Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2025 Sep 1.
Published in final edited form as: Environ Res. 2024 May 18;256:119178. doi: 10.1016/j.envres.2024.119178

A Comparison of PM2.5 Exposure Estimates from Different Estimation Methods and their Associations with Cognitive Testing and Brain MRI Outcomes

Melinda C Power 1, Katie M Lynch 1, Erin E Bennett 1, Qi Ying 2, Eun Sug Park 3, Xiaohui Xu 4, Richard L Smith 5,6, James D Stewart 7, Jeff D Yanosky 8, Duanping Liao 8, Aaron van Donkelaar 9, Joel D Kaufman 10,11,12, Lianne Sheppard 12,13, Adam A Szpiro 13, Eric A Whitsel 7,14
PMCID: PMC11186721  NIHMSID: NIHMS1998659  PMID: 38768885

Abstract

Background:

Reported associations between particulate matter with aerodynamic diameter ≤ 2.5μm (PM2.5) and cognitive outcomes remain mixed. Differences in exposure estimation method may contribute to this heterogeneity.

Objectives:

To assess agreement between PM2.5 exposure concentrations across 11 exposure estimation methods and to compare resulting associations between PM2.5 and cognitive or MRI outcomes.

Methods:

We used Visit 5 (2011–2013) cognitive testing and brain MRI data from the Atherosclerosis Risk in Communities (ARIC) Study. We derived address-linked average 2000–2007 PM2.5 exposure concentrations in areas immediately surrounding the four ARIC recruitment sites (Forsyth County, NC; Jackson, MS; suburbs of Minneapolis, MN; Washington County, MD) using 11 estimation methods. We assessed agreement between method-specific PM2.5 concentrations using descriptive statistics and plots, overall and by site. We used adjusted linear regression to estimate associations of method-specific PM2.5 exposure estimates with cognitive scores (n=4,678) and MRI outcomes (n=1,518) stratified by study site and combined site-specific estimates using meta-analyses to derive overall estimates. We explored the potential impact of unmeasured confounding by spatially patterned factors.

Results:

Exposure estimates from most methods had high agreement across sites, but low agreement within sites. Within-site exposure variation was limited for some methods. Consistently null findings for the PM2.5-cognitive outcome associations regardless of method precluded empirical conclusions about the potential impact of method on study findings in contexts where positive associations are observed. Not accounting for study site led to consistent, adverse associations, regardless of exposure estimation method, suggesting the potential for substantial bias due to residual confounding by spatially patterned factors.

Discussion:

PM2.5 estimation methods agreed across sites but not within sites. Choice of estimation method may impact findings when participants are concentrated in small geographic areas. Understanding unmeasured confounding by factors that are spatially patterned may be particularly important in studies of air pollution and cognitive or brain health.

Keywords: Air pollution, particulate matter, exposure assessment, dementia, cognition

1. INTRODUCTION

Ambient air pollution is an important environmental health risk factor due to its adverse toxicological effects on human health and its global health impact (Health Effects Institute 2020; Landrigan et al. 2018; U.S. EPA. 2019). According to the World Health Organization, air pollution contributes to the global burden of noncommunicable diseases, including stroke, heart attack, lung cancer, and chronic obstructive pulmonary diseases (World Health Assembly 71 2018). High exposure to particulate matter <2.5 μm in diameter (PM2.5) appears particularly detrimental to human health. A U.S. Environmental Protection Agency (EPA) assessment concluded there is a causal relationship of PM2.5 exposure with cardiovascular effects and non-accidental mortality, and a likely causal relationship of PM2.5 exposure with cancers and respiratory and nervous system effects (U.S. EPA 2019).

The effects of PM2.5 on human health likely extend beyond these established effects. One area of growing interest is the potential impact of PM2.5 on late-life cognitive health. While the weight of evidence currently suggests an association between higher PM2.5 exposure and accelerated cognitive decline, a limited number of studies report on this outcome. Further, studies of PM2.5 and cognitive level, neuroimaging findings, and incident cognitive impairment are less consistent (Weuve et al. 2021). Overall, there is substantial heterogeneity in findings across studies; whether this is attributable to heterogeneity of affect across populations or pollutants, or whether this is attributable to bias remains unclear (Weuve et al. 2021).

One factor that may contribute to this heterogeneity is the use of different exposure estimation methods across studies. Broadly, methods for air pollution exposure estimation include geostatistical interpolation, geographic information system (GIS)-based statistical models, air dispersion and chemical transport models, satellite models, and hybrid models (Diao et al. 2019; Jerrett et al. 2005; Xie et al. 2017). Choice of estimation approach is not straightforward, as there is no gold standard for estimation of air pollution exposures across space and time, and each approach has strengths and limitations. A limited number of studies have explicitly compared long-term average air pollution exposures across estimation methods or quantified the impact of estimation method choice on health effect associations in national or regional samples (Bauwelinck et al. 2022; de Hoogh et al. 2014; Hart et al. 2015; Jerrett et al. 2017; Jin et al. 2019; Kelly et al. 2021; Klompmaker et al. 2021; McGuinn et al. 2017; Wang et al. 2015); several compare dispersion and land use regression (LUR) methods in Europe (Bauwelinck et al. 2022; de Hoogh et al. 2014; Klompmaker et al. 2021; Wang et al. 2015). Overall, they report similar mean exposure estimates, with some differences in estimated variability. Associations between PM2.5 and mortality, cardiovascular and respiratory outcomes across exposure methods were often similar, although differences in magnitude of the estimates were reported in some studies. However, differences across methods have been reported in the handful of studies of PM2.5 and cognition that report estimates from a second model for the pollutant as a sensitivity analysis, which can provide insight into the potential impact of exposure method on associations with cognitive outcomes (Cullen et al. 2018; Tzivian et al. 2016; Yuchi et al. 2020). Specifically, two studies found differences in significance of associations between methods (Tzivian et al. 2016; Yuchi et al. 2020), and one found differences in magnitude (Cullen et al. 2018).

Thus, our aim is to compare estimates of long-term average air pollution exposures produced by different research groups and to evaluate whether the choice of estimation method influences estimated associations of PM2.5 with cognitive test performance and dementia-related neuroimaging outcomes. To do so, we build upon prior work in the Atherosclerosis Risk in Communities (ARIC) Study cohort (Power et al. 2018; The ARIC Investigators 1989; Wright et al. 2021), which quantified the association between particulate matter exposures and brain MRI findings using a single method of PM2.5 exposure estimation. Here, we compare estimates of long-term average PM2.5 exposures at ARIC participant addresses and examine associations with brain MRI findings and concurrent cognitive status using a total of 11 air pollution estimation methods developed by multiple research groups.

2. METHODS

2.1. Cohort Description

The Atherosclerosis Risk in Communities (ARIC) Study is a cohort of individuals recruited in 1987 to 1989 from four sites in the United States: Forsyth County, North Carolina (NC); Jackson, Mississippi (MS); the suburbs of Minneapolis, Minnesota (MN); and Washington County, Maryland (MD). At recruitment, 15,792 45- to 64-year-olds were enrolled. Through 2020, the cohort has completed eight study visits; we use cognitive and neuroimaging data collected at Visit 5 (2011–2013). Our methods are similar to a prior ARIC analysis of particulate matter and Visit 5 MRI outcomes (Power et al. 2018), but extend the analysis to consider associations between PM2.5 and cognitive function at Visit 5. The ARIC study was approved by the institutional review boards of all participating institutions, and written informed consent was obtained from participants for each study visit.

2.2. Air Pollution Exposure

Air pollution exposure estimates were assigned to participants based on their accurately and reliably geocoded residential addresses (Whitsel et al. 2006; Whitsel et al. 2004) by one author (JS). Addresses from recruitment through 2018 were obtained from data collected at each study visit through Visit 4 (1996–1998) and at annual or semi-annual follow-up calls from Visit 4 through 2018. We considered average PM2.5 exposures from 2000–2007 using 11 different estimation approaches (Table 1). We note that this averaging period was slightly different than the averaging periods used in the prior work (1990–1998, 1999–2007, and 1990–2007) (Power et al. 2018). Here we use the average from 2000–2007 because all models under consideration generated estimates for this time period and because we felt this averaging period remained a reasonable proxy for long-term prior exposures, which we believe to be etiologically relevant.

Table 1.

Methods used to estimate average PM2.5 exposure from 2000 to 2007 at ARIC participant addresses

Method Label Method Description Spatial Resolution
M01 GAMM with LUR (Yanosky et al. 2014) Point estimates
M02 National Log-Normal Ordinary Kriging (Liao et al. 2006; Liao D 2013) Point estimates
M03 National Universal Kriging with LUR & PLSR (National Historical PM2.5 Historical Model) (Kim et al. 2017) Point estimates
M04 Regionalized Universal Kriging with LUR & PLSR (National Spatiotemporal PM2.5 Model) (Keller et al. 2015) Point estimates
M05 Satellite-Based, using a Chemical Transport Model and Ground-Based Observations(Van Donkelaar et al. 2019) 0.01° × 0.01° grid (~1.1km at the equator)
M06 CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 1.33km grid
M07 CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 4km grid
M08 CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 12km grid
M09 CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 36km grid
M10 Nearest Neighbor Point estimates
M11 Inverse Distance Weighted Mean Point estimates
Open in a new tab

Abbreviations: Community Multiscale Air Quality chemical transport model (CMAQ); Emission Database for Global Atmospheric Research (EDGAR); generalized additive mixed model (GAMM); land use regression (LUR); the National Emissions Inventory (NEI); partial least squares regression (PLSR); particulate matter of size 2.5 microns or less in diameter (PM2.5)

A more detailed description of each PM2.5 estimation method is available in the Appendix and the cited literature. Briefly, Method 1 (M01) is a GIS-based spatiotemporal generalized additive mixed model (GAMM) which was developed using monitor PM2.5 concentrations, geographic data, and meteorological data, along with spatial smoothing to predict monthly average air pollution levels (Yanosky et al. 2014). Method 2 (M02) is a national log-normal ordinary kriging approach based solely on daily average monitor PM2.5 concentrations, derived using ArcGIS Version 10.2 (Liao et al. 2006; Liao D 2013). Method 3 (M03) (Kim et al. 2017) and Method 4 (M04) (Keller et al. 2015; Kirwa et al. 2021) estimate PM2.5 concentrations using a spatiotemporal modeling framework involving universal kriging with land-use regression and partial least squares regression (PLSR) to reduce geographic covariates, as well as information on time trends in exposure; Method 3 directly models annual estimates and uses a national approach, while Method 4 produces one-week average estimates using a regional approach. Method 5 (M05) is a satellite-based method that transforms observations of aerosol optical depth (AOD) to PM2.5 estimates using the GEOS-Chem chemical transport model (CTM), which are then calibrated using PM2.5 monitor data to produce gridded 0.01° × 0.01° resolution estimates (Van Donkelaar et al. 2019). Methods 6 to 9 (M06 to M09) combined the output of two runs of the Community Multiscale Air Quality model (CMAQ) using different emissions inventories, the National Emissions Inventory (NEI) and the Emission Database for Global Atmospheric Research (EDGAR), and then calibrated the combined estimates with PM2.5 monitor data to produce gridded exposure estimates within four domains, with variable grid sizes for predictions, covering nested areas of the United States. The 36-km resolution estimates were derived for the continental U.S., whereas 1.33km, 4km, and 12km grid resolution estimates were derived only for areas surrounding the ARIC study sites, with area covered decreasing with increasing resolution (see Appendix). Finally, two basic interpolation methods were used for comparison. Method 10 (M10) estimated annual PM2.5 concentrations at participant addresses using a nearest neighbor approach, while Method 11 (M11) used inverse distance weighting to derive mean annual average PM2.5 concentrations when participant addresses were within 50 km of at least one PM2.5 monitor, and a nearest neighbor approach when there were no PM2.5 monitors within 50km.

2.3. Outcomes

Overall and domain-specific (executive function, memory, and language) summary measures of cognition were derived based on previously established methods from participant scores on a battery of cognitive tests administered at Visit 5 (Gross et al. 2015; Rawlings et al. 2016; Schneider et al. 2015; The ARIC Investigators 2020). The executive function domain included the Digit Symbol Substitution Test (DSST) (Wechsler 1987) and Trail Making Tests Part A and B (Reitan 1958). The memory domain included the Delayed Word Recall Test (DWRT) (Knopman and Ryberg 1989), Incidental Learning Test (Ryan 2001), and the Logical Memory Test (Wechsler 1987). The language domain included the Word Fluency Test (Benton 1976), Animal Naming (Benton 1976), and the Boston Naming Test (Williams et al. 1989). The global cognition score was derived from all tests included in the three domains, along with the Digit Span Backwards Test (Wechsler 1987). Each summary measure was standardized based on the sample mean and standard deviation.

A subset of ARIC Visit 5 participants were selected to complete brain MRI. After excluding those with contraindications, all ARIC Visit 5 participants with cognitive impairment (i.e., low cognitive test scores/declines on cognitive tests over time) or previous completion of a brain MRI in the ARIC study were invited to complete a brain MRI (Knopman et al. 2015). Of the remaining participants, an age-stratified random sample at each site was also selected. Standardized protocols were used to conduct 3T brain MRI at each study site. All imaging analysis was conducted at the Mayo Clinic, Minnesota. Details for scanning and processing are available elsewhere (Knopman et al. 2015; Power et al. 2018; Schneider et al. 2017). Available MRI outcomes mirrored those used previously (Power et al. 2018) and included brain volumes (total brain, deep gray matter, frontal lobe, white matter hyperintensity, occipital lobe, parietal lobe, temporal lobe, hippocampal, and Alzheimer’s Disease (AD) signature region of interest (Dickerson et al. 2011)), and the presence of infarcts, lacunes, severe white matter hyperintensities, and microbleeds. White matter hyperintensity volume (WMH) was base-2 log-transformed for use in analyses due to its skewed distribution, and the remaining volumetric measures were standardized based on the sample mean and standard deviation to aid in interpretation. All other MRI outcomes were dichotomous.

2.4. Statistical Methods

2.4.1. Masking

During analyses, manuscript development, and initial co-author review, the air pollution methods remained masked to all authors except the honest broker who created the air pollution dataset (JS). The air pollution estimation methods were unmasked after the initial analysis, interpretation, and first round of co-author comments were received, but before manuscript completion.

2.4.2. Study Sample

We compare exposure estimates and compute PM2.5-outcome associations in two nested samples, labeled the “NCS” (neurocognitive study) sample and the “MRI” sample. Eligible ARIC participants for both samples included those who attended Visit 5 and who were in a sufficiently large race-site category (White in Forsyth County, NC; Black in Forsyth County, NC; Black in Jackson, MS; White in Minneapolis, MN; and White in Washington County, MD; n=6,066). Among these eligible participants, those missing the global cognitive score or any cognitive domain score (n=265), missing exposure estimates from any of the 11 PM2.5 estimation methods (n=749), or missing important covariates (n=374) were excluded from the NCS sample for a final sample size of n=4,678. Note that the requirement to have exposure estimates for all 11 PM2.5 models limited the analysis to the domains in which 1.33km CMAQ-NEI/CMAQ-EDGAR + observation data fusing estimates were available, which are relatively small geographic areas centered on the four ARIC sites (see Figure 1; the 1.33km domains are approximately 10,000 km2 or 3,860 square miles, which is approximately the combined area of Delaware and Rhode Island or the area of Cyprus). For the MRI sample, we further restricted the NCS sample to the subset of persons who completed 3T MRI at Visit 5 (n=1,535), and then excluded those with missing total brain volume (n=4); a history of surgery or radiation to the head, multiple sclerosis, brain tumors (n=12); or implausible or missing estimated total intracranial volume (n=1), for a final analytical sample of n=1,518.

Figure 1.

Figure 1.

Open in a new tab

Nested modeling domains for the CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing method centered on the ARIC study sites

2.4.3. Comparison of Exposure Estimates Across Estimation Methods

First, we used descriptive statistics and plots to assess agreement between the 2000–2007 average PM2.5 estimates in both the NCS and MRI samples. Specifically, we calculated the means, standard deviations, ranges, minima, maxima, and percentiles, and we created box and whisker and density plots to illustrate the distributions of estimated exposures overall and stratified by study site. Second, we evaluated absolute differences in long-term air pollution exposure estimates across methods. We calculated mean bias and mean error between all possible pairs of estimation methods and created heat maps of those results to aid interpretation. Third, we assessed relative agreement between air pollution estimates. We calculated spatial R2s for each pair of methods, defined as the square of the Pearson correlation coefficient between long-term average exposures derived from the different methods across the common set of addresses. We also created scatterplots with Deming regression for estimates from each method versus the global mean (i.e., the mean of exposures at a given location across all methods) and for each method pair. Finally, we derived two types of intraclass correlation coefficients (ICCs): two-way single-rater absolute agreement and two-way single-rater consistency (i.e. rank order) agreement (Koo and Li 2016). All comparisons were completed in SAS (Version 9.4) and R (Version 4.1.0). The R “deming” package was used for the Deming regression and the “IRR” package was used to calculate ICCs.

2.4.4. Associations of Air Pollution with Cognitive and Neuroimaging Outcomes

Generally, we follow the approach of Power et al., 2018, where we initially performed all analyses separately by study site and meta-analyzed site-specific findings using a random-effects model, to quantify associations between PM2.5 and cognitive/neuroimaging outcomes (Power et al. 2018). We used adjusted linear and logistic regression to quantify the association between method-specific PM2.5 (per 1 μg/m3 unit difference in PM2.5) and cognitive and neuroimaging outcomes, as appropriate. Here we focus reporting on frontal lobe volumes, white matter hyperintensity volumes (WMH), and global cognition scores, although we consider robustness of findings across additional outcomes in sensitivity analyses. All models were adjusted for race, sex, education level, age at Visit 5, smoking status at Visit 4, apolipoprotein E4 (APOE e4) allele status, and area-level socioeconomic status at Visit 4 (see Appendix). MRI analyses were further adjusted for estimated intracranial volumes (Jack et al. 2014) and were weighted using study-provided sampling weights derived to account for the known sampling strategy and refusals associated with selection into the MRI sub-sample.

The original decision to perform all analyses separately by study site and meta-analyze site-specific findings using a random-effects model (Power et al. 2018), replicated here, reflects the presence of large among-site variation in PM2.5 relative to within-site variation in PM2.5, as well as known regional differences in PM2.5 composition. To better understand the potential impact of spatial confounding that could occur through naïve use of the combined data in this extreme case, where air pollution and outcome distributions vary substantially by site, we further completed two combined-site regression analyses; one adjusted for a combined race-site variable and the other included race as a covariate but omitted adjustment for study site. We created forest plots to allow visual comparison of effect estimates and precision across exposure estimation methods.

Sensitivity analyses considered additional outcomes and associations when re-scaling exposures by method-specific standard deviations. We report 95% confidence intervals (CIs). Exposure-outcome analyses were conducted in SAS (Version 9.4), R (Version 4.1.0) and Stata/MP (Version 15.1).

3. RESULTS

Participant characteristics varied substantially by site (Appendix Tables A1 and A2). Of note, on average, those in the MN and NC sites were more educated and had higher neighborhood socioeconomic status. On average, those in MN had the highest cognitive test scores, largest brain volumes, and least cerebrovascular pathology, while those in MS had the lowest cognitive test scores, smallest brain volumes, and most cerebrovascular pathology. Air pollution levels also varied by site, regardless of estimation method; generally, MN participants had the lowest exposure levels while NC and MD participants generally had the highest levels (Appendix Tables A3 and A4).

3.1. Comparison of Exposure Estimates Across Estimation Methods

Relative and absolute agreement between PM2.5 estimation methods were broadly similar across the NCS and MRI datasets; therefore, we report results from the NCS dataset only. When considered overall, distributions of PM2.5 were reasonably similar across methods, reflecting similar ordering of air pollution levels predicted across sites; however, predicted exposure levels varied substantially within site (Figure 2, Appendix Tables A3 and A4). Several of the methods had narrow within-site distributions in one or more sites (i.e., M02 (National Log-Normal Ordinary Kriging), M09 (CMAQ NEI/CMAQ- EDGAR + Observation Data Fusing, 36km), M10 (nearest neighbor), and M11 (inverse distance weighted mean). M01 (GAMM with LUR) produced substantially higher concentrations in MN and MD, as well as substantially lower concentrations in NC and MS, compared to the other methods.

Figure 2.

Figure 2.

Open in a new tab

Distributions of Predicted PM2.5 Exposures (Mean of 2000–2007) for Each Exposure Estimation Method by ARIC Study Site.

M01, GAMM with LUR Point estimates; M02, National Log-Normal Ordinary Kriging Point estimates; M03, National Universal Kriging with LUR & PLSR (National Historical PM2.5 Historical Model) Point estimates; M04, Regionalized Universal Kriging with LUR & PLSR (National Spatiotemporal PM2.5 Model) Point estimates; M05, Satellite-Based, using a Chemical Transport Model and Ground-Based Observations 0.01° × 0.01° grid or ~1.1km at the equator; M06, CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 1.33km grid; M07, CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 4km grid; M08, CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 12km grid; M09, CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 36km grid; M10, Nearest Neighbor Point estimates; M11, Inverse Distance Weighted Mean, Point estimates

On an absolute basis, pairwise mean bias was within ±1μg/m3 (and generally within ±0.5μg/m3), excluding comparisons against M01 (GAMM with LUR) or M05 (Satellite-Based, using a Chemical Transport Model and Ground-Based Observations) overall and within sites (Appendix Figure A1). Patterns for mean error were similar (Appendix Figure A2). On a relative basis, pairwise R2 and scatterplots with Deming regression illustrated that all models except M01 (GAMM with LUR) agreed on which sites had higher or lower air pollution exposures (Appendix Figure A3); however, the ordering of participant-level within-site exposures varied substantially by method (Figure 3, Appendix Figures A4 to A7). ICCs for the combined-site data were moderate (two-way single-rater absolute agreement: 0.65, 95% CI: 0.62, 0.68; two-way single-rater consistency agreement: 0.69, 95% CI: 0.68, 0.70, Appendix Table A5). However, ICCs within site were poor, ranging from 0.03 to 0.06 for two-way single-rater absolute agreement and 0.26 to 0.50 for two-way single-rater consistency (i.e. rank-order) agreement.

Figure 3.

Figure 3.

Open in a new tab

Pairwise, inter-method spatial R2s of Predicted PM2.5 Exposures (Mean of 2000–2007) for Each Pair of Exposure Estimation Methods by ARIC Study Site

M01, GAMM with LUR Point estimates; M02, National Log-Normal Ordinary Kriging Point estimates; M03, National Universal Kriging with LUR & PLSR (National Historical PM2.5 Historical Model) Point estimates; M04, Regionalized Universal Kriging with LUR & PLSR (National Spatiotemporal PM2.5 Model) Point estimates; M05, Satellite-Based, using a Chemical Transport Model and Ground-Based Observations 0.01° × 0.01° grid or ~1.1km at the equator; M06, CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 1.33km grid; M07, CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 4km grid; M08, CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 12km grid; M09, CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 36km grid; M10, Nearest Neighbor Point estimates; M11, Inverse Distance Weighted Mean, Point estimates

3.2. Associations of Air Pollution with Cognitive and Neuroimaging Outcomes

The was little support for associations between method-specific PM2.5 concentrations and our primary outcomes (frontal lobe volume, WMH volume, and our global cognition summary measure) across all methods and sites (Figure 4). Precision varied substantially across methods for all outcomes and locations. As expected, methods that produced limited variation in PM2.5 concentrations within study site (e.g., M02 (National Lognormal Ordinary Kriging), M11 (Inverse Distance Weighted Mean)) often had large confidence intervals. Overall patterns were similar when considering other outcomes and when scaling the exposures by the method-specific standard deviation.

Figure 4.

Figure 4.

Open in a new tab

Association between a 1 μg/m3 Difference in Average PM2.5 (2000–2007) and Cognitive or Neuroimaging Outcomes by Exposure Estimation Method and ARIC Study Site Effect estimates are reported as difference in mm3 for WMH volume and difference in standard deviation units for frontal lobe volume and global score.

M01, GAMM with LUR Point estimates; M02, National Log-Normal Ordinary Kriging Point estimates; M03, National Universal Kriging with LUR & PLSR (National Historical PM2.5 Historical Model) Point estimates; M04, Regionalized Universal Kriging with LUR & PLSR (National Spatiotemporal PM2.5 Model) Point estimates; M05, Satellite-Based, using a Chemical Transport Model and Ground-Based Observations 0.01° × 0.01° grid or ~1.1km at the equator; M06, CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 1.33km grid; M07, CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 4km grid; M08, CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 12km grid; M09, CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 36km grid; M10, Nearest Neighbor Point estimates; M11, Inverse Distance Weighted Mean, Point estimates

With a few exceptions, race-site-adjusted results using data from all sites combined were similar to the results obtained from meta-analysis of site-specific findings (Figure 5). However, omission of adjustment for site created divergent findings, with notably different point estimates and (misleadingly) narrower confidence intervals. Without adjustment for site, increasing PM2.5 was associated with higher WMH volumes and lower overall cognitive status for nearly all PM2.5 exposure methods (Figure 5).

Figure 5.

Figure 5.

Open in a new tab

Association per 1 μg/m3 Difference in PM2.5 (2000–2007) and Cognitive or Neuroimaging Outcomes by Exposure Estimation Method and Analytic Approach to Combined-Site Analyses Effect estimates are reported as difference in mm3 for WMH volume and difference in standard deviation units for frontal lobe volume and global score.

M01, GAMM with LUR Point estimates; M02, National Log-Normal Ordinary Kriging Point estimates; M03, National Universal Kriging with LUR & PLSR (National Historical PM2.5 Historical Model) Point estimates; M04, Regionalized Universal Kriging with LUR & PLSR (National Spatiotemporal PM2.5 Model) Point estimates; M05, Satellite-Based, using a Chemical Transport Model and Ground-Based Observations 0.01° × 0.01° grid or ~1.1km at the equator; M06, CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 1.33km grid; M07, CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 4km grid; M08, CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 12km grid; M09, CMAQ-NEI/CMAQ-EDGAR + Observational Data Fusing 36km grid; M10, Nearest Neighbor Point estimates; M11, Inverse Distance Weighted Mean, Point estimates

4. DISCUSSION

All considered PM2.5 exposure estimation methods similarly represented between-site variation in long-term average PM2.5 exposures, with the exception of M01 (GAMM with LUR), which produced substantially higher concentrations in MN and MD but substantially lower concentrations in NC and MS. However, within-site, predicted exposure concentrations and exposure variation varied substantially by method. Several methods (e.g., M02 (National Log-Normal Ordinary Kriging), M09 (CMAQ NEI/CMAQ- EDGAR + Observation Data Fusing, 36km), M10 (nearest neighbor), M11 (inverse distance-weighted mean) yielded estimates with little to no variation within site, as expected given coarse resolution or reliance only on monitor data, and thus are not an appropriate choice when the goal is to capture exposure variability in small geographic areas. Among methods that yield within-site variation in exposure estimates, limited within-site correlation across estimates derived from different methods suggests that at least some subset of the methods we considered do not accurately capture small-scale spatial variation within the study site areas (which are areas of approximately 10,000 km2 or 3,860 square miles, or approximately the combined area of Delaware and Rhode Island or the area of Cyprus). Further work to understand which methods are more accurately capturing PM2.5 variation across smaller geographic areas is clearly needed.

Several other studies have compared long-term PM2.5 concentration estimates produced using different methods, and generally report reasonable relative agreement across methods. In Europe, pairwise spatial correlations ranged from 0.22 to 0.86 (R2 0.05 to 0.74) for annual-average PM2.5 between LUR, dispersion, and hybrid models depending on the spatial area, location, and calendar year(s) (Bauwelinck et al. 2022; de Hoogh et al. 2014; Klompmaker et al. 2021; Wang et al. 2015). Likewise, U.S.-based studies have reported pairwise spatial correlations ranging from 0.54 to 0.99 (R2 0.29 to 0.98) across PM2.5 estimation methods (including BME, CTM, LUR with BME kriging, and satellite-based approaches) for multi-year averages in the conterminous U.S. (Jerrett et al. 2017) and 0.60 to 0.88 (R2 0.36 to 0.77) for PM2.5 annual averages derived from monitor data, CTM, and satellite-based approaches in North Carolina (McGuinn et al. 2017). These findings are similar to our combined-site results and suggest that our finding that many methods produce similar broad spatial patterns is generalizable across many geographies. However, our study adds to the literature by illustrating that similar across-site or broad patterns do not guarantee similar within-site or local patterns. For example, one study comparing seven publicly-available products for PM2.5 concentrations across the monitor-rich state of New York (~141,000 km2 or 55,000 mi2) reported pairwise spatial correlation coefficients between methods for multi-year averages of 0.65 to 0.90 (R2: 0.42 to 0.81), including a pairwise spatial correlation of 0.83 (R2 : 0.69) between an IDW and the Satellite-based method used in our analyses (Jin et al. 2019). While this aligns with our overall findings (the overall pairwise R2 between our IDW method (M11) and the Satellite-based using a CTM and Ground Observations methods (M05) was 0.77 in our analyses), within-site R2 for M05 and M11 ranged from 0.00 (NC) to 0.32 (MN) in our study, demonstrating that reported statistics using data representing exposures from a broad spatial area may mask poor within-area agreement, particularly if there are factors impacting local model performance. Supporting our own findings of more limited agreement at local scales, another study which compared 2011 annual average PM2.5 estimates across the continental U.S. derived using nine different PM2.5 exposure methods -- including instances of CTMs, inverse distance weighting, Bayesian downscaling, machine learning models, hybrid models, and a satellite-based approach incorporating a CTM and ground observations -- reported that methods generally agreed when considering nationwide patterns of exposure, but differed more when considering exposures within urban areas using finer (1km grid) spatial resolutions (Kelly et al. 2021).

In this study, associations between PM2.5 and cognitive or neuroimaging outcomes were mostly consistent with no effect, regardless of method; for some methods, limited variability in estimated PM2.5 within-site also limited precision of estimates. The absence of evidence supporting an effect of PM2.5 on our outcomes makes it difficult to draw strong conclusions about the influence of exposure assessment method choice on health effect estimates based on our analyses. A few other studies have reported associations between PM2.5 and cognition or dementia using estimates from more than one exposure estimation method. In a study of middle aged and older adults in Metro Vancouver, elevated associations between long-term PM2.5 exposures estimated using a land use regression (LUR) model and incident non-Alzheimer’s disease dementia, multiple sclerosis, and Parkinson’s disease were not replicated in sensitivity analyses using national satellite-based PM2.5 estimates (Yuchi et al. 2020). In a study of older adults in the German Ruhr area, long-term residential PM2.5 concentrations were associated with mild cognitive impairment (MCI) when using exposure estimates from a LUR model based on the European Study of Cohorts for Air Pollution Effects procedures (ESCAPE-LUR) but not when using the European Air Pollution Dispersion and Chemistry Transport Model (EURAD-CTM) (Tzivian et al. 2016). Analyses in the U.K. Biobank suggested some differences in the association between PM2.5 with baseline cognition and cognitive decline when PM2.5 was estimated using two different exposure methods, ESCAPE-LUR and a database provided by the U.K. Government Department for Environment, Food and Rural Affairs (DEFRA); however, regardless of exposure assessment method, no strong associations were found, resulting in similar conclusions regardless of method (Cullen et al. 2018). Collectively, these studies suggest that variation in exposure estimation method may impact findings in studies of air pollution and cognitive or brain health, particularly when participants are recruited from smaller geographic areas.

Although the literature is limited, PM2.5 estimation method effects may be more pronounced for associations of long-term PM2.5 exposures with cognitive and brain health outcomes relative to other health outcomes. For example, a Dutch study that compared associations between PM2.5 exposure and childhood lung function using PM2.5 estimates from a dispersion and LUR model found some estimates were stronger but less precise when using LUR (Wang et al. 2015). A study in North Carolina found PM2.5 estimation method did not substantially influence conclusions about the association between PM2.5 exposure and cardiovascular outcomes (McGuinn et al. 2017). Similarly, 3-year (and one 6-year) average PM2.5 exposures from several estimation methods were associated with higher cardiovascular mortality in the U.S.; however, effect estimates were typically larger when using models that incorporated ground-based information compared to those that only used remote sensing (Jerrett et al. 2017). A few other studies, including two in Europe comparing the use of LUR and dispersion models (Bauwelinck et al. 2022; Klompmaker et al. 2021), a study in the state of New York comparing the use of seven publicly available approaches (Jin et al. 2019), and a study in continental U.S. (Di et al. 2017) report similar findings for PM2.5 and mortality across exposure estimation method, although some reported differences in magnitude.

PM2.5 exposures and brain health appear spatially patterned in similar ways. In our analyses, there were clear differences in both PM2.5 and cognitive or neuroimaging outcomes by ARIC study site, with participants at the MN site having the lowest PM2.5 exposures as well as highest average cognitive performance and least evidence of pathologic brain changes. This similar spatial patterning is likely to hold true more broadly given the observed patterns of ambient air pollution by socioeconomic position (SEP) (Brochu et al. 2011; Hajat et al. 2021) and the association between SEP and dementia or cognitive test performance (Cadar et al. 2018; Deckers et al. 2019). Our analyses using meta-analysis of site-specific findings and combined analyses with adjustment for race-site produced similar results. However, confounding by site was substantial enough to change study conclusions regardless of exposure estimation method in analyses that did not adjust for study site, even when they also adjusted for individual-level education and neighborhood-level SES. While we recognize that this is an extreme example, where we have four distinct sites with distinct distributions of exposures, confounders, and outcomes, it strongly suggests that unmeasured confounding by spatially patterned factors may be of particular concern in analyses of air pollution and dementia-related outcomes. Adjusting for area or conducting area-specific analyses can minimize the impact of unmeasured confounding when the unmeasured confounder varies by area; however, while it will not further bias the results, adjustment for area will not address bias due to confounding by factors that are directly associated with the exposure and outcome, independent of area (Pedersen et al. 2013). Adjusting for measures of geographic location may reduce bias by spatially-patterned unmeasured confounders that are not captured by adjusting for larger areas, like state or U.S. census region, but will also substantially increase the variance of the effect estimates due to collinearity with the exposure estimates (Paciorek 2010).

Our study has several strengths. We use data from 11 different exposure estimation methods developed by distinct groups previously used in the epidemiologic literature on PM2.5 and health, and compared agreement both across site and within site. We also explicitly consider the impact on cognitive outcomes and the influence of spatial confounding on study findings. As our goal was to understand agreement and disagreement between methods, we cannot comment on the accuracy of individual methods. While cross-validated statistics comparing estimates to measured air pollution at monitoring sites are frequently reported, the existing reporting is insufficient to provide an assessment of the relative accuracy of models given the heterogeneity in the choice of validation statistics reported, variable spatiotemporal resolution of models and data used in validation exercises, and differences in geographic areas covered. Additionally, cross-validation using existing monitoring sites may not adequately reflect agreement at residences or other prediction locations, so comparisons of modelled data against that from existing monitoring stations may not be an ideal validation approach. However, a better understanding of agreement in air pollution or health effects estimates across approaches should help advance the field. For example, our investigation clearly illustrates the need to better understand relative model performance within smaller areas rather than at the national or regional levels, as well as the need to better understand the potentially unique impact of spatial confounding on studies of air pollution and cognition or related outcomes.

Here, we only considered the average air pollution exposure between 2000 and 2007 in four areas in the eastern United States for a subset of methods available to or used by the epidemiologic community. It is possible that our results would differ if we focused on another part of the country or the world, on different time periods, or on different sets of methods. We did not see strong support for an association between PM2.5 and cognitive or neuroimaging outcomes with any method, making it difficult to draw strong conclusions about the impact of method on heterogeneity in findings in other contexts. Lack of strong support for an association between PM2.5 may reflect the limitations of our data rather than absence of an effect. For example, we were unable to evaluate the potential for a non-linear association given limited within-site sample size and within-site exposure variation. Similarly, as we focused on ambient air pollution exposures given its relevance to policy, ambient air pollution exposures arguably misclassify personal exposure, which may mask an etiologic effect of air pollution on health. Finally, we did not consider other types of air pollution in our study. These methods may or may not perform similarly or lead to different impacts on effect estimates for other pollutants.

5. CONCLUSION

In conclusion, PM2.5 estimation methods agreed across sites, but not within site. Choice of estimation method may impact findings in epidemiologic studies when participants are concentrated in small geographic areas. Further work to understand the relative accuracy of different methods across smaller geographic areas is needed. Epidemiologic research on PM2.5 and cognition or related outcomes may also be particularly susceptible to spatial confounding. Further work to understand the potential extent and magnitude of this potential bias in other settings is also warranted.

Supplementary Material

1

HIGHLIGHTS:

  • Eleven PM2.5 exposure estimation methods agreed across, but not within, study sites.

  • Findings were generally null for PM2.5-cognitive and MRI outcome associations regardless of estimation method used.

  • Choice of estimation method is expected to impact findings when there truly is an association and participants are concentrated in small geographic areas.

  • Not accounting for geographic location may bias results due to unmeasured, spatially patterned confounders.

ACKNOWLEDGEMENTS

The authors thank the staff and participants of the ARIC study for their important contributions.

FUNDING

This work was supported by the National Institutes for Environmental Health Sciences and National Institute on Aging [grant number [R01ES029509]] to Melinda C. Power. The Atherosclerosis Risk in Communities Study is carried out as a collaborative study supported by National Heart, Lung, and Blood Institute contracts (75N92022D00001, 75N92022D00002, 75N92022D00003, 75N92022D00004, 75N92022D00005). The ARIC Neurocognitive Study is supported by U01HL096812, U01HL096814, U01HL096899, U01HL096902, and U01HL096917 from the NIH (NHLBI, NINDS, NIA and NIDCD). The authors thank the staff and participants of the ARIC study for their important contributions.

DECLARATION OF INTEREST

Melinda C. Power reports research grants from the United States (U.S.) National Institutes of Health, U.S. Department of Defense, and District of Columbia Department of Health and consulting fees from Biogen for service on the Health Climate, Healthy Lives Scientific Advisory Board.

Erin E. Bennett reports receiving consulting fees from Massachusetts General Hospital.

Katie M. Lynch reports no disclosures.

Qi Ying reports no disclosures.

Eun Sug Park reports a research grant from the U.S. National Institutes of Health.

Xiaohui Xu reports no disclosures.

Richard Smith reports no disclosures.

James Stewart reports no disclosures.

Jeff D. Yanosky reports research grants from the U.S. National Institutes of Health.

Duanping Liao reports research grants from the U.S. National Institutes of Health.

Joel D. Kaufman reports research grants from the US National Institutes of Health and Environmental Protection Agency.

Aaron van Donkelaar reports no competing financial interest disclosures.

Joel D. Kaufman reports grant funding from U.S. National Institutes of Health and Environmental Protection Agency.

Lianne Sheppard reports research grants from the U.S. National Institutes of Health and the Health Effects Institute.

Adam Szpiro reports grant funding from the U.S. National Institutes of Health and the Health Effects Institute and consulting fees from Health Effects Institute.

Eric A. Whitsel reports research grants from the US National Institutes of Health and Federal Aviation Administration.

Declaration of interests

Melinda Power reports financial support was provided by National Institutes of Health.

Melinda Power reports a relationship with National Institutes of Health that includes: funding grants.

Melinda Power reports a relationship with US Department of Defense that includes: funding grants.

Melinda Power reports a relationship with DC DEPARTMENT OF HEALTH that includes: funding grants.

Melinda Power reports a relationship with Biogen that includes: consulting or advisory.

Erin Bennett reports a relationship with Massachusetts General Hospital that includes: consulting or advisory.

Eun Sug Park reports a relationship with National Institutes of Health that includes: funding grants.

Jeff D. Yanosky reports a relationship with National Institutes of Health that includes: funding grants.

Duanping Liao reports a relationship with National Institutes of Health that includes: funding grants.

Joel D. Kaufman reports a relationship with National Institutes of Health that includes: funding grants.

Joel D. Kaufman reports a relationship with US Environmental Protection Agency that includes: funding grants.

Lianne Sheppard reports a relationship with National Institutes of Health that includes: funding grants.

Lianne Sheppard reports a relationship with Health Effects Institute that includes: funding grants.

Adam Szpiro reports a relationship with National Institutes of Health that includes: funding grants.

Adam Szpiro reports a relationship with Health Effects Institute that includes: funding grants.

Adam Szpiro reports a relationship with Health Effects Institute that includes: consulting or advisory.

Eric Whitsel reports a relationship with National Institutes of Health that includes: funding grants.

Eric Whitsel reports a relationship with US Federal Aviation Administration that includes: funding grants.

If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Abbreviations:

ARIC

Atherosclerosis Risk in Communities

MRI

magnetic resonance imaging

PM2.5

particulate matter of size <2.5 microns in diameter

EPA

U.S. Environmental Protection Agency

GIS

geographic information system

LUR

land use regression

AOD

aerosol optical depth

CTM

chemical transport model

GAMM

General Additive Mixed Model

PLSR

partial least squares regression

CMAQ

Community Multiscale Air Quality

NEI

National Emissions Inventory

EDGAR

Emission Database for Global Atmospheric Research

DSST

Digit Symbol Substitution Test

WMH

white matter hyperintensity

NCS

neurocognitive study

ICC

interclass correlation coefficients

APOE e4

apolipoprotein E4

CIs

confidence intervals

M01-M11

Method1-Method 11

IDW

Inverse distance-weighted

ESCAPE

European Study of Cohorts for Air Pollution Effects

EURAD

European Air Pollution Dispersion

DEFRA

Department for Environment, Food and Rural Affairs

SEP

socioeconomic position

SES

socioeconomic status

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

HUMAN SUBJECTS APPROVALS

The ARIC study was approved by the institutional review boards of all participating institutions, and written informed consent was obtained from participants for each study visit. Geocoding and air pollution exposure estimate linkage was conducted under approval from the UNC IRB. This analysis was determined to be not human subjects research by the George Washington University IRB.

DATA STATEMENT

ARIC data can be accessed through established cohort procedures. Please see https://aric.cscc.unc.edu/aric9/ for more information.

REFERENCES

  1. Bauwelinck M; Chen J; de Hoogh K; Katsouyanni K; Rodopoulou S; Samoli E; Andersen ZJ; Atkinson R; Casas L; Deboosere P; Demoury C; Janssen N; Klompmaker JO; Lefebvre W; Mehta AJ; Nawrot TS; Oftedal B; Renzi M; Stafoggia M; Strak M; Vandenheede H; Vanpoucke C; Van Nieuwenhuyse A; Vienneau D; Brunekreef B; Hoek G Variability in the association between long-term exposure to ambient air pollution and mortality by exposure assessment method and covariate adjustment: A census-based country-wide cohort study. Sci Total Environ 2022;804:150091. [DOI] [PubMed] [Google Scholar]
  2. Benton A, Hamsher K Multilingual Aphasia Examination edêds. Iowa City, IA: University of Iowa; 1976 [Google Scholar]
  3. Brochu PJ; Yanosky JD; Paciorek CJ; Schwartz J; Chen JT; Herrick RF; Suh HH Particulate air pollution and socioeconomic position in rural and urban areas of the Northeastern United States. Am J Public Health 2011;101 (Suppl1):S224–230 [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Cadar D; Lassale C; Davies H; Llewellyn DJ; Batty GD; Steptoe A Individual and Area-Based Socioeconomic Factors Associated With Dementia Incidence in England: Evidence From a 12-Year Follow-up in the English Longitudinal Study of Ageing. JAMA Psychiatry 2018;75:723–732 [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Cullen B; Newby D; Lee D; Lyall DM; Nevado-Holgado AJ; Evans JJ; Pell JP; Lovestone S; Cavanagh J Cross-sectional and longitudinal analyses of outdoor air pollution exposure and cognitive function in UK Biobank. Sci Rep 2018;8:12089. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. de Hoogh K; Korek M; Vienneau D; Keuken M; Kukkonen J; Nieuwenhuijsen MJ; Badaloni C; Beelen R; Bolignano A; Cesaroni G; Pradas MC; Cyrys J; Douros J; Eeftens M; Forastiere F; Forsberg B; Fuks K; Gehring U; Gryparis A; Gulliver J; Hansell AL; Hoffmann B; Johansson C; Jonkers S; Kangas L; Katsouyanni K; Künzli N; Lanki T; Memmesheimer M; Moussiopoulos N; Modig L; Pershagen G; Probst-Hensch N; Schindler C; Schikowski T; Sugiri D; Teixidó O; Tsai M-Y; Yli-Tuomi T; Brunekreef B; Hoek G; Bellander T Comparing land use regression and dispersion modelling to assess residential exposure to ambient air pollution for epidemiological studies. Environ Int 2014;73:382–392 [DOI] [PubMed] [Google Scholar]
  7. Deckers K; Cadar D; van Boxtel MPJ; Verhey FRJ; Steptoe A; Köhler S Modifiable Risk Factors Explain Socioeconomic Inequalities in Dementia Risk: Evidence from a Population-Based Prospective Cohort Study. J Alzheimers Dis 2019;71:549–557 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Di Q; Wang Y; Zanobetti A; Wang Y; Koutrakis P; Choirat C; Dominici F; Schwartz JD Air Pollution and Mortality in the Medicare Population. The New England journal of medicine 2017;376:2513–2522 [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Diao M; Holloway T; Choi S; O’Neill SM; Al-Hamdan MZ; Van Donkelaar A; Martin RV; Jin X; Fiore AM; Henze DK; Lacey F; Kinney PL; Freedman F; Larkin NK; Zou Y; Kelly JT; Vaidyanathan A Methods, availability, and applications of PM(2.5) exposure estimates derived from ground measurements, satellite, and atmospheric models. J Air Waste Manag Assoc 2019;69:1391–1414 [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Dickerson BC; Stoub TR; Shah RC; Sperling RA; Killiany RJ; Albert MS; Hyman BT; Blacker D; Detoledo-Morrell L Alzheimer-signature MRI biomarker predicts AD dementia in cognitively normal adults. Neurology 2011;76:1395–1402 [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Gross AL; Power MC; Albert MS; Deal JA; Gottesman RF; Griswold M; Wruck LM; Mosley TH Jr.; Coresh J; Sharrett AR; Bandeen-Roche K Application of Latent Variable Methods to the Study of Cognitive Decline When Tests Change over Time. Epidemiology 2015;26:878–887 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Hajat A; MacLehose RF; Rosofsky A; Walker KD; Clougherty JE Confounding by Socioeconomic Status in Epidemiological Studies of Air Pollution and Health: Challenges and Opportunities. Environmental health perspectives 2021;129:65001–65001 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Hart JE; Liao X; Hong B; Puett RC; Yanosky JD; Suh H; Kioumourtzoglou MA; Spiegelman D; Laden F The association of long-term exposure to PM2.5 on all-cause mortality in the Nurses’ Health Study and the impact of measurement-error correction. Environ Health 2015;14:38. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Health Effects Institute. State of Global Air 2020. Special Report Boston, MA: Health Effects Institute; 2020 [Google Scholar]
  15. Jack CR Jr.; Wiste HJ; Weigand SD; Rocca WA; Knopman DS; Mielke MM; Lowe VJ; Senjem ML; Gunter JL; Preboske GM; Pankratz VS; Vemuri P; Petersen RC Age-specific population frequencies of cerebral β-amyloidosis and neurodegeneration among people with normal cognitive function aged 50–89 years: a cross-sectional study. Lancet Neurol 2014;13:997–1005 [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Jerrett M; Arain A; Kanaroglou P; Beckerman B; Potoglou D; Sahsuvaroglu T; Morrison J; Giovis C A review and evaluation of intraurban air pollution exposure models. Journal of Exposure Science & Environmental Epidemiology 2005;15:185–204 [DOI] [PubMed] [Google Scholar]
  17. Jerrett M; Turner MC; Beckerman BS; Pope CA; van Donkelaar A; Martin RV; Serre M; Crouse D; Gapstur SM; Krewski D; Diver WR; Coogan PF; Thurston GD; Burnett RT Comparing the Health Effects of Ambient Particulate Matter Estimated Using Ground-Based versus Remote Sensing Exposure Estimates. Environ Health Perspect 2017;125:552–559 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Jin X; Fiore AM; Civerolo K; Bi J; Liu Y; van Donkelaar A; Martin RV; Al-Hamdan M; Zhang Y; Insaf TZ; Kioumourtzoglou M-A; He MZ; Kinney PL Comparison of multiple PM2.5 exposure products for estimating health benefits of emission controls over New York State, USA. Environmental Research Letters 2019;14:084023 [Google Scholar]
  19. Keller JP; Olives C; Kim S-Y; Sheppard L; Sampson PD; Szpiro AA; Oron AP; Lindström J; Vedal S; Kaufman JD A Unified Spatiotemporal Modeling Approach for Predicting Concentrations of Multiple Air Pollutants in the Multi-Ethnic Study of Atherosclerosis and Air Pollution. Environmental Health Perspectives 2015;123:301–309 [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Kelly JT; Jang C; Timin B; Di Q; Schwartz J; Liu Y; van Donkelaar A; Martin RV; Berrocal V; Bell ML Examining PM2.5 concentrations and exposure using multiple models. Environ Res 2021;196:110432. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Kim S-Y; Olives C; Sheppard L; Sampson PD; Larson TV; Keller JP; Kaufman JD Historical Prediction Modeling Approach for Estimating Long-Term Concentrations of PM2.5 in Cohort Studies before the 1999 Implementation of Widespread Monitoring. Environmental health perspectives 2017;125:38–46 [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Kirwa K; Szpiro AA; Sheppard L; Sampson PD; Wang M; Keller JP; Young MT; Kim SY; Larson TV; Kaufman JD Fine-Scale Air Pollution Models for Epidemiologic Research: Insights From Approaches Developed in the Multi-ethnic Study of Atherosclerosis and Air Pollution (MESA Air). Curr Environ Health Rep 2021;8:113–126 [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Klompmaker JO; Janssen N; Andersen ZJ; Atkinson R; Bauwelinck M; Chen J; de Hoogh K; Houthuijs D; Katsouyanni K; Marra M; Oftedal B; Rodopoulou S; Samoli E; Stafoggia M; Strak M; Swart W; Wesseling J; Vienneau D; Brunekreef B; Hoek G Comparison of associations between mortality and air pollution exposure estimated with a hybrid, a land-use regression and a dispersion model. Environ Int 2021;146:106306. [DOI] [PubMed] [Google Scholar]
  24. Knopman DS; Griswold ME; Lirette ST; Gottesman RF; Kantarci K; Sharrett AR; Jack CR Jr.; Graff-Radford J; Schneider ALC; Windham BG; Coker LH; Albert MS; Mosley TH Jr.; Investigators AN Vascular imaging abnormalities and cognition: mediation by cortical volume in nondemented individuals: atherosclerosis risk in communities-neurocognitive study. Stroke 2015;46:433–440 [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Knopman DS; Ryberg S A verbal memory test with high predictive accuracy for dementia of the Alzheimer type. Arch Neurol 1989;46:141–145 [DOI] [PubMed] [Google Scholar]
  26. Koo TK; Li MY A Guideline of Selecting and Reporting Intraclass Correlation Coefficients for Reliability Research. J Chiropr Med 2016;15:155–163 [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Landrigan PJ; Fuller R; Acosta NJR; Adeyi O; Arnold R; Basu NN; Balde AB; Bertollini R; Bose-O’Reilly S; Boufford JI; Breysse PN; Chiles T; Mahidol C; Coll-Seck AM; Cropper ML; Fobil J; Fuster V; Greenstone M; Haines A; Hanrahan D; Hunter D; Khare M; Krupnick A; Lanphear B; Lohani B; Martin K; Mathiasen KV; McTeer MA; Murray CJL; Ndahimananjara JD; Perera F; Potocnik J; Preker AS; Ramesh J; Rockstrom J; Salinas C; Samson LD; Sandilya K; Sly PD; Smith KR; Steiner A; Stewart RB; Suk WA; van Schayck OCP; Yadama GN; Yumkella K; Zhong M The Lancet Commission on pollution and health. Lancet 2018;391:462–512 [DOI] [PubMed] [Google Scholar]
  28. Liao D; Peuquet DJ; Duan Y; Whitsel EA; Dou J; Smith RL; Lin HM; Chen JC; Heiss G GIS approaches for the estimation of residential-level ambient PM concentrations. Environ Health Perspect 2006;114:1374–1380 [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Liao D, W. E, Wellenius G, Smith R, Szpiro A, Coull B, Yanosky J. Daily kriging to estimate large scale location specific PM10 – comparisons of measurement error options. Abstract presented at: International Society for Environmental Epidemiology (ISEE) annual meeting: Environ Health Perspect; 2013 [Google Scholar]
  30. McGuinn LA; Ward-Caviness C; Neas LM; Schneider A; Di Q; Chudnovsky A; Schwartz J; Koutrakis P; Russell AG; Garcia V; Kraus WE; Hauser ER; Cascio W; Diaz-Sanchez D; Devlin RB Fine particulate matter and cardiovascular disease: Comparison of assessment methods for long-term exposure. Environ Res 2017;159:16–23 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Paciorek CJ The importance of scale for spatial-confounding bias and precision of spatial regression estimators. Stat Sci 2010;25:107–125 [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Pedersen M; Siroux V; Pin I; Charles MA; Forhan A; Hulin A; Galineau J; Lepeule J; Giorgis-Allemand L; Sunyer J; Annesi-Maesano I; Slama R Does consideration of larger study areas yield more accurate estimates of air pollution health effects? An illustration of the bias-variance trade-off in air pollution epidemiology. Environ Int 2013;60:23–30 [DOI] [PubMed] [Google Scholar]
  33. Power MC; Lamichhane AP; Liao D; Xu X; Jack CR; Gottesman RF; Mosley T; Stewart JD; Yanosky JD; Whitsel EA The Association of Long-Term Exposure to Particulate Matter Air Pollution with Brain MRI Findings: The ARIC Study. Environmental Health Perspectives 2018;126:027009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Rawlings AM; Bandeen-Roche K; Gross AL; Gottesman RF; Coker LH; Penman AD; Sharrett AR; Mosley TH Factor structure of the ARIC-NCS Neuropsychological Battery: An evaluation of invariance across vascular factors and demographic characteristics. Psychological assessment 2016;28:1674–1683 [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Reitan RM Validity of the Trail Making Test as an Indicator of Organic Brain Damage. Perceptual and Motor Skills 1958;8:271–276 [Google Scholar]
  36. Ryan J, Lopez S Understanding psychological assessment: Perspectives on individual differences edêds. New York, NY: Kluwer Academic/Plenum Publishers; 2001 [Google Scholar]
  37. Schneider AL; Sharrett AR; Gottesman RF; Coresh J; Coker L; Wruck L; Selnes OA; Deal J; Knopman D; Mosley TH Normative data for 8 neuropsychological tests in older blacks and whites from the atherosclerosis risk in communities (ARIC) study. Alzheimer Dis Assoc Disord 2015;29:32–44 [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Schneider ALC; Selvin E; Sharrett AR; Griswold M; Coresh J; Jack CR Jr.; Knopman D; Mosley T; Gottesman RF Diabetes, Prediabetes, and Brain Volumes and Subclinical Cerebrovascular Disease on MRI: The Atherosclerosis Risk in Communities Neurocognitive Study (ARIC-NCS). Diabetes Care 2017;40:1514–1521 [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. The ARIC Investigators. The Atherosclerosis Risk in the Community (ARIC) Study: Design and Objectives American Journal of Epidemiology 1989;129:687–702 [PubMed] [Google Scholar]
  40. The ARIC Investigators. Manual 30: ARIC Analysis Manual Version 1.2 ed^eds; 2020
  41. Tzivian L; Dlugaj M; Winkler A; Hennig F; Fuks K; Sugiri D; Schikowski T; Jakobs H; Erbel R; Jöckel KH; Moebus S; Hoffmann B; Weimar C Long-term air pollution and traffic noise exposures and cognitive function:A cross-sectional analysis of the Heinz Nixdorf Recall study. J Toxicol Environ Health A 2016;79:1057–1069 [DOI] [PubMed] [Google Scholar]
  42. U.S. EPA. Integrated science assessment for particulate matter. Washington, DC: U.S. Environmental Protection Agency; 2019 [PubMed] [Google Scholar]
  43. U.S. EPA. Integrated Science Assessment (ISA) for Particulate Matter. Washington, DC: U.S. Environmental Protection Agency; 2019 [PubMed] [Google Scholar]
  44. Van Donkelaar A; Martin RV; Li C; Burnett RT Regional Estimates of Chemical Composition of Fine Particulate Matter Using a Combined Geoscience-Statistical Method with Information from Satellites, Models, and Monitors. Environmental Science & Technology 2019;53:2595–2611 [DOI] [PubMed] [Google Scholar]
  45. Wang M; Gehring U; Hoek G; Keuken M; Jonkers S; Beelen R; Eeftens M; Postma DS; Brunekreef B Air Pollution and Lung Function in Dutch Children: A Comparison of Exposure Estimates and Associations Based on Land Use Regression and Dispersion Exposure Modeling Approaches. Environ Health Perspect 2015;123:847–851 [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Wechsler D Wechsler Memory Scale-Revised. San Antonio, TX: Psychological Corporation; 1987 [Google Scholar]
  47. Weuve J; Bennett EE; Ranker L; Gianattasio KZ; Pedde M; Adar SD; Yanosky JD; Power MC Exposure to Air Pollution in Relation to Risk of Dementia and Related Outcomes: An Updated Systematic Review of the Epidemiological Literature. Environ Health Perspect 2021;129:96001. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Whitsel EA; Quibrera PM; Smith RL; Catellier DJ; Liao D; Henley AC; Heiss G Accuracy of commercial geocoding: assessment and implications. Epidemiol Perspect Innov 2006;3:8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Whitsel EA; Rose KM; Wood JL; Henley AC; Liao D; Heiss G Accuracy and repeatability of commercial geocoding. American journal of epidemiology 2004;160:1023–1029 [DOI] [PubMed] [Google Scholar]
  50. Williams BW; Mack W; Henderson VW Boston Naming Test in Alzheimer’s disease. Neuropsychologia 1989;27:1073–1079 [DOI] [PubMed] [Google Scholar]
  51. World Health Assembly 71. Health, environment and climate change: road map for an enhanced global response to the adverse health effects of air pollution: report by the Director-General. World Health Organization; 2018 [Google Scholar]
  52. Wright JD; Folsom AR; Coresh J; Sharrett AR; Couper D; Wagenknecht LE; Mosley TH Jr.; Ballantyne CM; Boerwinkle EA; Rosamond WD; Heiss G The ARIC (Atherosclerosis Risk In Communities) Study: JACC Focus Seminar 3/8. J Am Coll Cardiol 2021;77:2939–2959 [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Xie X; Semanjski I; Gautama S; Tsiligianni E; Deligiannis N; Rajan R; Pasveer F; Philips W A Review of Urban Air Pollution Monitoring and Exposure Assessment Methods. ISPRS International Journal of Geo-Information 2017;6:389 [Google Scholar]
  54. Yanosky JD; Paciorek CJ; Laden F; Hart JE; Puett RC; Liao D; Suh HH Spatio-temporal modeling of particulate air pollution in the conterminous United States using geographic and meteorological predictors. Environmental Health 2014;13:63. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Yuchi W; Sbihi H; Davies H; Tamburic L; Brauer M Road proximity, air pollution, noise, green space and neurologic disease incidence: a population-based cohort study. Environmental Health 2020;19:8. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1
NIHMS1998659-supplement-1.pdf (5.9MB, pdf)

Data Availability Statement

ARIC data can be accessed through established cohort procedures. Please see https://aric.cscc.unc.edu/aric9/ for more information.

RESOURCES