Machine learning identifies traffic-related pollutant mixtures as potential factors suggestive of association with osteoporosis in a cross-sectional analysis of NHANES

Clinical Snapshot

Key Findings

• P Value: P = 0.412

• Effect Size: Mixture odds ratio 1.21

• Primary Outcome: Osteoporosis prevalence 12.97%

• Nnt Or Sensitivity: LASSO model achieved highest AUC 0.81 with CV SD 0.0167

• Confidence Interval: 95% CI: 0.77-2.63

Clinical Application

Limited clinical utility given non-significant findings; may inform occupational health screening Relevance to Australian practice limited by different exposure patterns; may inform TGA environmental health guidelines and occupational health standards Adults with occupational or environmental exposure to traffic-related pollutants

Abstract

The combined effect of traffic-related pollutant mixtures on osteoporosis (OP) remains unclear. This study aimed to evaluate such associations using machine learning and mixture modeling approaches. A cross-sectional analysis was conducted among 3053 participants from the National Health and Nutrition Examination Survey 2015 to 2018. Eighteen exposures were assessed, including heavy metals, polycyclic aromatic hydrocarbons, phthalates, per- and polyfluoroalkyl substances, volatile organic compounds, diesel exhaust, and behavioral factors. OP was defined by femoral neck bone mineral density T-score ≤-2.5. Weighted quantile sum regression and 5 machine learning models were applied, accounting for complex survey design. OP prevalence was 12.97%. Weighted quantile sum regression showed a positive but nonsignificant mixture odds ratio (odds ratio = 1.21; 95% confidence interval: 0.77-2.63; P = .412). The largest positive weights were daily outdoor work duration (0.240) and occupational diesel exhaust (0.136). Random forest identified female sex, age, vitamin D, blood lead, and body mass index as top predictors. Least Absolute Shrinkage and Selection Operator achieved the highest test area under the receiver operating characteristic curve (0.81) with the lowest cross-validation standard deviation (0.0167). Its coefficients highlighted occupational diesel exhaust (β = 0.40) as a notable environmental factor alongside female sex (β = 0.95) and age (β = 0.70). Traffic-related pollutant mixtures showed a cross-sectional association with OP in a nationally representative US population, with occupational diesel exhaust and prolonged outdoor work emerging as key contributors in mixture models. However, the nonsignificant overall mixture effect and potential for reverse causation warrant caution in interpretation. These findings support considering environmental and occupational factors in OP prevention, although the overall mixture effect did not reach statistical significance.

References

1. 1.Wang, Y., Deng, M., Xu, Y., Zha, Z., Feng, Y., Zhuang, T., He, J., Peng, H., & Wu, W. (2026). Machine learning identifies traffic-related pollutant mixtures as potential factors suggestive of association with osteoporosis in a cross-sectional analysis of NHANES. Medicine, 105(22), e48884. https://doi.org/10.1097/MD.0000000000048884