Estimation of residential fine particulate matter infiltration in Shanghai, China.

Ambient concentrations of fine particulate matter (PM2.5 ) concentration is often used as an exposure surrogate to estimate PM2.5 health effects in epidemiological studies. Ignoring the potential variations in the amount of outdoor PM2.5 infiltrating into indoor environments will cause exposure misclassification, especially when people spend most of their time indoors. As it is not feasible to measure the PM2.5 infiltration factor (Finf ) for each individual residence, we aimed to build models for residential PM2.5 Finf prediction and to evaluate seasonal Finf variations among residences. We repeated collected paired indoor and outdoor PM2.5 filter samples for 7 continuous days in each of the three seasons (hot, cold and transitional seasons) from 48 typical homes of Shanghai, China. PM2.5 -bound sulfur on the filters was measured by X-ray fluorescence for PM2.5 Finf calculation. We then used stepwise-multiple linear regression to construct season-specific models with climatic variables and questionnaire-based predictors. All models were evaluated by the coefficient of determination (R2 ) and root mean square error (RMSE) from a leave-one-out-cross-validation (LOOCV). The 7-day mean (±SD) of PM2.5 Finf across all observations was 0.83 (±0.18). Finf was found higher and more varied in transitional season (12-25 °C) than hot (>25 °C) and cold (<12 °C) seasons. Air conditioning use and meteorological factors were the most important predictors during hot and cold seasons; Floor of residence and building age were the best transitional season predictors. The models predicted 60.0%-68.4% of the variance in 7-day averages of Finf , The LOOCV analysis showed an R2 of 0.52 and an RMSE of 0.11. Our finding of large variation in residential PM2.5 Finf between seasons and across residences within season indicated the important source of outdoor-generated PM2.5 exposure heterogeneity in epidemiologic studies. Our models based on readily available data may potentially improve the accuracy of estimates of the health effects of PM2.5 exposure.