Identifying groups of children's social mobility opportunity for public health applications using k-means clustering.

Background: The Opportunity Atlas project is a pioneering effort to trace social mobility and adulthood socioeconomic outcomes back to childhood residence. Half of the variation in adulthood socioeconomic outcomes was explainable by neighborhood-level socioeconomic characteristics during childhood. Clustering census tracts by Opportunity Atlas characteristics would allow for further exploration of variance in social mobility. Our objectives here are to identify and describe spatial clustering trends within Opportunity Atlas outcomes. Methods: We utilized a k-means clustering machine learning approach with four outcome variables (individual income, incarceration rate, employment, and percent of residents living in a neighborhood with low levels of poverty) each given at five parental income levels (1st, 25th, 50th, 75th, and 100th percentiles of the national distribution) to create clusters of census tracts across the contiguous United States (US) and within each Environmental Protection Agency region. Results: At the national level, the algorithm identified seven distinct clusters; the highest opportunity clusters occurred in the Northern Midwest and Northeast, and the lowest opportunity clusters occurred in rural areas of the Southwest and Southeast. For regional analyses, we identified between five to nine clusters within each region. PCA loadings fluctuate across parental income levels; income and low poverty neighborhood residence explain a substantial amount of variance across all variables, but there are differences in contributions across parental income levels for many components. Conclusions: Using data from the Opportunity Atlas, we have taken four social mobility opportunity outcome variables each stratified at five parental income levels and created nationwide and EPA region-specific clusters that group together census tracts with similar opportunity profiles. The development of clusters that can serve as a combined index of social mobility opportunity is an important contribution of this work, and this in turn can be employed in future investigations of factors associated with children's social mobility.

Human exposure factors as potential determinants of the heterogeneity in city-specific associations between PM2.5 and mortality.

Multi-city population-based epidemiological studies of short-term fine particulate matter (PM2.5) exposures and mortality have observed heterogeneity in risk estimates between cities. Factors affecting exposures, such as pollutant infiltration, which are not captured by central-site monitoring data, can differ between communities potentially explaining some of this heterogeneity. This analysis evaluates exposure factors as potential determinants of the heterogeneity in 312 core-based statistical areas (CBSA)-specific associations between PM2.5 and mortality using inverse variance weighted linear regression. Exposure factor variables were created based on data on housing characteristics, commuting patterns, heating fuel usage, and climatic factors from national surveys. When survey data were not available, air conditioning (AC) prevalence was predicted utilizing machine learning techniques. Across all CBSAs, there was a 0.95% (Interquartile range (IQR) of 2.25) increase in non-accidental mortality per 10 µg/m3 increase in PM2.5 and significant heterogeneity between CBSAs. CBSAs with larger homes, more heating degree days, a higher percentage of home heating with oil had significantly (p < 0.05) higher health effect estimates, while cities with more gas heating had significantly lower health effect estimates. While univariate models did not explain much of heterogeneity in health effect estimates (R2 < 1%), multivariate models began to explain some of the observed heterogeneity (R2 = 13%).

Influence of exposure differences on city-to-city heterogeneity in PM2.5-mortality associations in US cities.

BACKGROUND: Multi-city population-based epidemiological studies have observed heterogeneity between city-specific fine particulate matter (PM2.5)-mortality effect estimates. These studies typically use ambient monitoring data as a surrogate for exposure leading to potential exposure misclassification. The level of exposure misclassification can differ by city affecting the observed health effect estimate. METHODS: The objective of this analysis is to evaluate whether previously developed residential infiltration-based city clusters can explain city-to-city heterogeneity in PM2.5 mortality risk estimates. In a prior paper 94 cities were clustered based on residential infiltration factors (e.g. home age/size, prevalence of air conditioning (AC)), resulting in 5 clusters. For this analysis, the association between PM2.5 and all-cause mortality was first determined in 77 cities across the United States for 2001-2005. Next, a second stage analysis was conducted evaluating the influence of cluster assignment on heterogeneity in the risk estimates. RESULTS: Associations between a 2-day (lag 0-1 days) moving average of PM2.5 concentrations and non-accidental mortality were determined for each city. Estimated effects ranged from -3.2 to 5.1% with a pooled estimate of 0.33% (95% CI: 0.13, 0.53) increase in mortality per 10 µg/m3 increase in PM2.5. The second stage analysis determined that cluster assignment was marginally significant in explaining the city-to-city heterogeneity. The health effects estimates in cities with older, smaller homes with less AC (Cluster 1) and cities with newer, smaller homes with a large prevalence of AC (Cluster 3) were significantly lower than the cluster consisting of cities with older, larger homes with a small percentage of AC. CONCLUSIONS: This is the first study that attempted to examine whether multiple exposure factors could explain the heterogeneity in PM2.5-mortality associations. The results of this study were found to explain a small portion (6%) of this heterogeneity.

Probabilistic estimation of residential air exchange rates for population-based human exposure modeling.

Residential air exchange rates (AERs) are a key determinant in the infiltration of ambient air pollution indoors. Population-based human exposure models using probabilistic approaches to estimate personal exposure to air pollutants have relied on input distributions from AER measurements. An algorithm for probabilistically estimating AER was developed based on the Lawrence Berkley National Laboratory Infiltration model utilizing housing characteristics and meteorological data with adjustment for window opening behavior. The algorithm was evaluated by comparing modeled and measured AERs in four US cities (Los Angeles, CA; Detroit, MI; Elizabeth, NJ; and Houston, TX) inputting study-specific data. The impact on the modeled AER of using publically available housing data representative of the region for each city was also assessed. Finally, modeled AER based on region-specific inputs was compared with those estimated using literature-based distributions. While modeled AERs were similar in magnitude to the measured AER they were consistently lower for all cities except Houston. AERs estimated using region-specific inputs were lower than those using study-specific inputs due to differences in window opening probabilities. The algorithm produced more spatially and temporally variable AERs compared with literature-based distributions reflecting within- and between-city differences, helping reduce error in estimates of air pollutant exposure.

A simulation study to quantify the impacts of exposure measurement error on air pollution health risk estimates in copollutant time-series models.

BACKGROUND: Exposure measurement error in copollutant epidemiologic models has the potential to introduce bias in relative risk (RR) estimates. A simulation study was conducted using empirical data to quantify the impact of correlated measurement errors in time-series analyses of air pollution and health. METHODS: ZIP-code level estimates of exposure for six pollutants (CO, NOx, EC, PM2.5, SO4, O3) from 1999 to 2002 in the Atlanta metropolitan area were used to calculate spatial, population (i.e. ambient versus personal), and total exposure measurement error. Empirically determined covariance of pollutant concentration pairs and the associated measurement errors were used to simulate true exposure (exposure without error) from observed exposure. Daily emergency department visits for respiratory diseases were simulated using a Poisson time-series model with a main pollutant RR = 1.05 per interquartile range, and a null association for the copollutant (RR = 1). Monte Carlo experiments were used to evaluate the impacts of correlated exposure errors of different copollutant pairs. RESULTS: Substantial attenuation of RRs due to exposure error was evident in nearly all copollutant pairs studied, ranging from 10 to 40% attenuation for spatial error, 3-85% for population error, and 31-85% for total error. When CO, NOx or EC is the main pollutant, we demonstrated the possibility of false positives, specifically identifying significant, positive associations for copollutants based on the estimated type I error rate. CONCLUSIONS: The impact of exposure error must be considered when interpreting results of copollutant epidemiologic models, due to the possibility of attenuation of main pollutant RRs and the increased probability of false positives when measurement error is present.

Comparing multipollutant emissions-based mobile source indicators to other single pollutant and multipollutant indicators in different urban areas.

A variety of single pollutant and multipollutant metrics can be used to represent exposure to traffic pollutant mixtures and evaluate their health effects. Integrated mobile source indicators (IMSIs) that combine air quality concentration and emissions data have recently been developed and evaluated using data from Atlanta, Georgia. IMSIs were found to track trends in traffic-related pollutants and have similar or stronger associations with health outcomes. In the current work, we apply IMSIs for gasoline, diesel and total (gasoline + diesel) vehicles to two other cities (Denver, Colorado and Houston, Texas) with different emissions profiles as well as to a different dataset from Atlanta. We compare spatial and temporal variability of IMSIs to single-pollutant indicators (carbon monoxide (CO), nitrogen oxides (NOx) and elemental carbon (EC)) and multipollutant source apportionment factors produced by Positive Matrix Factorization (PMF). Across cities, PMF-derived and IMSI gasoline metrics were most strongly correlated with CO (r = 0.31-0.98), while multipollutant diesel metrics were most strongly correlated with EC (r = 0.80-0.98). NOx correlations with PMF factors varied across cities (r = 0.29-0.67), while correlations with IMSIs were relatively consistent (r = 0.61-0.94). In general, single-pollutant metrics were more correlated with IMSIs (r = 0.58-0.98) than with PMF-derived factors (r = 0.07-0.99). A spatial analysis indicated that IMSIs were more strongly correlated (r > 0.7) between two sites in each city than single pollutant and PMF factors. These findings provide confidence that IMSIs provide a transferable, simple approach to estimate mobile source air pollution in cities with differing topography and source profiles using readily available data.

An empirical assessment of exposure measurement error and effect attenuation in bipollutant epidemiologic models.

BACKGROUND: Using multipollutant models to understand combined health effects of exposure to multiple pollutants is becoming more common. However, complex relationships between pollutants and differing degrees of exposure error across pollutants can make health effect estimates from multipollutant models difficult to interpret. OBJECTIVES: We aimed to quantify relationships between multiple pollutants and their associated exposure errors across metrics of exposure and to use empirical values to evaluate potential attenuation of coefficients in epidemiologic models. METHODS: We used three daily exposure metrics (central-site measurements, air quality model estimates, and population exposure model estimates) for 193 ZIP codes in the Atlanta, Georgia, metropolitan area from 1999 through 2002 for PM2.5 and its components (EC and SO4), as well as O3, CO, and NOx, to construct three types of exposure error: δspatial (comparing air quality model estimates to central-site measurements), δpopulation (comparing population exposure model estimates to air quality model estimates), and δtotal (comparing population exposure model estimates to central-site measurements). We compared exposure metrics and exposure errors within and across pollutants and derived attenuation factors (ratio of observed to true coefficient for pollutant of interest) for single- and bipollutant model coefficients. RESULTS: Pollutant concentrations and their exposure errors were moderately to highly correlated (typically, > 0.5), especially for CO, NOx, and EC (i.e., "local" pollutants); correlations differed across exposure metrics and types of exposure error. Spatial variability was evident, with variance of exposure error for local pollutants ranging from 0.25 to 0.83 for δspatial and δtotal. The attenuation of model coefficients in single- and bipollutant epidemiologic models relative to the true value differed across types of exposure error, pollutants, and space. CONCLUSIONS: Under a classical exposure-error framework, attenuation may be substantial for local pollutants as a result of δspatial and δtotal with true coefficients reduced by a factor typically < 0.6 (results varied for δpopulation and regional pollutants).

Clustering cities with similar fine particulate matter exposure characteristics based on residential infiltration and in-vehicle commuting factors.

Epidemiological studies have observed between city heterogeneity in PM2.5-mortality risk estimates. These differences could potentially be due to the use of central-site monitors as a surrogate for exposure which do not account for an individual's activities or ambient pollutant infiltration to the indoor environment. Therefore, relying solely on central-site monitoring data introduces exposure error in the epidemiological analysis. The amount of exposure error produced by using the central-site monitoring data may differ by city. The objective of this analysis was to cluster cities with similar exposure distributions based on residential infiltration and in-vehicle commuting characteristics. Factors related to residential infiltration and commuting were developed from the American Housing Survey (AHS) from 2001 to 2005 for 94 Core-Based Statistical Areas (CBSAs). We conducted two separate cluster analyses using a k-means clustering algorithm to cluster CBSAs based on these factors. The first only included residential infiltration factors (i.e. percent of homes with central air conditioning (AC) mean year home was built, and mean home size) while the second incorporated both infiltration and commuting (i.e. mean in-vehicle commuting time and mean in-vehicle commuting distance) factors. Clustering on residential infiltration factors resulted in 5 clusters, with two having distinct exposure distributions. Cluster 1 consisted of cities with older, smaller homes with less central AC while homes in Cluster 2 cities were newer, larger, and more likely to have central AC. Including commuting factors resulted in 10 clusters. Clusters with shorter in-vehicle commuting times had shorter in-vehicle commuting distances. Cities with newer homes also tended to have longer commuting times and distances. This is the first study to employ cluster analysis to group cities based on exposure factors. Identifying cities with similar exposure distributions may help explain city-to-city heterogeneity in PM2.5 mortality risk estimates.

Exposure prediction approaches used in air pollution epidemiology studies: key findings and future recommendations.

Many epidemiologic studies of the health effects of exposure to ambient air pollution use measurements from central-site monitors as their exposure estimate. However, measurements from central-site monitors may lack the spatial and temporal resolution required to capture exposure variability in a study population, thus resulting in exposure error and biased estimates. Articles in this dedicated issue examine various approaches to predict or assign exposures to ambient pollutants. These methods include combining existing central-site pollution measurements with local- and/or regional-scale air quality models to create new or "hybrid" models for pollutant exposure estimates and using exposure models to account for factors such as infiltration of pollutants indoors and human activity patterns. Key findings from these articles are summarized to provide lessons learned and recommendations for additional research on improving exposure estimation approaches for future epidemiological studies. In summary, when compared with use of central-site monitoring data, the enhanced spatial resolution of air quality or exposure models can have an impact on resultant health effect estimates, especially for pollutants derived from local sources such as traffic (e.g., EC, CO, and NO(x)). In addition, the optimal exposure estimation approach also depends upon the epidemiological study design. We recommend that future research develops pollutant-specific infiltration data (including for PM species) and improves existing data on human time-activity patterns and exposure to local source (e.g., traffic), in order to enhance human exposure modeling estimates. We also recommend comparing how various approaches to exposure estimation characterize relationships between multiple pollutants in time and space and investigating the impact of improved exposure estimates in chronic health studies.

Development and evaluation of alternative approaches for exposure assessment of multiple air pollutants in Atlanta, Georgia.

Measurements from central site (CS) monitors are often used as estimates of exposure in air pollution epidemiological studies. As these measurements are typically limited in their spatiotemporal resolution, true exposure variability within a population is often obscured, leading to potential measurement errors. To fully examine this limitation, we developed a set of alternative daily exposure metrics for each of the 169 ZIP codes in the Atlanta, GA, metropolitan area, from 1999 to 2002, for PM(2.5) and its components (elemental carbon (EC), SO(4)), O(3), carbon monoxide (CO), and nitrogen oxides (NOx). Metrics were applied in a study investigating the respiratory health effects of these pollutants. The metrics included: (i) CS measurements (one CS per pollutant); (ii) air quality model results for regional background pollution; (iii) local-scale AERMOD air quality model results; (iv) hybrid air quality model estimates (a combination of (ii) and (iii)); and (iv) population exposure model predictions (SHEDS and APEX). Differences in estimated spatial and temporal variability were compared by exposure metric and pollutant. Comparisons showed that: (i) both hybrid and exposure model estimates exhibited high spatial variability for traffic-related pollutants (CO, NO(x), and EC), but little spatial variability among ZIP code centroids for regional pollutants (PM(2.5), SO(4), and O(3)); (ii) for all pollutants except NO(x), temporal variability was consistent across metrics; (iii) daily hybrid-to-exposure model correlations were strong (r>0.82) for all pollutants, suggesting that when temporal variability of pollutant concentrations is of main interest in an epidemiological application, the use of estimates from either model may yield similar results; (iv) exposure models incorporating infiltration parameters, time-location-activity budgets, and other exposure factors affect the magnitude and spatiotemporal distribution of exposure, especially for local pollutants. The results of this analysis can inform the development of more appropriate exposure metrics for future epidemiological studies of the short-term effects of particulate and gaseous ambient pollutant exposure in a community.

Refined ambient PM2.5 exposure surrogates and the risk of myocardial infarction.

Using a case-crossover study design and conditional logistic regression, we compared the relative odds of transmural (full-wall) myocardial infarction (MI) calculated using exposure surrogates that account for human activity patterns and the indoor transport of ambient PM(2.5) with those calculated using central-site PM(2.5) concentrations to estimate exposure to PM(2.5) of outdoor origin (exposure to ambient PM(2.5)). Because variability in human activity and indoor PM(2.5) transport contributes exposure error in epidemiologic analyses when central-site concentrations are used as exposure surrogates, we refer to surrogates that account for this variability as "refined" surrogates. As an alternative analysis, we evaluated whether the relative odds of transmural MI associated with increases in ambient PM(2.5) is modified by residential air exchange rate (AER), a variable that influences the fraction of ambient PM(2.5) that penetrates and persists indoors. Use of refined exposure surrogates did not result in larger health effect estimates (ORs=1.10-1.11 with each interquartile range (IQR) increase), narrower confidence intervals, or better model fits compared with the analysis that used central-site PM(2.5). We did observe evidence for heterogeneity in the relative odds of transmural MI with residential AER (effect-modification), with residents of homes with higher AERs having larger ORs than homes in lower AER tertiles. For the level of exposure-estimate refinement considered here, our findings add support to the use of central-site PM(2.5) concentrations for epidemiological studies that use similar case-crossover study designs. In such designs, each subject serves as his or her own matched control. Thus, exposure error related to factors that vary spatially or across subjects should only minimally impact effect estimates. These findings also illustrate that variability in factors that influence the fraction of ambient PM(2.5) in indoor air (e.g., AER) could possibly bias health effect estimates in study designs for which a spatiotemporal comparison of exposure effects across subjects is conducted.

Air pollution exposure prediction approaches used in air pollution epidemiology studies.

Epidemiological studies of the health effects of outdoor air pollution have traditionally relied upon surrogates of personal exposures, most commonly ambient concentration measurements from central-site monitors. However, this approach may introduce exposure prediction errors and misclassification of exposures for pollutants that are spatially heterogeneous, such as those associated with traffic emissions (e.g., carbon monoxide, elemental carbon, nitrogen oxides, and particulate matter). We review alternative air quality and human exposure metrics applied in recent air pollution health effect studies discussed during the International Society of Exposure Science 2011 conference in Baltimore, MD. Symposium presenters considered various alternative exposure metrics, including: central site or interpolated monitoring data, regional pollution levels predicted using the national scale Community Multiscale Air Quality model or from measurements combined with local-scale (AERMOD) air quality models, hybrid models that include satellite data, statistically blended modeling and measurement data, concentrations adjusted by home infiltration rates, and population-based human exposure model (Stochastic Human Exposure and Dose Simulation, and Air Pollutants Exposure models) predictions. These alternative exposure metrics were applied in epidemiological applications to health outcomes, including daily mortality and respiratory hospital admissions, daily hospital emergency department visits, daily myocardial infarctions, and daily adverse birth outcomes. This paper summarizes the research projects presented during the symposium, with full details of the work presented in individual papers in this journal issue.

Influence of human activity patterns, particle composition, and residential air exchange rates on modeled distributions of PM2.5 exposure compared with central-site monitoring data.

Central-site monitors do not account for factors such as outdoor-to-indoor transport and human activity patterns that influence personal exposures to ambient fine-particulate matter (PM(2.5)). We describe and compare different ambient PM(2.5) exposure estimation approaches that incorporate human activity patterns and time-resolved location-specific particle penetration and persistence indoors. Four approaches were used to estimate exposures to ambient PM(2.5) for application to the New Jersey Triggering of Myocardial Infarction Study. These include: Tier 1, central-site PM(2.5) mass; Tier 2A, the Stochastic Human Exposure and Dose Simulation (SHEDS) model using literature-based air exchange rates (AERs); Tier 2B, the Lawrence Berkeley National Laboratory (LBNL) Aerosol Penetration and Persistence (APP) and Infiltration models; and Tier 3, the SHEDS model where AERs were estimated using the LBNL Infiltration model. Mean exposure estimates from Tier 2A, 2B, and 3 exposure modeling approaches were lower than Tier 1 central-site PM(2.5) mass. Tier 2A estimates differed by season but not across the seven monitoring areas. Tier 2B and 3 geographical patterns appeared to be driven by AERs, while seasonal patterns appeared to be due to variations in PM composition and time activity patterns. These model results demonstrate heterogeneity in exposures that are not captured by the central-site monitor.

Examining the effects of air pollution composition on within region differences in PM2.5 mortality risk estimates.

Multi-city population-based epidemiological studies have observed significant heterogeneity in both the magnitude and direction of city-specific risk estimates, but tended to focus on regional differences in PM2.5 mortality risk estimates. Interpreting differences in risk estimates is complicated by city-to-city heterogeneity observed within regions due to city-to-city variations in the PM2.5 composition and the concentration of gaseous pollutants. We evaluate whether variations in PM2.5 composition and gaseous pollutant concentrations have a role in explaining the heterogeneity in PM2.5 mortality risk estimates observed in 27 US cities from 1997 to 2002. Within each region, we select the two cities with the largest and smallest mortality risk estimate. We compare for each region the within- and between-city concentrations and correlations of PM2.5 constituents and gaseous pollutants. We also attempt to identify source factors through principal component analysis (PCA) for each city. The results of this analysis indicate that identifying a PM constituent(s) that explains the differences in the PM2.5 mortality risk estimates is not straightforward. The difference in risk estimates between cities in the same region may be attributed to a group of pollutants, possibly those related to local sources such as traffic.

Variability in the fraction of ambient fine particulate matter found indoors and observed heterogeneity in health effect estimates.

Exposure to ambient (outdoor-generated) fine particulate matter (PM(2.5)) occurs predominantly indoors. The variable efficiency with which ambient PM(2.5) penetrates and persists indoors is a source of exposure error in air pollution epidemiology and could contribute to observed temporal and spatial heterogeneity in health effect estimates. We used a mass balance approach to model F for several scenarios across which heterogeneity in effect estimates has been observed: with geographic location of residence, residential roadway proximity, socioeconomic status, and central air-conditioning use. We found F is higher in close proximity to primary combustion sources (e.g. proximity to traffic) and for lower income homes. F is lower when PM(2.5) is enriched in nitrate and with central air-conditioning use. As a result, exposure error resulting from variability in F will be greatest when these factors have high temporal and/or spatial variability. The circumstances for which F is lower in our calculations correspond to circumstances for which lower effect estimates have been observed in epidemiological studies and higher F values correspond to higher effect estimates. Our results suggest that variability in exposure misclassification resulting from variability in F is a possible contributor to heterogeneity in PM-mediated health effect estimates.

Evaluating heterogeneity in indoor and outdoor air pollution using land-use regression and constrained factor analysis.

Previous studies have identified associations between traffic exposures and a variety of adverse health effects, but many of these studies relied on proximity measures rather than measured or modeled concentrations of specific air pollutants, complicating interpretability of the findings. An increasing number of studies have used land-use regression (LUR) or other techniques to model small-scale variability in concentrations of specific air pollutants. However, these studies have generally considered a limited number of pollutants, focused on outdoor concentrations (or indoor concentrations of ambient origin) when indoor concentrations are better proxies for personal exposures, and have not taken full advantage of statistical methods for source apportionment that may have provided insight about the structure of the LUR models and the interpretability of model results. Given these issues, the primary objective of our study was to determine predictors of indoor and outdoor residential concentrations of multiple traffic-related air pollutants within an urban area, based on a combination of central site monitoring data; geographic information system (GIS) covariates reflecting traffic and other outdoor sources; questionnaire data reflecting indoor sources and activities that affect ventilation rates; and factor-analytic methods to better infer source contributions. As part of a prospective birth cohort study assessing asthma etiology in urban Boston, we collected indoor and/or outdoor 3-to-4 day samples of nitrogen dioxide (NO2) and fine particulate matter with an aerodynamic diameter or = 2.5 pm (PM2.5) at 44 residences during multiple seasons of the year from 2003 through 2005. We performed reflectance analysis, x-ray fluorescence spectroscopy (XRF), and high-resolution inductively coupled plasma-mass spectrometry (ICP-MS) on particle filters to estimate the concentrations of elemental carbon (EC), trace elements, and water-soluble metals, respectively. We derived multiple indicators of traffic using Massachusetts Highway Department (MHD) data and traffic counts collected outside the residences where the air monitoring was conducted. We used a standardized questionnaire to collect data on home characteristics and occupant behaviors. Additional housing information was collected through property tax records. Ambient concentrations of pollutants as well as meteorological data were collected from centrally located ambient monitors. We used GIS-based LUR models to explain spatial and temporal variability in residential outdoor concentrations of PM2.5, EC, and NO2. We subsequently derived latent-source factors for residential outdoor concentrations using confirmatory factor analysis constrained to nonnegative loadings. We developed LUR models to determine whether GIS covariates and other predictors explain factor variability and thereby support initial factor interpretations. To evaluate indoor concentrations, we developed physically interpretable regression models that explored the relationship between measured indoor and outdoor concentrations, relying on questionnaire data to characterize indoor sources and activities. Because outdoor pollutant concentrations measured directly outside of homes are unlikely to be available for most large epidemiologic studies, we developed regression models to explain indoor concentrations of PM2.5, EC, and NO2 as a function of other, more readily available data: GIS covariates, questionnaire data reflecting both sources and ventilation, and central site monitoring data. As we did for outdoor concentrations, we then derived latent-source factors for residential indoor concentrations and developed regression models explaining variability in these indoor latent-source factors. Finally, to provide insight about the effects of improved characterization of exposures for the results of subsequent epidemiologic investigations, we developed a simulation framework to quantitatively compare the implications of using exposure models derived from validation studies with the use of other surrogate models with varying amounts of measurement error. The concentrations of outdoor PM2.5 were strongly associated with the central site monitor data, whereas EC concentrations showed greater spatial variability, especially during colder months, and were predicted by the length of roadway within 200 m of the home. Outdoor NO2 also showed significant spatial variability, predicted in part by population density and roadway length within 50 m of the home. Our constrained factor analysis of outdoor concentrations produced loadings indicating long-range transport, brake wear and traffic exhaust, diesel exhaust, fuel oil combustion, and resuspended road dust as sources; corresponding LUR models largely corroborated these factor interpretations through covariate significance. For example, long-range transport was predicted by central site PM2.5, and season, brake wear and traffic exhaust and resuspended road dust by traffic and residential density, diesel exhaust by the percentage of diesel traffic on the nearest major road, and fuel oil combustion by population density. Our modeling of the concentrations of indoor pollutants demonstrated substantial variability in indoor-outdoor relationships across constituents, helping to separate constituents dominated by outdoor sources (e.g., S, Se, and V) from those dominated by indoor sources (e.g., Ca and Si). Regression models indicated that indoor PM2.5 was not influenced substantially by local traffic but had significant indoor sources (cooking activity and occupant density), while EC was associated with distance to the nearest designated truck route, and NO2 was associated with both traffic density within 50 m of the home and gas stove usage. Our constrained factor analysis of indoor concentrations helped to separate outdoor-dominated factors from indoor-dominated factors, though some factors appeared to be influenced by both indoor and outdoor sources. Subsequent factor analyses of the indoor-attributable fractions from indoor-outdoor regression models provided generally consistent interpretations of indoor-dominated factors. The use of regression models on indoor factors demonstrated the limited predictive power of questionnaire data related to indoor sources, but reinforced the viability of modeling indoor concentrations of pollutants of ambient origin. In spite of the relatively weak predictive power of some of the indoor-concentration regression models, our epidemiologic simulations illustrated that exposure models with fairly modest R2 values (in the range of 0.3 through 0.4, corresponding with the regression models for PM2.5 and NO2) yielded substantial improvements in epidemiologic study performance relative to the use of exposure proxies that could be applied in the absence of validation studies. In spite of limitations related to sample size and available covariate data, our study demonstrated significant outdoor spatial variability within an urban area in NO2 and in several constituents of airborne particles. LUR techniques combined with constrained factor analysis helped to disentangle the contributions to temporal variability of local, long-range transport, and other sources, ultimately allowing exposures from defined source categories to be investigated in epidemiologic studies. For the indoor residential environment, we demonstrated substantial variability in indoor-outdoor relationships among particle constituents; then, using information from public databases and focused questionnaire data, we were able to predict indoor concentrations for a subset of key pollutants. Constrained factor analysis methods applied to the indoor environment helped to separate indoor sources from outdoor sources. The corresponding indoor regression models had limited predictive power, reinforcing the complexity of characterizing the indoor environment when only limited information about key predictors is available. This finding also underscores the likelihood that these regression models might characterize indoor concentrations of pollutants with ambient origins better than they can the indoor concentrations from all sources. Our findings provide direction for future studies characterizing indoor exposure sources and patterns, and our epidemiologic simulation reinforced the importance of reducing measurement error in a context where many traffic-related air pollutants are influenced by both indoor and outdoor sources. The combination of analytical techniques used in our study could ultimately allow for more refined exposure characterization and evaluation of the relative contributions of various sources to health outcomes in epidemiologic studies.

Effects of exposure measurement error in the analysis of health effects from traffic-related air pollution.

In large epidemiological studies, many researchers use surrogates of air pollution exposure such as geographic information system (GIS)-based characterizations of traffic or simple housing characteristics. It is important to evaluate quantitatively these surrogates against measured pollutant concentrations to determine how their use affects the interpretation of epidemiological study results. In this study, we quantified the implications of using exposure models derived from validation studies, and other alternative surrogate models with varying amounts of measurement error on epidemiological study findings. We compared previously developed multiple regression models characterizing residential indoor nitrogen dioxide (NO(2)), fine particulate matter (PM(2.5)), and elemental carbon (EC) concentrations to models with less explanatory power that may be applied in the absence of validation studies. We constructed a hypothetical epidemiological study, under a range of odds ratios, and determined the bias and uncertainty caused by the use of various exposure models predicting residential indoor exposure levels. Our simulations illustrated that exposure models with fairly modest R(2) (0.3 to 0.4 for the previously developed multiple regression models for PM(2.5) and NO(2)) yielded substantial improvements in epidemiological study performance, relative to the application of regression models created in the absence of validation studies or poorer-performing validation study models (e.g., EC). In many studies, models based on validation data may not be possible, so it may be necessary to use a surrogate model with more measurement error. This analysis provides a technique to quantify the implications of applying various exposure models with different degrees of measurement error in epidemiological research.

Uncertainty and variability in health-related damages from coal-fired power plants in the United States.

The health-related damages associated with emissions from coal-fired power plants can vary greatly across facilities as a function of plant, site, and population characteristics, but the degree of variability and the contributing factors have not been formally evaluated. In this study, we modeled the monetized damages associated with 407 coal-fired power plants in the United States, focusing on premature mortality from fine particulate matter (PM(2.5)). We applied a reduced-form chemistry-transport model accounting for primary PM(2.5) emissions and the influence of sulfur dioxide (SO(2)) and nitrogen oxide (NO(x)) emissions on secondary particulate formation. Outputs were linked with a concentration-response function for PM(2.5)-related mortality that incorporated nonlinearities and model uncertainty. We valued mortality with a value of statistical life approach, characterizing and propagating uncertainties in all model elements. At the median of the plant-specific uncertainty distributions, damages across plants ranged from $30,000 to $500,000 per ton of PM(2.5), $6,000 to $50,000 per ton of SO(2), $500 to $15,000 per ton of NO(x), and $0.02 to $1.57 per kilowatt-hour of electricity generated. Variability in damages per ton of emissions was almost entirely explained by population exposure per unit emissions (intake fraction), which itself was related to atmospheric conditions and the population size at various distances from the power plant. Variability in damages per kilowatt-hour was highly correlated with SO(2) emissions, related to fuel and control technology characteristics, but was also correlated with atmospheric conditions and population size at various distances. Our findings emphasize that control strategies that consider variability in damages across facilities would yield more efficient outcomes.

Land use regression modeling of intra-urban residential variability in multiple traffic-related air pollutants.

BACKGROUND: There is a growing body of literature linking GIS-based measures of traffic density to asthma and other respiratory outcomes. However, no consensus exists on which traffic indicators best capture variability in different pollutants or within different settings. As part of a study on childhood asthma etiology, we examined variability in outdoor concentrations of multiple traffic-related air pollutants within urban communities, using a range of GIS-based predictors and land use regression techniques. METHODS: We measured fine particulate matter (PM2.5), nitrogen dioxide (NO2), and elemental carbon (EC) outside 44 homes representing a range of traffic densities and neighborhoods across Boston, Massachusetts and nearby communities. Multiple three to four-day average samples were collected at each home during winters and summers from 2003 to 2005. Traffic indicators were derived using Massachusetts Highway Department data and direct traffic counts. Multivariate regression analyses were performed separately for each pollutant, using traffic indicators, land use, meteorology, site characteristics, and central site concentrations. RESULTS: PM2.5 was strongly associated with the central site monitor (R2 = 0.68). Additional variability was explained by total roadway length within 100 m of the home, smoking or grilling near the monitor, and block-group population density (R2 = 0.76). EC showed greater spatial variability, especially during winter months, and was predicted by roadway length within 200 m of the home. The influence of traffic was greater under low wind speed conditions, and concentrations were lower during summer (R2 = 0.52). NO2 showed significant spatial variability, predicted by population density and roadway length within 50 m of the home, modified by site characteristics (obstruction), and with higher concentrations during summer (R2 = 0.56). CONCLUSION: Each pollutant examined displayed somewhat different spatial patterns within urban neighborhoods, and were differently related to local traffic and meteorology. Our results indicate a need for multi-pollutant exposure modeling to disentangle causal agents in epidemiological studies, and further investigation of site-specific and meteorological modification of the traffic-concentration relationship in urban neighborhoods.