance with an accuracy of 0.89, sensitivity of 0.92, and specificity of 0.83. Multivariate logistic regression confirmed BMI as the strongest predictor (OR 2.640, Conclusion: ML models can effectively identify critical health indicators related to FFW in high-altitude environments. These tools offer the potential to improve occupational health assessments and support preventive decision making in vulnerable worker populations.