Bayesian modeling of air pollution health effects with missing exposure data

A scalable two-stage Bayesian approach accounting for exposure measurement error in environmental epidemiology

Accounting for exposure measurement errors has been recognized as a crucial problem in environmental epidemiology for over two decades. Bayesian hierarchical models offer a coherent probabilistic framework for evaluating associations between environmental exposures and health effects, which take into account exposure measurement errors introduced by uncertainty in the estimated exposure as well as spatial misalignment between the exposure and health outcome data. While two-stage Bayesian analyses are often regarded as a good alternative to fully Bayesian analyses when joint estimation is not feasible, there has been minimal research on how to properly propagate uncertainty from the first-stage exposure model to the second-stage health model, especially in the case of a large number of participant locations along with spatially correlated exposures. We propose a scalable two-stage Bayesian approach, called a sparse multivariate normal (sparse MVN) prior approach, based on the Vecchia approximation for assessing associations between exposure and health outcomes in environmental epidemiology. We compare its performance with existing approaches through simulation. Our sparse MVN prior approach shows comparable performance with the fully Bayesian approach, which is a gold standard but is impossible to implement in some cases. We investigate the association between source-specific exposures and pollutant (nitrogen dioxide [NO2])-specific exposures and birth weight of full-term infants born in 2012 in Harris County, Texas, using several approaches, including the newly developed method.

Improving modeled air pollution concentration maps by residual interpolation

Models that are used to map air pollutant concentrations are not free of errors. A possible approach for improving the final concentration map is to interpolate the residuals of the initial model concentration estimates. Due to the possible spatial autocorrelation of the residuals of the initial model estimates, Bayesian inference schemes were suggested for this task, since they can correctly adjust the level of fitting of the residuals to the random measurement errors. However, the complexity of Bayesian methods often discourages their use. Here, we present an alternative and simpler approach, using a leave-one-out cross-validation to determine the optimal level of fitting of the residual correction. We show that the optimal correction level is related to the extent of the spatial autocorrelation of the cross-validated residuals. Namely, when the residuals are not autocorrelated residual correction is unnecessary, and if employed may actually degrade the quality of the final concentration map. Moreover, our approach enables to optimize the residual correction based on different target performance measures, with a possibly different optimal correction depending on the performance measure used. Hence, different target performance measures can be chosen to fit best the specific application of interest. The method is demonstrated using output of three different models used for estimating NO_x and NO₂ concentrations over Israel. We show that our approach can be used as an exploratory step, for assessing the potential benefit of residual correction, and as a simple alternative to Bayesian schemes.

Long term exposure to NO2 and diabetes incidence in the Black Women's Health Study

While laboratory studies show that air pollutants can potentiate insulin resistance, the epidemiologic evidence regarding the association of air pollution with diabetes incidence is conflicting. The purpose of the present study was to assess the association of the traffic-related nitrogen dioxide (NO2) with the incidence of diabetes in a longitudinal cohort study of African American women. We used Cox proportional hazards models to calculate hazard ratios and 95% confidence intervals (CI) for diabetes associated with exposure to NO2 among 43,003 participants in the Black Women's Health Study (BWHS). Pollutant levels at participant residential locations were estimated with 1) a land use regression model for participants living in 56 metropolitan areas, and 2) a dispersion model for participants living in 27 of the cities. From 1995 to 2011, 4387 cases of diabetes occurred. The hazard ratios per interquartile range of NO2 (9.7 ppb), adjusted for age, metropolitan area, education, vigorous exercise, body mass index, smoking, and diet, were 0.96 (95% CI 0.88-1.06) using the land use regression model estimates and 0.94 (95% CI 0.80, 1.10) using the dispersion model estimates. The present results do not support the hypothesis that exposure to NO2 contributes to diabetes incidence in African American women.

Long term effects of traffic noise on mortality in the city of Barcelona, 2004-2007

Numerous studies showing statistically significant associations between environmental noise and adverse health effects already exist for short-term (over one day at most) and long-term (over a year or more) noise exposure, both for morbidity and (albeit to a lesser extent) mortality. Recently, several studies have shown this association to be independent from confounders, mainly those of air pollutants. However, what has not been addressed is the problem of misalignment (i.e. the exposure data locations and health outcomes have different spatial locations). Without any explicit control of such misalignment inference is seriously compromised. Our objective is to assess the long-term effects of traffic noise on mortality in the city of Barcelona (Spain) during 2004-2007. We take into account the control of confounding, for both air pollution and socioeconomic factors at a contextual level and, in particular, we explicitly address the problem of misalignment. We employed a case-control design with individual data. We used deaths resulting from myocardial infarction, hypertension, or Type II diabetes mellitus in Barcelona between 2004 and 2007 as cases for the study, while for controls we used deaths (likewise in Barcelona and over the same period of time) resulting from AIDS or external causes (e.g. accidental falls, accidental poisoning by psychotropic drugs, drugs of abuse, suicide and self-harm, or injuries resulting from motor vehicle accidents). The controls were matched with the cases by sex and age. We used the annual average equivalent A-weighted sound pressure levels for daytime (7-21h), evening-time (21-23h) and night-time (23-7h), and controlled for the following confounders: i) air pollutants (NO2, PM10 and benzene), ii) material deprivation (at a census tract level) and iii) land use and other spatial variables. We explicitly controlled for heterogeneity (uneven distribution of both response and environmental exposures within an area), spatial dependency (of the observations of the response variables), temporal trends (long-term behaviour of the response variables) and spatial misalignment (between response and environmental exposure locations). We used a fully Bayesian method, through the Integrated Nested Laplace Approximation (INLA). Specifically, we plugged the whole model for the exposure into the health model and obtained a linear predictor defined on the entire spatial domain. Separate analyses were carried out for men and for women. After adjusting for confounders, we found that traffic noise was associated with myocardial infarction mortality along with Type II diabetes mellitus in men (in both cases, odds ratios (OR) were around 1.02) and mortality from hypertension in women (ORs around 1.01). Nevertheless, only in the case of hypertension in women, does the association remain statistically significant for all age groups considered (all ages, ≥65 years and ≥75 years).

Determinants of manganese in prenatal dentin of shed teeth from CHAMACOS children living in an agricultural community

Manganese (Mn) is an essential nutrient, but overexposure can be neurotoxic. Over 800 000 kg of Mn-containing fungicides are applied each year in California. Manganese levels in teeth are a promising biomarker of perinatal exposure. Participants in our analysis included 207 children enrolled in the Center for the Health Assessment of Mothers and Children of Salinas (CHAMACOS), a longitudinal birth cohort study in an agricultural area of California. Mn was measured in teeth using laser-ablation-inductively coupled plasma-mass spectrometry. Our purpose was to determine environmental and lifestyle factors related to prenatal Mn levels in shed teeth. We found that storage of farmworkers' shoes in the home, maternal farm work, agricultural use of Mn-containing fungicides within 3 km of the residence, residence built on Antioch Loam soil and Mn dust loading (μg/m(2) of floor area) during pregnancy were associated with higher Mn levels in prenatal dentin (p < 0.05). Maternal smoking during pregnancy was inversely related to Mn levels in prenatal dentin (p < 0.01). Multivariable regression models explained 22-29% of the variability of Mn in prenatal dentin. Our results suggest that Mn measured in prenatal dentin provides retrospective and time specific levels of fetal exposure resulting from environmental and occupational sources.

Childhood exposure to fine particulate matter and black carbon and the development of new wheeze between ages 5 and 7 in an urban prospective cohort

BACKGROUND

While exposures to urban fine particulate matter (PM(2.5)) and soot-black carbon (soot-BC) have been associated with asthma exacerbations, there is limited evidence on whether these pollutants are associated with the new development of asthma or allergy among young inner city children. We hypothesized that childhood exposure to PM(2.5) and the soot-BC component would be associated with the report of new wheeze and development of seroatopy in an inner city birth cohort.

METHODS

As part of the research being conducted by the Columbia Center of Children's Environmental Health (CCCEH) birth cohort study in New York City, two-week integrated residential monitoring of PM(2.5), soot-BC (based on a multi-wavelength integrating sphere method), and modified absorption coefficient (Abs*; based on the smoke stain reflectometer) was conducted between October 2005 and May 2011 for 408 children at ages 5-6 years old. Residential monitoring was repeated 6 months later (n=262) to capture seasonal variability. New wheeze was identified through the International Study of Asthma and Allergies in Childhood (ISAAC) questionnaires during up to 3 years of follow-up and compared to a reference group that reported never wheeze, remitted wheeze, or persistent wheeze. Specific immunoglobulin (Ig) E against cockroach, mouse, cat, and dust mite and total IgE levels was measured in sera at ages 5 and 7 years.

RESULTS

PM(2.5), soot-BC, and Abs* measured at the first visit were correlated moderately with those at the second visit (Pearson r>0.44). Using logistic regression models, a positive association between PM(2.5) and new wheeze was found with adjusted odds ratio [95% confidence intervals] of 1.51 [1.05-2.16] per interquartile range (IQR). Positive but non-significant association was found between the development of new wheeze and soot-BC and (OR 1.40 [0.96-2.05]), and Abs* (OR 1.57 [0.91-2.68]); Significantly positive associations were found between air pollutant measurements and new wheeze when restricting to those participants with repeat home indoor measurements 6 months apart. Associations between pollutants and IgE levels were not detected.

CONCLUSIONS

Our findings suggest that childhood exposure to indoor air pollution, much of which penetrated readily from outdoor sources, may contribute to the development of wheeze symptoms among children ages 5 to 7 years.

Examining the representativeness of home outdoor PM(2.5), EC, and OC estimates for daily personal exposures in Southern California

Recent studies have linked acute respiratory and cardiovascular outcomes to measurements or estimates of traffic-related air pollutants at homes or schools. However, few studies have evaluated these outdoor measurements and estimates against personal exposure measurements. We compared measured and modeled home outdoor concentrations with personal measurements of traffic-related air pollutants in the Los Angeles air basin (Whittier and Riverside). Personal exposure of 63 children with asthma and 15 homes were assessed for particulate matter with an aerodynamic diameter less than 2.5 μm (PM(2.5)), elemental carbon (EC), and organic carbon (OC) during sixteen 10-day monitoring runs. Regression models to predict daily home outdoor PM(2.5), EC, and OC were constructed using home outdoor measurements, geographical and meteorological parameters, as well as CALINE4 estimates at outdoor home sites, which represent the concentrations from local traffic sources. These home outdoor models showed the variance explained (R(2)) was 0.97 and 0.94 for PM(2.5), 0.91 and 0.83 for OC, and 0.76 and 0.87 for EC in Riverside and Whittier, respectively. The PM(2.5) outdoor estimates correlated well with the personal measurements (Riverside R(2) = 0.65 and Whittier R(2) = 0.69). However, excluding potentially inaccurate samples from Riverside, the correlation between personal exposure to carbonaceous species and home outdoor estimates in Whittier was moderate for EC (R(2) = 0.37) and poor for OC (R(2) = 0.08). The CALINE4 estimates alone were not correlated with personal measurements of EC or other pollutants. While home outdoor estimates provide good approximations for daily personal PM(2.5) exposure, they may not be adequate for estimating daily personal exposure to EC and OC. ELECTRONIC SUPPLEMENTARY MATERIAL: The online version of this article (doi:10.1007/s11869-010-0099-y) contains supplementary material, which is available to authorized users.

Statistical modeling of crystalline silica exposure by trade in the construction industry using a database compiled from the literature

A quantitative determinants-of-exposure analysis of respirable crystalline silica (RCS) levels in the construction industry was performed using a database compiled from an extensive literature review. Statistical models were developed to predict work-shift exposure levels by trade. Monte Carlo simulation was used to recreate exposures derived from summarized measurements which were combined with single measurements for analysis. Modeling was performed using Tobit models within a multimodel inference framework, with year, sampling duration, type of environment, project purpose, project type, sampling strategy and use of exposure controls as potential predictors. 1346 RCS measurements were included in the analysis, of which 318 were non-detects and 228 were simulated from summary statistics. The model containing all the variables explained 22% of total variability. Apart from trade, sampling duration, year and strategy were the most influential predictors of RCS levels. The use of exposure controls was associated with an average decrease of 19% in exposure levels compared to none, and increased concentrations were found for industrial, demolition and renovation projects. Predicted geometric means for year 1999 were the highest for drilling rig operators (0.238 mg m(-3)) and tunnel construction workers (0.224 mg m(-3)), while the estimated exceedance fraction of the ACGIH TLV by trade ranged from 47% to 91%. The predicted geometric means in this study indicated important overexposure compared to the TLV. However, the low proportion of variability explained by the models suggests that the construction trade is only a moderate predictor of work-shift exposure levels. The impact of the different tasks performed during a work shift should also be assessed to provide better management and control of RCS exposure levels on construction sites.

Bayesian time-series analysis of a repeated-measures poisson outcome with excess zeroes

In this article, the authors demonstrate a time-series analysis based on a hierarchical Bayesian model of a Poisson outcome with an excessive number of zeroes. The motivating example for this analysis comes from the intensive care unit (ICU) of an urban university teaching hospital (New Haven, Connecticut, 2002-2004). Studies of medication use among older patients in the ICU are complicated by statistical factors such as an excessive number of zero doses, periodicity, and within-person autocorrelation. Whereas time-series techniques adjust for autocorrelation and periodicity in outcome measurements, Bayesian analysis provides greater precision for small samples and the flexibility to conduct posterior predictive simulations. By applying elements of time-series analysis within both frequentist and Bayesian frameworks, the authors evaluate differences in shift-based dosing of medication in a medical ICU. From a small sample and with adjustment for excess zeroes, linear trend, autocorrelation, and clinical covariates, both frequentist and Bayesian models provide evidence of a significant association between a specific nursing shift and dosing level of a sedative medication. Furthermore, the posterior distributions from a Bayesian random-effects Poisson model permit posterior predictive simulations of related results that are potentially difficult to model.

Traffic-related air pollution and asthma onset in children: a prospective cohort study with individual exposure measurement

BACKGROUND

The question of whether air pollution contributes to asthma onset remains unresolved.

OBJECTIVES

In this study, we assessed the association between asthma onset in children and traffic-related air pollution.

METHODS

We selected a sample of 217 children from participants in the Southern California Children's Health Study, a prospective cohort designed to investigate associations between air pollution and respiratory health in children 10-18 years of age. Individual covariates and new asthma incidence (30 cases) were reported annually through questionnaires during 8 years of follow-up. Children had nitrogen dioxide monitors placed outside their home for 2 weeks in the summer and 2 weeks in the fall-winter season as a marker of traffic-related air pollution. We used multilevel Cox models to test the associations between asthma and air pollution.

RESULTS

In models controlling for confounders, incident asthma was positively associated with traffic pollution, with a hazard ratio (HR) of 1.29 [95% confidence interval (CI), 1.07-1.56] across the average within-community interquartile range of 6.2 ppb in annual residential NO2. Using the total interquartile range for all measurements of 28.9 ppb increased the HR to 3.25 (95% CI, 1.35-7.85).

CONCLUSIONS

In this cohort, markers of traffic-related air pollution were associated with the onset of asthma. The risks observed suggest that air pollution exposure contributes to new-onset asthma.

Use of multiple imputation in the epidemiologic literature

The authors attempted to catalog the use of procedures to impute missing data in the epidemiologic literature and to determine the degree to which imputed results differed in practice from unimputed results. The full text of articles published in 2005 and 2006 in four leading epidemiologic journals was searched for the text imput. Sixteen articles utilizing multiple imputation, inverse probability weighting, or the expectation-maximization algorithm to impute missing data were found. The small number of relevant manuscripts and diversity of detail provided precluded systematic analysis of the use of imputation procedures. To form a bridge between current and future practice, the authors suggest details that should be included in articles that utilize these procedures.

Panel discussion review: session three--issues involved in interpretation of epidemiologic analyses--statistical modeling

The Clean Air Act mandates that the US Environmental Protection Agency (EPA) develop National Ambient Air Quality Standards for criteria air pollutants and conduct periodic reviews of the standards based on new scientific evidence. In recent reviews, evidence from epidemiologic studies has played a key role. Epidemiologic studies often provide evidence for effects of several air pollutants. Determining whether there are independent effects of the separate pollutants is a challenge. Among the many issues confronting the interpretation of epidemiologic studies of multi-pollutant exposures and health effects are those specifically related to statistical modeling. The EPA convened a workshop on 13 and 14 December 2006 in Chapel Hill, North Carolina, USA, to discuss these and other issues; Session Three of the workshop was devoted specifically to statistical modeling. Prominent statistical modeling issues in epidemiologic studies of air pollution include (1) measurement error across the co-pollutants; (2) correlation and multi-collinearity among the co-pollutants; (3) the timing of the concentration-response function; (4) confounding; and (5) spatial analyses.

Error and bias in determining exposure potential of children at school locations using proximity-based GIS techniques

BACKGROUND

The widespread availability of powerful tools in commercial geographic information system (GIS) software has made address geocoding a widely employed technique in spatial epidemiologic studies.

OBJECTIVE

The objective of this study was to determine the effect of the positional error in geocoding on the analysis of exposure to traffic-related air pollution of children at school locations.

METHODS

For a case study of Orange County, Florida, we determined the positional error of geocoding of school locations through comparisons with a parcel database and digital orthophotography. We used four different geocoding techniques for comparison to establish the repeatability of geocoding, and an analysis of proximity to major roads to determine bias and error in environmental exposure assessment.

RESULTS

RESULTS INDICATE THAT THE POSITIONAL ERROR IN GEOCODING OF SCHOOLS IS VERY SUBSTANTIAL: We found that the 95% root mean square error was 196 m using street centerlines, 306 m using TIGER roads, and 210 and 235 m for two commercial geocoding firms. We found bias and error in proximity analysis to major roads to be unacceptably large at distances of < 500 m. Bias and error are introduced by lack of positional accuracy and lack of repeatability of geocoding of school locations.

CONCLUSIONS

These results suggest that typical geocoding is insufficient for fine-scale analysis of school locations and more accurate alternatives need to be considered.

Assessing uncertainty in spatial exposure models for air pollution health effects assessment

BACKGROUND

Although numerous epidemiologic studies now use models of intraurban exposure, there has been little systematic evaluation of the performance of different models.

OBJECTIVES

In this present article we proposed a modeling framework for assessing exposure model performance and the role of spatial autocorrelation in the estimation of health effects.

METHODS

We obtained data from an exposure measurement substudy of subjects from the Southern California Children's Health Study. We examined how the addition of spatial correlations to a previously described unified exposure and health outcome modeling framework affects estimates of exposure-response relationships using the substudy data. The methods proposed build upon the previous work, which developed measurement-error techniques to estimate long-term nitrogen dioxide exposure and its effect on lung function in children. In this present article, we further develop these methods by introducing between- and within-community spatial autocorrelation error terms to evaluate effects of air pollution on forced vital capacity. The analytical methods developed are set in a Bayesian framework where multistage models are fitted jointly, properly incorporating parameter estimation uncertainty at all levels of the modeling process.

RESULTS

Results suggest that the inclusion of residual spatial error terms improves the prediction of adverse health effects. These findings also demonstrate how residual spatial error may be used as a diagnostic for comparing exposure model performance.