Encoder-Decoder Full Residual Deep Networks for Robust Regression and Spatiotemporal Estimation.

Although increasing hidden layers can improve the ability of a neural network in modeling complex nonlinear relationships, deep layers may result in degradation of accuracy due to the problem of vanishing gradient. Accuracy degradation limits the applications of deep neural networks to predict continuous variables with a small sample size and/or weak or little invariance to translations. Inspired by residual convolutional neural network in computer vision, we developed an encoder-decoder full residual deep network to robustly regress and predict complex spatiotemporal variables. We embedded full shortcuts from each encoding layer to its corresponding decoding layer in a systematic encoder-decoder architecture for efficient residual mapping and error signal propagation. We demonstrated, theoretically and experimentally, that the proposed network structure with full residual connections can successfully boost the backpropagation of signals and improve learning outcomes. This novel method has been extensively evaluated and compared with four commonly used methods (i.e., plain neural network, cascaded residual autoencoder, generalized additive model, and XGBoost) across different testing cases for continuous variable predictions. For model evaluation, we focused on spatiotemporal imputation of satellite aerosol optical depth with massive nonrandomness missingness and spatiotemporal estimation of atmospheric fine particulate matter [Formula: see text] (PM2.5). Compared with the other approaches, our method achieved the state-of-the-art accuracy, had less bias in predicting extreme values, and generated more realistic spatial surfaces. This encoder-decoder full residual deep network can be an efficient and powerful tool in a variety of applications that involve complex nonlinear relationships of continuous variables, varying sample sizes, and spatiotemporal data with weak or little invariance to translation.

Ensemble-based deep learning for estimating PM2.5 over California with multisource big data including wildfire smoke.

INTRODUCTION: Estimating PM2.5 concentrations and their prediction uncertainties at a high spatiotemporal resolution is important for air pollution health effect studies. This is particularly challenging for California, which has high variability in natural (e.g, wildfires, dust) and anthropogenic emissions, meteorology, topography (e.g. desert surfaces, mountains, snow cover) and land use. METHODS: Using ensemble-based deep learning with big data fused from multiple sources we developed a PM2.5 prediction model with uncertainty estimates at a high spatial (1 km × 1 km) and temporal (weekly) resolution for a 10-year time span (2008-2017). We leveraged autoencoder-based full residual deep networks to model complex nonlinear interrelationships among PM2.5 emission, transport and dispersion factors and other influential features. These included remote sensing data (MAIAC aerosol optical depth (AOD), normalized difference vegetation index, impervious surface), MERRA-2 GMI Replay Simulation (M2GMI) output, wildfire smoke plume dispersion, meteorology, land cover, traffic, elevation, and spatiotemporal trends (geo-coordinates, temporal basis functions, time index). As one of the primary predictors of interest with substantial missing data in California related to bright surfaces, cloud cover and other known interferences, missing MAIAC AOD observations were imputed and adjusted for relative humidity and vertical distribution. Wildfire smoke contribution to PM2.5 was also calculated through HYSPLIT dispersion modeling of smoke emissions derived from MODIS fire radiative power using the Fire Energetics and Emissions Research version 1.0 model. RESULTS: Ensemble deep learning to predict PM2.5 achieved an overall mean training RMSE of 1.54 µg/m3 (R2: 0.94) and test RMSE of 2.29 µg/m3 (R2: 0.87). The top predictors included M2GMI carbon monoxide mixing ratio in the bottom layer, temporal basis functions, spatial location, air temperature, MAIAC AOD, and PM2.5 sea salt mass concentration. In an independent test using three long-term AQS sites and one short-term non-AQS site, our model achieved a high correlation (>0.8) and a low RMSE (<3 µg/m3). Statewide predictions indicated that our model can capture the spatial distribution and temporal peaks in wildfire-related PM2.5. The coefficient of variation indicated highest uncertainty over deciduous and mixed forests and open water land covers. CONCLUSION: Our method can be generalized to other regions, including those having a mix of major urban areas, deserts, intensive smoke events, snow cover and complex terrains, where PM2.5 has previously been challenging to predict. Prediction uncertainty estimates can also inform further model development and measurement error evaluations in exposure and health studies.

Exposure Measurement Error in Air Pollution Studies: The Impact of Shared, Multiplicative Measurement Error on Epidemiological Health Risk Estimates.

Spatiotemporal air pollution models are increasingly being used to estimate health effects in epidemiological studies. Although such exposure prediction models typically result in improved spatial and temporal resolution of air pollution predictions, they remain subject to shared measurement error, a type of measurement error common in spatiotemporal exposure models which occurs when measurement error is not independent of exposures. A fundamental challenge of exposure measurement error in air pollution assessment is the strong correlation and sometimes identical (shared) error of exposure estimates across geographic space and time. When exposure estimates with shared measurement error are used to estimate health risk in epidemiological analyses, complex errors are potentially introduced, resulting in biased epidemiological conclusions. We demonstrate the influence of using a three-stage spatiotemporal exposure prediction model and introduce formal methods of shared, multiplicative measurement error (SMME) correction of epidemiological health risk estimates. Using our three-stage, ensemble learning based nitrogen oxides (NOx) exposure prediction model, we quantified SMME. We conducted an epidemiological analysis of wheeze risk in relation to NOx exposure among school-aged children. To demonstrate the incremental influence of exposure modeling stage, we iteratively estimated the health risk using assigned exposure predictions from each stage of the NOx model. We then determined the impact of SMME on the variance of the health risk estimates under various scenarios. Depending on the stage of the spatiotemporal exposure model used, we found that wheeze odds ratio ranged from 1.16 to 1.28 for an interquartile range increase in NOx. With each additional stage of exposure modeling, the health effect estimate moved further away from the null (OR=1). When corrected for observed SMME, the health effects confidence intervals slightly lengthened, but our epidemiological conclusions were not altered. When the variance estimate was corrected for the potential "worst case scenario" of SMME, the standard error further increased, having a meaningful influence on epidemiological conclusions. Our framework can be expanded and used to understand the implications of using exposure predictions subject to shared measurement error in future health investigations.

Spatiotemporal Imputation of MAIAC AOD Using Deep Learning with Downscaling.

Aerosols have adverse health effects and play a significant role in the climate as well. The Multiangle Implementation of Atmospheric Correction (MAIAC) provides Aerosol Optical Depth (AOD) at high temporal (daily) and spatial (1 km) resolution, making it particularly useful to infer and characterize spatiotemporal variability of aerosols at a fine spatial scale for exposure assessment and health studies. However, clouds and conditions of high surface reflectance result in a significant proportion of missing MAIAC AOD. To fill these gaps, we present an imputation approach using deep learning with downscaling. Using a baseline autoencoder, we leverage residual connections in deep neural networks to boost learning and parameter sharing to reduce overfitting, and conduct bagging to reduce error variance in the imputations. Downscaled through a similar auto-encoder based deep residual network, Modern-Era Retrospective analysis for Research and Applications Version 2 (MERRA-2) GMI Replay Simulation (M2GMI) data were introduced to the network as an important gap-filling feature that varies in space to be used for missingness imputations. Imputing weekly MAIAC AOD from 2000 to 2016 over California, a state with considerable geographic heterogeneity, our full (non-full) residual network achieved mean R2 = 0.94 (0.86) [RMSE = 0.007 (0.01)] in an independent test, showing considerably better performance than a regular neural network or non-linear generalized additive model (mean R2 = 0.78-0.81; mean RMSE = 0.013-0.015). The adjusted imputed as well as combined imputed and observed MAIAC AOD showed strong correlation with Aerosol Robotic Network (AERONET) AOD (R = 0.83; R2 = 0.69, RMSE = 0.04). Our results show that we can generate reliable imputations of missing AOD through a deep learning approach, having important downstream air quality modeling applications.

Association between ambient air pollution and breast cancer risk: The multiethnic cohort study.

Previous studies using different exposure methods to assess air pollution and breast cancer risk among primarily whites have been inconclusive. Air pollutant exposures of particulate matter and oxides of nitrogen were estimated by kriging (NOx , NO2 , PM10 , PM2.5 ), land use regression (LUR, NOx , NO2 ) and California Line Source Dispersion model (CALINE4, NOx , PM2.5 ) for 57,589 females from the Multiethnic Cohort, residing largely in Los Angeles County from recruitment (1993-1996) through 2010. Cox proportional hazards models were used to examine the associations between time-varying air pollution and breast cancer incidence adjusting for confounding factors. Stratified analyses were conducted by race/ethnicity and distance to major roads. Among all women, breast cancer risk was positively but not significantly associated with NOx (per 50 parts per billion [ppb]) and NO2 (per 20 ppb) determined by kriging and LUR and with PM2.5 and PM10 (per 10 µg/m3 ) determined by kriging. However, among women who lived within 500 m of major roads, significantly increased risks were observed with NOx (hazard ratio [HR] = 1.35, 95% confidence interval [95% CI]: 1.02-1.79), NO2 (HR = 1.44, 95% CI: 1.04-1.99), PM10 (HR = 1.29, 95% CI: 1.07-1.55) and PM2.5 (HR = 1.85, 95% CI: 1.15-2.99) determined by kriging and NOx (HR = 1.21, 95% CI:1.01-1.45) and NO2 (HR = 1.26, 95% CI: 1.00-1.59) determined by LUR. No overall associations were observed with exposures assessed by CALINE4. Subgroup analyses suggested stronger associations of NOx and NO2 among African Americans and Japanese Americans. Further studies of multiethnic populations to confirm the effects of air pollution, particularly near-roadway exposures, on the risk of breast cancer is warranted.

Cluster-based bagging of constrained mixed-effects models for high spatiotemporal resolution nitrogen oxides prediction over large regions.

BACKGROUND: Accurate estimation of nitrogen dioxide (NO2) and nitrogen oxide (NOx) concentrations at high spatiotemporal resolutions is crucial for improving evaluation of their health effects, particularly with respect to short-term exposures and acute health outcomes. For estimation over large regions like California, high spatial density field campaign measurements can be combined with more sparse routine monitoring network measurements to capture spatiotemporal variability of NO2 and NOx concentrations. However, monitors in spatially dense field sampling are often highly clustered and their uneven distribution creates a challenge for such combined use. Furthermore, heterogeneities due to seasonal patterns of meteorology and source mixtures between sub-regions (e.g. southern vs. northern California) need to be addressed. OBJECTIVES: In this study, we aim to develop highly accurate and adaptive machine learning models to predict high-resolution NO2 and NOx concentrations over large geographic regions using measurements from different sources that contain samples with heterogeneous spatiotemporal distributions and clustering patterns. METHODS: We used a comprehensive Kruskal-K-means method to cluster the measurement samples from multiple heterogeneous sources. Spatiotemporal cluster-based bootstrap aggregating (bagging) of the base mixed-effects models was then applied, leveraging the clusters to obtain balanced and less correlated training samples for less bias and improvement in generalization. Further, we used the machine learning technique of grid search to find the optimal interaction of temporal basis functions and the scale of spatial effects, which, together with spatiotemporal covariates, adequately captured spatiotemporal variability in NO2 and NOx at the state and local levels. RESULTS: We found an optimal combination of four temporal basis functions and 200 m scale spatial effects for the base mixed-effects models. With the cluster-based bagging of the base models, we obtained robust predictions with an ensemble cross validation R2 of 0.88 for both NO2 and NOx [RMSE (RMSEIQR): 3.62 ppb (0.28) and 9.63 ppb (0.37) respectively]. In independent tests of random sampling, our models achieved similarly strong performance (R2 of 0.87-0.90; RMSE of 3.97-9.69 ppb; RMSEIQR of 0.21-0.27), illustrating minimal over-fitting. CONCLUSIONS: Our approach has important implications for fusing data from highly clustered and heterogeneous measurement samples from multiple data sources to produce highly accurate concentration estimates of air pollutants such as NO2 and NOx at high resolution over a large region.

Exposure measurement error in air pollution studies: A framework for assessing shared, multiplicative measurement error in ensemble learning estimates of nitrogen oxides.

BACKGROUND: Increasingly ensemble learning-based spatiotemporal models are being used to estimate residential air pollution exposures in epidemiological studies. While these machine learning models typically have improved performance, they suffer from exposure measurement error that is inherent in all models. Our objective is to develop a framework to formally assess shared, multiplicative measurement error (SMME) in our previously published three-stage, ensemble learning-based nitrogen oxides (NOx) model to identify its spatial and temporal patterns and predictors. METHODS: By treating the ensembles as an external dosimetry system, we quantified shared and unshared, multiplicative and additive (SUMA) measurement error components in our exposure model. We used generalized additive models (GAMs) with a smooth term for location to identify geographic locations with significantly elevated SMME and explain their spatial and temporal determinants. RESULTS: We found evidence of significant shared and unshared multiplicative error (p < 0.0001) in our ensemble-learning based spatiotemporal NOx model predictions. Unshared multiplicative error was 26 times larger than SMME. We observed significant geographic (p < 0.0001) and temporal variation in SMME with the majority (43%) of predictions with elevated SMME occurring in the earliest time-period (1992-2000). Densely populated urban prediction regions with complex air pollution sources generally exhibited highest odds of elevated SMME. CONCLUSIONS: We developed a novel statistical framework to formally evaluate the magnitude and drivers of SMME in ensemble learning-based exposure models. Our framework can be used to inform building future improved exposure models.

Source Characterization and Exposure Modeling of Gas-Phase Polycyclic Aromatic Hydrocarbon (PAH) Concentrations in Southern California.

Airborne exposures to polycyclic aromatic hydrocarbons (PAHs) are associated with adverse health outcomes. Because personal air measurements of PAHs are labor intensive and costly, spatial PAH exposure models are useful for epidemiological studies. However, few studies provide adequate spatial coverage to reflect intra-urban variability of ambient PAHs. In this study, we collected 39-40 weekly gas-phase PAH samples in southern California twice in summer and twice in winter, 2009, in order to characterize PAH source contributions and develop spatial models that can estimate gas-phase PAH concentrations at a high resolution. A spatial mixed regression model was constructed, including such variables as roadway, traffic, land-use, vegetation index, commercial cooking facilities, meteorology, and population density. Cross validation of the model resulted in an R2 of 0.66 for summer and 0.77 for winter. Results showed higher total PAH concentrations in winter. Pyrogenic sources, such as fossil fuels and diesel exhaust, were the most dominant contributors to total PAHs. PAH sources varied by season, with a higher fossil fuel and wood burning contribution in winter. Spatial autocorrelation accounted for a substantial amount of the variance in total PAH concentrations for both winter (56%) and summer (19%). In summer, other key variables explaining the variance included meteorological factors (9%), population density (15%), and roadway length (21%). In winter, the variance was also explained by traffic density (16%). In this study, source characterization confirmed the dominance of traffic and other fossil fuel sources to total measured gas-phase PAH concentrations while a spatial exposure model identified key predictors of PAH concentrations. Gas-phase PAH source characterization and exposure estimation is of high utility to epidemiologist and policy makers interested in understanding the health impacts of gas-phase PAHs and strategies to reduce emissions.

Long-Term Ambient Temperature and Externalizing Behaviors in Adolescents.

The climate-violence relationship has been debated for decades, and yet most of the supportive evidence has come from ecological or cross-sectional analyses with very limited long-term exposure data. We conducted an individual-level, longitudinal study to investigate the association between ambient temperature and externalizing behaviors of urban-dwelling adolescents. Participants (n = 1,287) in the Risk Factors for Antisocial Behavior Study, in California, were examined during 2000-2012 (aged 9-18 years) with repeated assessments of their externalizing behaviors (e.g., aggression, delinquency). Ambient temperature data were obtained from the local meteorological information system. In adjusted multilevel models, aggressive behaviors significantly increased with rising average temperatures (per 1°C increment) in the preceding 1, 2, or 3 years (respectively, ß = 0.23, 95% confidence interval (CI): 0.00, 0.46; ß = 0.35, 95% CI: 0.06, 0.63; or ß = 0.41, 95% CI: 0.08, 0.74), equivalent to 1.5-3.0 years of delay in age-related behavioral maturation. These associations were slightly stronger among girls and families of lower socioeconomic status but greatly diminished in neighborhoods with more green space. No significant associations were found with delinquency. Our study provides the first individual-level epidemiologic evidence supporting the adverse association of long-term ambient temperature and aggression. Similar approaches to studying meteorology and violent crime might further inform scientific debates on climate change and collective violence.

A spatiotemporal mixed model to assess the influence of environmental and socioeconomic factors on the incidence of hand, foot and mouth disease.

BACKGROUND: As a common infectious disease, hand, foot and mouth disease (HFMD) is affected by multiple environmental and socioeconomic factors, and its pathogenesis is complex. Furthermore, the transmission of HFMD is characterized by strong spatial clustering and autocorrelation, and the classical statistical approach may be biased without consideration of spatial autocorrelation. In this paper, we propose to embed spatial characteristics into a spatiotemporal additive model to improve HFMD incidence assessment. METHODS: Using incidence data (6439 samples from 137 monitoring district) for Shandong Province, China, along with meteorological, environmental and socioeconomic spatial and spatiotemporal covariate data, we proposed a spatiotemporal mixed model to estimate HFMD incidence. Geo-additive regression was used to model the non-linear effects of the covariates on the incidence risk of HFMD in univariate and multivariate models. Furthermore, the spatial effect was constructed to capture spatial autocorrelation at the sub-regional scale, and clusters (hotspots of high risk) were generated using spatiotemporal scanning statistics as a predictor. Linear and non-linear effects were compared to illustrate the usefulness of non-linear associations. Patterns of spatial effects and clusters were explored to illustrate the variation of the HFMD incidence across geographical sub-regions. To validate our approach, 10-fold cross-validation was conducted. RESULTS: The results showed that there were significant non-linear associations of the temporal index, spatiotemporal meteorological factors and spatial environmental and socioeconomic factors with HFMD incidence. Furthermore, there were strong spatial autocorrelation and clusters for the HFMD incidence. Spatiotemporal meteorological parameters, the normalized difference vegetation index (NDVI), the temporal index, spatiotemporal clustering and spatial effects played important roles as predictors in the multivariate models. Efron's cross-validation R2 of 0.83 was acquired using our approach. The spatial effect accounted for 23% of the R2, and notable patterns of the posterior spatial effect were captured. CONCLUSIONS: We developed a geo-additive mixed spatiotemporal model to assess the influence of meteorological, environmental and socioeconomic factors on HFMD incidence and explored spatiotemporal patterns of such incidence. Our approach achieved a competitive performance in cross-validation and revealed strong spatial patterns for the HFMD incidence rate, illustrating important implications for the epidemiology of HFMD.

Longitudinal Analysis of Particulate Air Pollutants and Adolescent Delinquent Behavior in Southern California.

Animal experiments and cross-sectional human studies have linked particulate matter (PM) with increased behavioral problems. We conducted a longitudinal study to examine whether the trajectories of delinquent behavior are affected by PM2.5 (PM with aerodynamic diameter ≤ 2.5 µm) exposures before and during adolescence. We used the parent-reported Child Behavior Checklist at age 9-18 with repeated measures every ~2-3 years (up to 4 behavioral assessments) on 682 children from the Risk Factors for Antisocial Behavior Study conducted in a multi-ethnic cohort of twins born in 1990-1995. Based on prospectively-collected residential addresses and a spatiotemporal model of ambient air concentrations in Southern California, monthly PM2.5 estimates were aggregated to represent long-term (1-, 2-, 3-year average) exposures preceding baseline and cumulative average exposure until the last assessment. Multilevel mixed-effects models were used to examine the association between PM2.5 exposure and individual trajectories of delinquent behavior, adjusting for within-family/within-individual correlations and potential confounders. We also examined whether psychosocial factors modified this association. The results sµggest that PM2.5 exposure at baseline and cumulative exposure during follow-up was significantly associated (p < 0.05) with increased delinquent behavior. The estimated effect sizes (per interquartile increase of PM2.5 by 3.12-5.18 µg/m3) were equivalent to the difference in delinquency scores between adolescents who are 3.5-4 years apart in age. The adverse effect was stronger in families with unfavorable parent-to-child relationships, increased parental stress or maternal depressive symptoms. Overall, these findings sµggest long-term PM2.5 exposure may increase delinquent behavior of urban-dwelling adolescents, with the resulting neurotoxic effect aggravated by psychosocial adversities.

Constrained Mixed-Effect Models with Ensemble Learning for Prediction of Nitrogen Oxides Concentrations at High Spatiotemporal Resolution.

Spatiotemporal models to estimate ambient exposures at high spatiotemporal resolutions are crucial in large-scale air pollution epidemiological studies that follow participants over extended periods. Previous models typically rely on central-site monitoring data and/or covered short periods, limiting their applications to long-term cohort studies. Here we developed a spatiotemporal model that can reliably predict nitrogen oxide concentrations with a high spatiotemporal resolution over a long time span (>20 years). Leveraging the spatially extensive highly clustered exposure data from short-term measurement campaigns across 1-2 years and long-term central site monitoring in 1992-2013, we developed an integrated mixed-effect model with uncertainty estimates. Our statistical model incorporated nonlinear and spatial effects to reduce bias. Identified important predictors included temporal basis predictors, traffic indicators, population density, and subcounty-level mean pollutant concentrations. Substantial spatial autocorrelation (11-13%) was observed between neighboring communities. Ensemble learning and constrained optimization were used to enhance reliability of estimation over a large metropolitan area and a long period. The ensemble predictions of biweekly concentrations resulted in an R2 of 0.85 (RMSE: 4.7 ppb) for NO2 and 0.86 (RMSE: 13.4 ppb) for NOx. Ensemble learning and constrained optimization generated stable time series, which notably improved the results compared with those from initial mixed-effects models.

An Ensemble Spatiotemporal Model for Predicting PM2.5 Concentrations.

Although fine particulate matter with a diameter of <2.5 µm (PM2.5) has a greater negative impact on human health than particulate matter with a diameter of <10 µm (PM10), measurements of PM2.5 have only recently been performed, and the spatial coverage of these measurements is limited. Comprehensively assessing PM2.5 pollution levels and the cumulative health effects is difficult because PM2.5 monitoring data for prior time periods and certain regions are not available. In this paper, we propose a promising approach for robustly predicting PM2.5 concentrations. In our approach, a generalized additive model is first used to quantify the non-linear associations between predictors and PM2.5, the bagging method is used to sample the dataset and train different models to reduce the bias in prediction, and the variogram for the daily residuals of the ensemble predictions is then simulated to improve our predictions. Shandong Province, China, is the study region, and data from 96 monitoring stations were included. To train and validate the models, we used PM2.5 measurement data from 2014 with other predictors, including PM10 data, meteorological parameters, remote sensing data, and land-use data. The validation results revealed that the R² value was improved and reached 0.89 when PM10 was used as a predictor and a kriging interpolation was performed for the residuals. However, when PM10 was not used as a predictor, our method still achieved a CV R² value of up to 0.86. The ensemble of spatial characteristics of relevant factors explained approximately 32% of the variance and improved the PM2.5 predictions. The spatiotemporal modeling approach to estimating PM2.5 concentrations presented in this paper has important implications for assessing PM2.5 exposure and its cumulative health effects.

Environmental Determinants of Aggression in Adolescents: Role of Urban Neighborhood Greenspace.

OBJECTIVE: Neighborhood greenspace improves mental health of urban-dwelling populations, but its putative neurobehavioral benefits in adolescents remain unclear. We conducted a prospective study on urban-dwelling adolescents to examine the association between greenspace in residential neighborhood and aggressive behaviors. METHOD: Participants (n = 1,287) of the Risk Factors for Antisocial Behavior Study, a multi-ethnic cohort of twins and triplets born in 1990 to 1995 and living in Southern California, were examined in 2000 to 2012 (aged 9-18 years) with repeated assessments of their aggressive behaviors by the parent-reported Child Behavior Checklist. Normalized Difference Vegetation Index (NDVI) derived from satellite imagery was used as a proxy for residential neighborhood greenspace aggregated over various spatiotemporal scales before each assessment. Multilevel mixed-effects models were used to estimate the effects of greenspace on aggressive behaviors, adjusting for within-family/within-individual correlations and other potential confounders. RESULTS: Both short-term (1- to 6-month) and long-term (1- to 3-year) exposures to greenspace within 1,000 meters surrounding residences were associated with reduced aggressive behaviors. The benefit of increasing vegetation over the range (∼0.12 in NDVI) commonly seen in urban environments was equivalent to approximately 2 to 2.5 years of behavioral maturation. Sociodemographic factors (e.g., age, gender, race/ethnicity, and socioeconomic status) and neighborhood quality did not confound or modify these associations, and the benefits remained after accounting for temperature. CONCLUSION: Our novel findings support the benefits of neighborhood greenspace in reducing aggressive behaviors of urban-dwelling adolescents. Community-based interventions are needed to determine the efficacy of greenspace as a preemptive strategy to reduce aggressive behaviors in urban environments.

Low birth weight and air pollution in California: Which sources and components drive the risk?

INTRODUCTION: Intrauterine growth restriction has been associated with exposure to air pollution, but there is a need to clarify which sources and components are most likely responsible. This study investigated the associations between low birth weight (LBW, <2500g) in term born infants (≥37 gestational weeks) and air pollution by source and composition in California, over the period 2001-2008. METHODS: Complementary exposure models were used: an empirical Bayesian kriging model for the interpolation of ambient pollutant measurements, a source-oriented chemical transport model (using California emission inventories) that estimated fine and ultrafine particulate matter (PM2.5 and PM0.1, respectively) mass concentrations (4km×4km) by source and composition, a line-source roadway dispersion model at fine resolution, and traffic index estimates. Birth weight was obtained from California birth certificate records. A case-cohort design was used. Five controls per term LBW case were randomly selected (without covariate matching or stratification) from among term births. The resulting datasets were analyzed by logistic regression with a random effect by hospital, using generalized additive mixed models adjusted for race/ethnicity, education, maternal age and household income. RESULTS: In total 72,632 singleton term LBW cases were included. Term LBW was positively and significantly associated with interpolated measurements of ozone but not total fine PM or nitrogen dioxide. No significant association was observed between term LBW and primary PM from all sources grouped together. A positive significant association was observed for secondary organic aerosols. Exposure to elemental carbon (EC), nitrates and ammonium were also positively and significantly associated with term LBW, but only for exposure during the third trimester of pregnancy. Significant positive associations were observed between term LBW risk and primary PM emitted by on-road gasoline and diesel or by commercial meat cooking sources. Primary PM from wood burning was inversely associated with term LBW. Significant positive associations were also observed between term LBW and ultrafine particle numbers modeled with the line-source roadway dispersion model, traffic density and proximity to roadways. DISCUSSION: This large study based on complementary exposure metrics suggests that not only primary pollution sources (traffic and commercial meat cooking) but also EC and secondary pollutants are risk factors for term LBW.

Refining Time-Activity Classification of Human Subjects Using the Global Positioning System.

BACKGROUND: Detailed spatial location information is important in accurately estimating personal exposure to air pollution. Global Position System (GPS) has been widely used in tracking personal paths and activities. Previous researchers have developed time-activity classification models based on GPS data, most of them were developed for specific regions. An adaptive model for time-location classification can be widely applied to air pollution studies that use GPS to track individual level time-activity patterns. METHODS: Time-activity data were collected for seven days using GPS loggers and accelerometers from thirteen adult participants from Southern California under free living conditions. We developed an automated model based on random forests to classify major time-activity patterns (i.e. indoor, outdoor-static, outdoor-walking, and in-vehicle travel). Sensitivity analysis was conducted to examine the contribution of the accelerometer data and the supplemental spatial data (i.e. roadway and tax parcel data) to the accuracy of time-activity classification. Our model was evaluated using both leave-one-fold-out and leave-one-subject-out methods. RESULTS: Maximum speeds in averaging time intervals of 7 and 5 minutes, and distance to primary highways with limited access were found to be the three most important variables in the classification model. Leave-one-fold-out cross-validation showed an overall accuracy of 99.71%. Sensitivities varied from 84.62% (outdoor walking) to 99.90% (indoor). Specificities varied from 96.33% (indoor) to 99.98% (outdoor static). The exclusion of accelerometer and ambient light sensor variables caused a slight loss in sensitivity for outdoor walking, but little loss in overall accuracy. However, leave-one-subject-out cross-validation showed considerable loss in sensitivity for outdoor static and outdoor walking conditions. CONCLUSIONS: The random forests classification model can achieve high accuracy for the four major time-activity categories. The model also performed well with just GPS, road and tax parcel data. However, caution is warranted when generalizing the model developed from a small number of subjects to other populations.

Spatial variability of the effect of air pollution on term birth weight: evaluating influential factors using Bayesian hierarchical models.

BACKGROUND: Epidemiological studies suggest that air pollution is adversely associated with pregnancy outcomes. Such associations may be modified by spatially-varying factors including socio-demographic characteristics, land-use patterns and unaccounted exposures. Yet, few studies have systematically investigated the impact of these factors on spatial variability of the air pollution's effects. This study aimed to examine spatial variability of the effects of air pollution on term birth weight across Census tracts and the influence of tract-level factors on such variability. METHODS: We obtained over 900,000 birth records from 2001 to 2008 in Los Angeles County, California, USA. Air pollution exposure was modeled at individual level for nitrogen dioxide (NO2) and nitrogen oxides (NOx) using spatiotemporal models. Two-stage Bayesian hierarchical non-linear models were developed to (1) quantify the associations between air pollution exposure and term birth weight within each tract; and (2) examine the socio-demographic, land-use, and exposure-related factors contributing to the between-tract variability of the associations between air pollution and term birth weight. RESULTS: Higher air pollution exposure was associated with lower term birth weight (average posterior effects: -14.7 (95 % CI: -19.8, -9.7) g per 10 ppb increment in NO2 and -6.9 (95 % CI: -12.9, -0.9) g per 10 ppb increment in NOx). The variation of the association across Census tracts was significantly influenced by the tract-level socio-demographic, exposure-related and land-use factors. Our models captured the complex non-linear relationship between these factors and the associations between air pollution and term birth weight: we observed the thresholds from which the influence of the tract-level factors was markedly exacerbated or attenuated. Exacerbating factors might reflect additional exposure to environmental insults or lower socio-economic status with higher vulnerability, whereas attenuating factors might indicate reduced exposure or higher socioeconomic status with lower vulnerability. CONCLUSIONS: Our Bayesian models effectively combined a priori knowledge with training data to infer the posterior association of air pollution with term birth weight and to evaluate the influence of the tract-level factors on spatial variability of such association. This study contributes new findings about non-linear influences of socio-demographic factors, land-use patterns, and unaccounted exposures on spatial variability of the effects of air pollution.

A Statewide Nested Case-Control Study of Preterm Birth and Air Pollution by Source and Composition: California, 2001-2008.

BACKGROUND: Preterm birth (PTB) has been associated with exposure to air pollution, but it is unclear whether effects might vary among air pollution sources and components. OBJECTIVES: We studied the relationships between PTB and exposure to different components of air pollution, including gases and particulate matter (PM) by size fraction, chemical composition, and sources. METHODS: Fine and ultrafine PM (respectively, PM2.5 and PM0.1) by source and composition were modeled across California over 2000-2008. Measured PM2.5, nitrogen dioxide, and ozone concentrations were spatially interpolated using empirical Bayesian kriging. Primary traffic emissions at fine scale were modeled using CALINE4 and traffic indices. Data on maternal characteristics, pregnancies, and birth outcomes were obtained from birth certificates. Associations between PTB (n = 442,314) and air pollution exposures defined according to the maternal residence at birth were examined using a nested matched case-control approach. Analyses were adjusted for maternal age, race/ethnicity, education and neighborhood income. RESULTS: Adjusted odds ratios for PTB in association with interquartile range (IQR) increases in average exposure during pregnancy were 1.133 (95% CI: 1.118, 1.148) for total PM2.5, 1.096 (95% CI: 1.085, 1.108) for ozone, and 1.079 (95% CI: 1.065, 1.093) for nitrogen dioxide. For primary PM, the strongest associations per IQR by source were estimated for onroad gasoline (9-11% increase), followed by onroad diesel (6-8%) and commercial meat cooking (4-7%). For PM2.5 composition, the strongest positive associations per IQR were estimated for nitrate, ammonium, and secondary organic aerosols (11-14%), followed by elemental and organic carbon (2-4%). Associations with local traffic emissions were positive only when analyses were restricted to births with residences geocoded at the tax parcel level. CONCLUSIONS: In our statewide nested case-control study population, exposures to both primary and secondary pollutants were associated with an increase in PTB. CITATION: Laurent O, Hu J, Li L, Kleeman MJ, Bartell SM, Cockburn M, Escobedo L, Wu J. 2016. A statewide nested case-control study of preterm birth and air pollution by source and composition: California, 2001-2008. Environ Health Perspect 124:1479-1486; http://dx.doi.org/10.1289/ehp.1510133.

Adverse Reproductive Health Outcomes and Exposure to Gaseous and Particulate-Matter Air Pollution in Pregnant Women.

Introduction: There is growing epidemiologic evidence of associations between maternal exposure to ambient air pollution and adverse birth outcomes, such as preterm birth (PTB). Recently, a few studies have also reported that exposure to ambient air pollution may also increase the risk of some common pregnancy complications, such as preeclampsia and gestational diabetes mellitus (GDM). Research findings, however, have been mixed. These inconsistent results could reflect genuine differences in the study populations, the study locations, the specific pollutants considered, the designs of the study, its methods of analysis, or random variation. Dr. Jun Wu of the University of California­ Irvine, a recipient of HEI's Walter A. Rosenblith New Investigator Award, and colleagues have examined the association between air pollution and adverse birth and pregnancy outcomes in California women. In addition, they examined the effect modification by socioeconomic status (SES) and other factors. Approach: A retrospective nested case­control study was conducted using birth certificate data from about 4.4 million birth records in California from 2001 to 2008. Wu and colleagues analyzed data on low birth weight (LBW) at term (infants born between 37 and 43 weeks of gestation and weighing less than 2500 g), PTB (infants born before 37 weeks of gestation), and preeclampsia (including eclampsia) of the mother during the pregnancy. In addition, they obtained data on GDM for the years 2006­ 2008. In the analyses, all outcomes were included as binary variables. Maternal residential addresses at the time of delivery were geocoded, and a large suite of air pollution exposure metrics was considered, such as (1) regulatory monitoring data on concentrations of criteria pollutants NO2, PM2.5 (particulate matter ≤ 2.5 µm in aerodynamic diameter), and ozone (O3) estimated by empirical Bayesian kriging; (2) concentrations of primary and secondary PM2.5 and PM0.1 components and sources estimated by the University of California­Davis Chemical Transport Model; (3) traffic-related ultrafine particles and concentrations of carbon monoxide (CO) and nitrogen oxides (NOx) estimated by a modified CALINE4 air pollution dispersion model; and (4) proximity to busy roads, road length, and traffic density calculated for different buffer sizes using geographic information system tools. In total, 50 different exposure metrics were available for the analyses. The exposure of primary interest was the mean of the entire pregnancy period for each mother. For the health analyses, controls were randomly selected from the source population. PTB controls were matched on conception year. Term LBW, preeclampsia, and GDM were analyzed using generalized additive mixed models with inclusion of a random effect per hospital. PTB analyses were conducted using conditional logistic regression, with no adjustment for hospital. The main results­ adjusted for race and education as categorical variables and adjusted for maternal age and median household income at the census-block level­were derived from single-pollutant models. Main results and interpretation: In its independent review of the study, the HEI Health Review Committee concluded that Wu and colleagues had conducted a comprehensive nested case­control study of air pollution and adverse birth and pregnancy outcomes. The very large data set and the extensive exposure assessment were strengths of the study. The study documented associations between increases in various air pollution metrics and increased risks of PTB, whereas the evidence was weaker overall for term LBW; in addition, decreases in many air pollution metrics were associated with an increased risk of preeclampsia and GDM, an unexpected result. The investigators suggested that underreporting in the registry data, especially in lower-SES groups, might have caused the many negative associations found for preeclampsia and GDM. In addition, poor geocoding was listed as a potential explanation, affecting in particular the results that were based on measures of proximity to busy roads and traffic density in the smallest buffer size (50 m). However, those issues were not fully explored. In general, the Committee thought that the analysis of road traffic indicators in the 50 m buffer was hampered by the lack of contrast and that the results are therefore difficult to interpret. Some other issues with the analytical approaches should be considered when interpreting the results. Only a subset of controls was used, to reduce computational demands. Hence, some models did not converge, especially in the subgroup analyses. Most of the results in the report were based on analyses using single-pollutant models, which is a reasonable approach but ignores that people are exposed to complex mixtures of pollutants. The Committee believed that the few two-pollutant models that were run provided important insights: these models showed the strongest association for PM2.5 mass, whereas components and source-specific positive associations largely disappeared after adjusting for PM2.5 mass. This study adds to the ongoing debate about whether some particle components and sources are of greater public health concern than others.

Using robust Bayesian network to estimate the residuals of fluoroquinolone antibiotic in soil.

Prediction of antibiotic pollution and its consequences is difficult, due to the uncertainties and complexities associated with multiple related factors. This article employed domain knowledge and spatial data to construct a Bayesian network (BN) model to assess fluoroquinolone antibiotic (FQs) pollution in the soil of an intensive vegetable cultivation area. The results show: (1) The relationships between FQs pollution and contributory factors: Three factors (cultivation methods, crop rotations, and chicken manure types) were consistently identified as predictors in the topological structures of three FQs, indicating their importance in FQs pollution; deduced with domain knowledge, the cultivation methods are determined by the crop rotations, which require different nutrients (derived from the manure) according to different plant biomass. (2) The performance of BN model: The integrative robust Bayesian network model achieved the highest detection probability (pd) of high-risk and receiver operating characteristic (ROC) area, since it incorporates domain knowledge and model uncertainty. Our encouraging findings have implications for the use of BN as a robust approach to assessment of FQs pollution and for informing decisions on appropriate remedial measures.