Assessing the Improvement in Predicting Personal Exposure to Elements in PM2.5 by Including Indoor PM2.5 Measurements and Home Characteristics to Outdoor PM2.5 Measurements

摘要

The elemental composition of PM2.5 has been associated with adverse health outcomes. Most epidemiological studies use outdoor PM2.5 concentrations at central monitoring sites as a surrogate of personal exposure. Here, it was tested if the addition of indoor PM2.5 measurements or other home and personal characteristics increased the prediction accuracy of personal PM2.5. The mass concentrations of PM2.5 and 36 elements were measured during 48-hour parallel indoor, outdoor, and personal sampling in nonsmoking households in three urban areas as part of the Relationship of Indoor, Outdoor, and Personal Air (RIOPA) study. Random forests were used to predict personal exposure for elements in PM2.5. A total of three models were developed, differing only with respect to the predictors used. Model 1 included only outdoor elemental measurements. Model 2 included both outdoor and indoor concentrations of elements in PM2.5. Model 3 included indoor and outdoor PM2.5 measurements and also home characteristics. The mean absolute prediction error percentage (MAPE%) for all models were calculated based on bootstrapped cross-validation and then were quantitatively compared. The inclusion of indoor PM2.5 measurements significantly improved the prediction of personal exposure for 17 of the 24 elements (Al, As, Ba, Ca, Cl, Fe, K, Mn, S, Sb, Se, Sn, Sr, Ti, V, Zn, Zr) compared to the model with outdoor PM2.5 data alone. For the remaining elements (Br, Cr, Cu, Ni, Pb, Si), the inclusion of indoor PM2.5 elemental concentrations did not significantly improve the prediction of personal exposure. Inclusion of home characteristics did not significantly improve the prediction of personal exposure for any of the elements. Overall, using outdoor PM2.5 data is not a perfect surrogate for personal PM2.5 exposure. For most elements, supplementing outdoor PM2.5 data with indoor PM2.5 data decreases the MAPE% for personal PM2.5 exposure while the addition of home characteristics does not.