Escape to America: Adapting European Study for Cohorts for Air Pollution Effects (ESCAPE) Methods to the Desert Southwestern U.S.

摘要

The influence of air pollution on respiratory illnesses is well-documented, however few long-running respiratory studies have measured air pollution at participant homes or other locations. To retrospectively predict air pollution exposures, various approaches have been taken, including dispersion modeling, geostatistical interpolation, and land use regression (LUR) modeling. LUR modeling as per the European Study for Cohorts for Air Pollution Effects (ESCAPE) has been used in >30 locations in Europe, but never in the US nor validated with values measured >10 years prior. In this study, we tested the ability of LUR models to predict air pollutant measures from nearly 30 years earlier in Tucson, AZ. Using ESCAPE sampling methods, the Tucson Air Pollution Study (TAPS) sampled NO2 in 40 homes and PM2.5 and PM10 in half of those homes for 3 two-week periods from 2015-2016 in Tucson, AZ. Using comparable sampling methods, the Pima County Workers Study (PCWS) measured N02 in 39 homes and PM2.5 and PM10 in 17 homes for 2 one-week periods in 1987 in Tucson, AZ. LUR models were developed for each pollutant in each dataset and internally validated with ESCAPE methods. Then, TAPS-based LUR models predicted PCWS pollutant levels. During internal validation, TAPS-based LUR models performed better than PCWS-based models: for TAPS, adjusted R2 values for NO2, PM2.5, and PM10 were 0.75, 0.55, and 0.68, respectively, versus, 0.47, 0.25, and 0.21 for PCWS. In retrospective prediction, adjusted R2 values (root-mean-square error) for TAPS-based LUR models for NO2, PM2.5, and PM10 were 0.37 (10.7 ppb), 0.70 (27.6 pg/m3), and 0.30 (17.3 pg/m3), respectively. While predictions for PM2.5 and PM10 levels had limited success, the N02 predictions were within the measured range but generally under-predicted (predicted vs. measured ranges: 2.83 - 8.41 vs. 5.17 - 24.8 ppb). Our project shows promise for using LUR to retrospectively model air pollution levels measured nearly 30 years earlier.