Association of inflammation and lung function decline caused by personal PM_2.5 exposure: a machine learning approach in time-series data

Numerous studies focused on the association between lung function impairment and inflammation caused by fine particulate matter (PM_2.5), but the causal relationships are difficult to clarify. In the current study, twenty healthy Chinese young adults who participated in 7 days of observation every four seasons were enrolled, and autoregression models (AM) and classification and regression trees (CART) in a machine learning framework were applied to analyze the association among PM_2.5 exposure, inflammation, and lung function from a data structure perspective. There were strong cross-correlations between personal dose of PM_2.5 (Dw) and lung functions (vital capacity (VC), forced vital capacity (FVC), etc.). These cross-correlation coefficients were associated with inflammatory indicators (uteroglobin (UG), serum amyloid (SAA), and fractional exhaled nitric oxide (FeNO)). CART reported that inflammatory indicators UG and SAA had the predictive ability of the directional association between Dw and FVC at 1-day lag and that high levels of UG and SAA predicted that PM_2.5 exposure induced lung function decline. Consistently, lower lung function indicators at a 2-day lag after personal PM_2.5 exposure predicted the high value of inflammatory indicator FeNO. Taken together, we applied machine learning algorithms to analyze repeated measurement data, finding that inflammation and lung function decline caused by PM_2.5 could affect each other.