Exploring the potential associations between single and mixed volatile compounds and preserved ratio impaired spirometry using five different approaches.

BACKGROUND

Although the relationship between environmental pollutants and respiratory health has received widespread attention, no studies have explored the association between volatile organic compounds (VOCs) and preserved ratio impaired spirometry (PRISm). The Systemic Inflammation Index (SII) is widely recognized as a reliable surrogate marker of systemic inflammatory status and has been identified as a mediating factor correlating various environmental pollutants to respiratory diseases.

OBJECTIVE

This study aims to investigate the potential associations between individual and combined VOCs and PRISm, and further explore the potential mediating role of SII.

METHODS

This study analyzed a subset of data from NHANES collected between 2007 and 2012. Multivariable logistic regression was employed to examine the association between individual VOCs and PRISm. Additionally, Weighted Quantile Sum (WQS) regression, the quantile g-computation (qgcomp) model, and Bayesian Kernel Machine Regression (BKMR) were used to assess the relationships between mixed VOCs and PRISm. We trained ten machine learning models to identify PRISm and assessed the relative importance of each feature using Shapley Additive Explanations (SHAP).

RESULTS

A total of 2616 participants were included in this study. In the fully adjusted model, multivariable logistic regression results indicated that each 1 ng/mL increase in benzene was associated with a 26 % increase in the prevalence of PRISm. The WQS regression, qgcomp model, and BKMR all suggested that a mixture of VOCs was associated with PRISm. Exploratory pathway analysis indicated that inflammation accounted for 7.69 % of the observed effect (Indirect Effect [IE]: Coefficient = 0.002, 95 % CI: 0.001-0.003, p-value = 0.006; Direct Effect [DE]: Coefficient = 0.024, 95 % CI: 0.017-0.031, p-value < 0.001). Additionally, the LightGBM model exhibited the highest predictive performance, with an AUC of 0.839. SHAP analysis identified race, body mass index (BMI), and 1,4-dichlorobenzene as the three most important factors.

CONCLUSION

Our findings demonstrate the correlations between VOCs and PRISm, with inflammation potentially mediating the relationship. The machine learning results highlight the potential of combining VOCs with demographic characteristics to improve PRISm identification, thereby supporting the development of prevention and intervention strategies.