Here, we investigated the use of breath-borne volatile organic compounds (VOCs) for rapid monitoring of air pollution health effects on humans. Forty-seven healthy college students were recruited, and their exhaled breath samples (n = 235) were collected and analyzed for VOCs before, on, and after two separate haze pollution episodes using gas chromatography-ion mobility spectrometry (GC-IMS). Using a paired t-test and machine learning model (Gradient Boosting Machine, GBM), six exhaled VOC species including propanol and isoprene were revealed to differ significantly among pre-, on-, and post-exposure in both haze episodes, while none was found between clean control days. The GBM model was shown capable of differentiating between pre- and on-exposure to haze pollution with a precision of 90-100% for both haze episodes. However, poor performance was detected for the same model between two different clean days. In addition to gender and particular haze occurrence influences, correlation analysis revealed that NH4+, NO3-, acetic acid, mesylate, CO, NO2, PM2.5, and O3 played important roles in the changes in breath-borne VOC fingerprints following haze air pollution exposure. This work has demonstrated direct evidence of human health impacts of haze pollution while identifying potential breath-borne VOC biomarkers such as propanol and isoprene for haze air pollution exposure.
Keywords: VOC; exhaled breath; haze air pollution; health effects; machine learning.