Historical Prediction Modeling Approach for Estimating Long-Term Concentrations of PM_2.5in Cohort Studies before the 1999 Implementation of Widespread Monitoring

Abstract

Introduction:Recent cohort studies have used exposure prediction models to estimate the association between long-term residential concentrations of fine particulate matter (PM2.5) and health. Because these prediction models rely on PM2.5 monitoring data, predictions for times before extensive spatial monitoring present a challenge to understanding long-term exposure effects. The U.S. Environmental Protection Agency (EPA) Federal Reference Method (FRM) network for PM2.5 was established in 1999.Objectives:We evaluated a novel statistical approach to produce high-quality exposure predictions from 1980 through 2010 in the continental United States for epidemiological applications.Methods:We developed spatio-temporal prediction models using geographic predictors and annual average PM2.5 data from 1999 through 2010 from the FRM and the Interagency Monitoring of Protected Visual Environments (IMPROVE) networks. Temporal trends before 1999 were estimated by using a) extrapolation based on PM2.5 data in FRM/IMPROVE, b) PM2.5 sulfate data in the Clean Air Status and Trends Network, and c) visibility data across the Weather Bureau Army Navy network. We validated the models using PM2.5 data collected before 1999 from IMPROVE, California Air Resources Board dichotomous sampler monitoring (CARB dichot), the Children’s Health Study (CHS), and the Inhalable Particulate Network (IPN).Results:In our validation using pre-1999 data, the prediction model performed well across three trend estimation approaches when validated using IMPROVE and CHS data (R2 = 0.84–0.91) with lower R2 values in early years. Model performance using CARB dichot and IPN data was worse (R2 = 0.00–0.85) most likely because of fewer monitoring sites and inconsistent sampling methods.Conclusions:Our prediction modeling approach will allow health effects estimation associated with long-term exposures to PM2.5 over extended time periods ≤ 30 years.Citation:Kim SY, Olives C, Sheppard L, Sampson PD, Larson TV, Keller JP, Kaufman JD. 2017. Historical prediction modeling approach for estimating long-term concentrations of PM2.5 in cohort studies before the 1999 implementation of widespread monitoring. Environ Health Perspect 125:38–46; http://dx.doi.org/10.1289/EHP131

Air pollution and subclinical interstitial lung disease: the Multi-Ethnic Study of Atherosclerosis (MESA) air–lung study

“…As a sensitivity analysis, we estimated twenty and thirty year exposure to PM 2.5 prior to full lung CT scan based on historic models of annual average exposure developed for periods pre-dating intensive PM monitoring. 22 These long-term residential estimates used address-weighted averages of annual PM 2.5 concentration.…”

Concentrations of criteria pollutants in the contiguous U.S., 1979 – 2015: Role of prediction model parsimony in integrated empirical geographic regression

“…The mean component is equivalent to the linear regression model often referred to as land use regression (LUR) with PLS data-reduction. Whereas other dimension reduction approaches such as principal component analysis solely rely on correlation of covariates, PLS predictors are estimated based on the correlation between covariates and the outcome; PLS was adopted in several previous prediction models [2,3,18,25,27,35]. The summary predictors not only incorporate various geographic characteristics, they also avoid producing extreme predictions [37].…”

“…Empirical models are a cost-effective approach to estimate fine-scale exposures to air pollution. Recent population-level studies of air pollution have relied on empirical models to estimate long-term concentrations of outdoor air pollution based largely on observation-driven geostatistical approaches [1][2][3] or hybrid approaches that incorporate satellite-based observations of air quality and theory-based mechanistic models with geostatistical approaches [4][5][6][7]. These model predictions are used to assess population-level characteristics of air pollution, such as health effects [8][9][10][11], the burden of disease [12,13], and exposure disparities [14,15].…”

A Deep Belief Network Combined with Modified Grey Wolf Optimization Algorithm for PM2.5 Concentration Prediction

“…The mean absolute error (MAE), the mean square error (MSE) and the R 2 are popular indicators for evaluating the prediction results [22][23][24][25]. Thus, we use the three indicators to evaluate the proposed prediction model.…”