Introduction

A growing body of epidemiological evidence supports late-life exposure to ambient air pollutants as an important modifiable risk factor for dementia [1]. These human data converge with neurotoxicological studies that demonstrate increased early markers of neurodegenerative disease (accumulation of amyloid-β and phosphorylation of tau), changes in hippocampal neuronal morphology, and increased cognitive deficits in animals with inhaled exposures to particles [2–8]. Neuroimaging studies in humans have also reported associations between increased fine particulate matter (PM_2.5; aerodynamic diameter <2.5 μm) and nitrogen dioxide (NO₂) and smaller brain volumes in gray matter [9–16], including areas vulnerable to Alzheimer disease (AD) neuropathologies [17,18]. Despite the suggestive evidence for a possible continuum of air pollution neurotoxicity on brain aging processes [19], the reported associations between exposures and cognitive decline have been mixed [20].

Since National Ambient Air Quality Standards were first put into effect 50 years ago, significant reductions in air pollution levels have been seen throughout the US [21]. During the period of 2000 to 2010, annual averages of PM_2.5 and NO₂ decreased by 27% and 35%, respectively [22]. Long-term reduction in the ambient levels of these air pollutants have been linked with increased life expectancy [23], reduced mortality [24], and improved respiratory health (lung function growth, decreased bronchitic symptoms, and lower asthma incidence) [25–27].

Decreasing air pollution levels across the nation provide the ideal environmental context for a quasi-experimental approach [28] to studying the potential benefits of improved air quality (AQ) on maintaining brain health of older people. To the best of our knowledge, no previous studies have explored whether AQ improvement may be associated with cognitive function. We leveraged a nationwide cohort of community-dwelling older women with individual-level air pollution exposure estimates (1996 to 2012) and annual assessments of late-life cognitive function (2008 to 2018). We hypothesized that improved AQ, as indicated by reductions in PM_2.5 and NO₂, was associated with slower rate of cognitive decline in older women. We further explored whether the putative associations might differ by age, region, education, Apolipoprotein E (ApoE) e4 genotype, and cardiovascular risk factors.

Methods

Estimation of air pollution exposure

Participants’ residential addresses were prospectively collected at each WHI assessment since its inception in 1993, updated at least biannually, and then geocoded using standardized procedures [33]. The exact date of address change was used in analyses when available; otherwise, the date when the change in residence was ascertained was used. We used validated regionalized national universal kriging models with partial least squares regression of geographic covariates and US Environmental Protection Agency (EPA) monitoring data to estimate annual mean concentrations of PM_2.5 (in μg/m³) and NO₂ (in parts per billion [ppb]; a proxy of traffic-related air pollutants) at each of these residential addresses. Over 300 geographic covariates covering land use, vegetative index, proximity to features, etc., were used in the historical models for pre-1999 PM_2.5 estimation or in the national models for post-1999 PM_2.5 estimation (average cross-validation R² = 0.88) [34,35]. Over 400 geographic covariates representing proximity and buffer measures as well as satellite-derived NO₂ data were used to estimate NO₂ (average cross-validation R² = 0.85) [36]. These annual estimates were then aggregated to the 3-year average at the WHIMS-ECHO enrollment (referred to as recent exposure) and the corresponding 3-year average 10 years earlier (referred to as remote exposure), accounting for residential mobility within each 3-year time window. The individual-level measure of long-term AQ improvement over the 10-year period was defined as the reduction from remote to recent exposures (Fig 1). We focused on AQ improvement over the 10 years prior to recruitment in order to avoid methodological concerns on the temporality between the defined period of AQ improvement and the concurrently observed slower cognitive decline during the follow-up.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flowchart of study population and demonstration of study timeline.

(A) Flowchart of study population. (B) Demonstration of study timeline.^{a a}The exposure time windows may vary slightly depending on each individual’s WHIMS-ECHO enrollment time. CVLT, California Verbal Learning Test; TICSm, modified Telephone Interview for Cognitive Status; WHI-CT (HT), Women’s Health Initiative-Clinical Trial (Hormone Therapy); WHIMS-ECHO, Women’s Health Initiative Memory Study-Epidemiology of Cognitive Health Outcomes.

https://doi.org/10.1371/journal.pmed.1003893.g001

Results

We excluded 346 women with prevalent dementia at WHIMS-ECHO enrollment or without follow-up visits, resulting in a sample of 2,534 women with at least 2 modified TICSm measures of general cognitive status (Fig 1). For the analyses on episodic memory decline, we further excluded 587 women without repeated measures of episodic memory assessed by CVLT. For both sets of analyses, we also excluded women with missing data on AQ measures or relevant covariates. This resulted in a final analytic sample of 2,232 women for the analyses on general cognitive status assessed by TICSm, a subset of which (n = 1,721) was used for the analyses on episodic memory assessed by CVLT (Fig 1).

Compared to the women excluded due to missing AQ measures or relevant covariates (Fig 1), women included in our analyses were younger with higher socioeconomic status and more likely to reside in the Northeast, self-identify as non-Hispanic White, and drink alcohol and have ApoE e4 genotype (Table 1). Compared to those excluded due to not having repeated CVLT measures (Fig 1), women with repeated CVLT measures were more likely to be younger, have higher education and income, and currently drink alcohol, but less likely to report having hypertension and carry the ApoE e4 genotype (S1 Table). Mean AQ improved significantly with reduced ambient levels for both PM_2.5 (Mean ± SD: 13.3 ± 2.7 to 10.6 ± 2.0 μg/m³; p < 0.001) and NO₂ (15.7 ± 7.2 to 10.4 ± 4.9 ppb; p < 0.001) over the 10 years before WHIMS-ECHO enrollment. Women residing in locations with initially high ambient air pollutants tended to experience greater AQ improvement for both PM_2.5 (correlation = 0.67; p < 0.001) and NO₂ (correlation = 0.80; p < 0.001) (S2 Table). Overall, non-Hispanic White women experienced less AQ improvement, while women who were older than 80, reported higher income, or resided in the Northeast and West experienced greater reductions in ambient PM_2.5 and NO₂, as compared to their counterparts (Table 2). Women with ApoE e4 genotype experienced less reduction in ambient PM_2.5, while nondrinkers experienced less reduction in ambient NO₂, as compared to their counterparts (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Distribution of population characteristics in the WHIMS-ECHO cohort, stratified by women included versus excluded.

https://doi.org/10.1371/journal.pmed.1003893.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Distribution of AQ measures by population characteristics in the WHIMS-ECHO cohort, 1998 to 2018.

https://doi.org/10.1371/journal.pmed.1003893.t002

During a median 6.2 (interquartile range [IQR] = 5) years of follow-up, women had a median 7 (IQR = 5) interviews for TICSm tests and 6 (IQR = 3) interviews for CVLT tests. The means of the cognitive test scores did not differ much over the visits (S3 Table), while significant cognitive declines (TICSm slope = −0.42/year; 95% CI: −0.44, −0.40; CVLT slope = −0.59/year; 95% CI: −0.64, −0.54) were observed (Fig 2). Non-Hispanic Black women or women older than 80 had lower mean cognitive scores at baseline for both TICSm and CVLT (S4 Table). Women residing in the Midwest and who had higher education or income, were currently employed, currently drinking alcohol, did not have hypertension, and did not carry the ApoE e4 genotype had higher mean cognitive scores on both TICSm and CVLT at baseline, compared to their counterparts (S4 Table). The distributions of cognitive scores at the last visit were similar across these population characteristics, except that some differences were no longer significant (S4 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Estimated cognitive trajectory over time with different levels of AQ improvement in WHIMS-ECHO cohort.

(A) Associations on general cognitive ability decline (N = 2,232). (B) Associations on episodic memory decline (N = 1,721). Estimated TICSm score (panel A) or CVLT score (panel B) change over time for low (25th percentile), median, and high (75th percentile) level of AQ improvement in PM_2.5 or NO₂ in the WHIMS-ECHO cohort. The estimated TICSm scores or CVLT scores were calculated using parameter estimates derived from Model III of Table 3, which were adjusted for spatial random effect, WHIMS-ECHO enrollment year, age, follow-up year, age interaction with follow-up year, time-varying propensity scores, demographic variables (geographic region and race/ethnicity), socioeconomic factors (education, income, and employment status) and neighborhood characteristics, lifestyle factors (smoking, drinking, and physical activities), prior hormone use, hormone therapy assignment, cardiovascular risk factors (hypertension, diabetes, and hypercholesterolemia), depression, BMI, and CVD histories. AQ, air quality; BMI, body mass index; CVD, cardiovascular disease; CVLT, California Verbal Learning Test; NO₂, nitrogen dioxide; PM_2.5, fine particulate matter; TICSm, modified Telephone Interview for Cognitive Status; WHIMS-ECHO, Women’s Health Initiative Memory Study-Epidemiology of Cognitive Health Outcomes.

https://doi.org/10.1371/journal.pmed.1003893.g002

Residing in locations with greater AQ improvement was associated with slower rates of decline in both general cognitive status and episodic memory (Table 3). Based on fully adjusted models (Table 3, Model III), each IQR increment of improved AQ in PM_2.5 and NO₂ (IQR_PM2.5 = 1.79 μg/m³ for both analytic samples; IQR_NO2 = 3.92 ppb for TICSm and 3.97 ppb for CVLT analyses) was associated with diminished cognitive declines over time by 0.026 to 0.034/year in TICSm (PM_2.5: (β = 0.026/year, 95% CI: 0.001, 0.05; NO₂: β = 0.034/year; 95% CI: 0.01, 0.06) and by 0.060 to 0.070/year in CVLT (PM_2.5: β = 0.070/year, 95% CI: 0.02, 0.12; NO₂: β = 0.060/year; 95% CI: 0.005, 0.12). These putative benefits suggested by the respective associations with TICSm and CVLT were equivalent to slower cognitive declines in women who were 0.9 to 1.2 years and 1.4 to 1.6 years younger at WHIMS-ECHO enrollment.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Summary of the associations between AQ improvement and cognitive declines in the WHIMS-ECHO cohort.

https://doi.org/10.1371/journal.pmed.1003893.t003

The associations between greater AQ improvement and slower cognitive declines remained robust in sensitivity analyses adjusting for potential confounders updated prior to the WHIMS-ECHO enrollment or temporal changes in relevant covariates from WHI inception (S5 Table). Using multiple imputation to include women with missing air pollution and covariate data, the associations were slightly attenuated, but remained statistically significant (S6 Table). The associations between greater AQ improvement and slower cognitive declines were strengthened in models further adjusting for recent or remote exposures (S7 Table). The results were largely unchanged after excluding prevalent or incident stroke cases (S8 Table). Excluding incident dementia cases (n = 398) resulted in a 41% to 81% reduction in the strength of the associations with slower declines in general cognitive status, which were no longer significant. For episodic memory decline, there was a 17% to 22% reduction in effect estimates, which remained statistically significant (S8 Table).

We did not find significant associations between AQ improvement and quadratic change in cognitive function (all p-values > 0.10, S9 Table). With the quadratic term of follow-up year included in models, the associations between AQ improvement and TICSm trajectory slope were similar to the estimates in the main analyses, while the estimated associations with CVLT trajectory slopes were slightly attenuated for both PM_2.5and NO₂ improvement (S10 Table). Except for some evidence supporting the quadratic term for PM_2.5 improvement on TICSm trajectory slope (p = 0.04 for the interaction between the quadratic term of PM_2.5 improvement and follow-up year, S11 Table), the overall results did not suggest nonlinear effects of AQ improvement on cognitive trajectory slopes (S11 Table).

We found no statistical evidence to suggest that the observed associations substantially differed by age, education, geographic region, ApoE e4 genotypes, or common cardiovascular risk factors after adjusting for multiple comparisons (false discovery rate corrected p-values > 0.05 for all tests, Figs 3 and 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Estimated associations^a between AQ improvement^b and cognitive ability decline^c, stratified by population characteristics.

The bars and whisker represent the regression coefficient beta and corresponding 95% CIs. ^aAssociation was represented by beta, the regression coefficient estimating the increase in TICSm score per year for each IQR increase of AQ improvement (IQR_PM2.5 = 1.79 μg/m³; IQR_NO2 = 3.92 ppb), adjusting for spatial random effect, WHIMS-ECHO enrollment year, age, follow-up year, age interaction with follow-up year, time-varying propensity scores, demographic variables (geographic region and race/ethnicity), socioeconomic factors (education, income, and employment status) and neighborhood characteristics, lifestyle factors (smoking, drinking, and physical activities), prior hormone use, hormone therapy assignment, cardiovascular risk factors (hypertension, diabetes, and hypercholesterolemia), depression, BMI, and CVD histories. ^bRecent exposures were the 3-year average exposures estimated at the WHIMS-ECHO enrollment. Remote exposures were the 3-year average exposures estimated 10 years before the WHIMS-ECHO enrollment. AQ improvement was defined by reduction from remote to recent exposures over the 10-year period. ^cp-Value was calculated using Wald t test for the interaction between AQ improvement and each subgroup unadjusted for multiple comparison. After controlling for multiple comparison using Benjamini–Hochberg approach, false discovery rate corrected p-values > 0.05 for all interaction tests. ApoE, Apolipoprotein E; AQ, air quality; BMI, body mass index; CVD, cardiovascular disease; GED, general educational development; IQR, interquartile range; NO₂, nitrogen dioxide; PM_2.5, fine particulate matter; ppb, parts per billion; TICSm, modified Telephone Interview for Cognitive Status; WHIMS-ECHO, Women’s Health Initiative Memory Study-Epidemiology of Cognitive Health Outcomes.

https://doi.org/10.1371/journal.pmed.1003893.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Estimated associations^a between AQ improvement^b and episodic memory decline^c, stratified by population characteristics.

The bars and whisker represent the regression coefficient beta and corresponding 95% CIs. ^aAssociation was represented by beta, the regression coefficient estimating the increase in CVLT score per year for each IQR increase of AQ improvement (IQR_PM2.5 = 1.79 μg/m³; IQR_NO2 = 3.97 ppb), adjusting for spatial random effect, WHIMS-ECHO enrollment year, age, follow-up year, age interaction with follow-up year, time-varying propensity scores, demographic variables (geographic region and race/ethnicity), socioeconomic factors (education, income, and employment status) and neighborhood characteristics, lifestyle factors (smoking, drinking, and physical activities), prior hormone use, hormone therapy assignment, cardiovascular risk factors (hypertension, diabetes, and hypercholesterolemia), depression, BMI, and CVD histories. ^bRecent exposures were the 3-year average exposures estimated at the WHIMS-ECHO enrollment. Remote exposures were the 3-year average exposures estimated 10 years before the WHIMS-ECHO enrollment. AQ improvement was defined by reduction from remote to recent exposures over the 10-year period. ^cp-Value was calculated using Wald t test for the interaction between AQ improvement and each subgroup unadjusted for multiple comparison. After controlling for multiple comparison using Benjamini–Hochberg approach, false discovery rate corrected p-values > 0.05 for all interaction tests. ApoE, Apolipoprotein E; AQ, air quality; BMI, body mass index; CVD, cardiovascular disease; CVLT, California Verbal Learning Test; GED, general educational development; IQR, interquartile range; NO₂, nitrogen dioxide; PM_2.5, fine particulate matter; WHIMS-ECHO, Women’s Health Initiative Memory Study-Epidemiology of Cognitive Health Outcomes.

https://doi.org/10.1371/journal.pmed.1003893.g004

Discussion

In this geographically diverse cohort of community-dwelling older women with up to 20 years of follow-up, we found that living in locations with 10-year improvements in ambient AQ in late life was associated with slower cognitive declines. The estimated associations, equivalent to slower declines in general cognitive status observed in women 0.9 to 1.2 years younger or in episodic memory observed in women 1.4 to 1.6 years younger, remained after adjusting for sociodemographics (age; geographic region; race/ethnicity; education; income; employment status; and neighborhood socioeconomic characteristics), lifestyle factors (smoking; alcohol; and physical activity), or clinical characteristics (BMI; depressive symptoms; diabetes; hypercholesterolemia; hypertension; CVD; and hormone therapy). The findings were robust in our sensitivity analyses accounting for various sources of spatiotemporal confounding and were largely unchanged after excluding women with prevalent or incident stroke. Excluding incident dementia cases greatly reduced the associations with slower declines in general cognitive status, but the associations with slower decline in episodic memory were modestly attenuated and remained statistically significant. There was no strong evidence that the slower declines in cognition associated with improved AQ differed by age, education, geographic region, APoE e4 genotype, or cardiovascular risk factors. To the best of our knowledge, this study provides the first epidemiologic evidence supporting the potential benefit of improved AQ on slowing cognitive aging.

To the best of our knowledge, our study adds novel epidemiologic data strengthening the evidence between late-life exposure to ambient air pollution and cognitive decline. Evidence from epidemiological and neurotoxicological studies point to a possible continuum of air pollution neurotoxicity on brain aging. Within the WHIMS suite of studies, we have found that PM_2.5 is associated with progression of gray matter atrophy in brain areas vulnerable to AD [18,41], episodic memory decline [41,42], and increased risk of clinically significant cognitive impairment [43], demonstrating this continuum in older women. In the ALzheimer and FAmilies cohort, investigators also found that higher exposure to PM was associated with lower cortical thickness in brain regions linked to AD in middle-aged men and women [17]. However, longitudinal epidemiological studies investigating the associations of cognitive declines with PM_2.5 and NO₂ have produced mixed results, including no associations [44–49], significant adverse effects [41,42,50–52], and significant positive associations (suggesting that PM_2.5 improved cognition function) [53]. In the present study, we found that improved AQ was significantly associated with slower declines in both general cognitive status and episodic memory. Although these findings alone do not prove causality, using a quasi-experimental design, our novel findings along with other supportive evidence from human studies and animal models strengthen the evidence of late-life exposure to ambient air pollution in its contribution to the progression of cognitive aging.

Our study findings have several important public health implications. First, we found that slower cognitive decline was associated with long-term reduction in ambient PM_2.5 and NO₂ levels. In the Children’s Health Study, investigators also showed that the benefits of AQ improvement in both PM_2.5 and NO₂ was associated with improved respiratory health [25–27]. These findings may imply that the observed health benefits of AQ improvement are due to the overall reduced ambient air pollution levels, rather than driven by specific control programs to mitigate either PM_2.5 or NO₂ in the US. Second, although the Clean Air Act mandates that the EPA sets AQ standards that provide a safe margin for susceptible populations [54], our analyses (Fig 3) revealed similar associations comparing subgroups of older women. This observation suggests the resulting benefit of slower cognitive decline associated with AQ improvement in late life may be universal in older women, including those already at greater risk for cognitive decline (e.g., women with high-risk cardiovascular profiles; ApoE e4 carriers). Third, the EPA’s projection that the AQ and health benefits achieved as a result of the Clean Air Act Amendments of 1990 are valued at approximately US$2 trillion in 2020 [55] was likely underestimated as it did not include the assessment of brain health, which costs the US economy US$159 to US$215 billion [56]. Fourth, our findings call for future studies to determine whether risk reductions may still be seen with improvements at lower-exposure levels. Increased dementia risks associated with low-level exposures had been reported in Sweden and Canada (range of mean PM_2.5: 7.6 to 10.4 μg/m³; range of mean NO₂: 12.1 to 16.2 ppb) [57–60]. In a subset of WHIMS participants, we previously found that exposure to late-life PM_2.5 at levels below the current EPA regulatory standard of 12 μg/m³ was associated with gray matter atrophy in brain areas vulnerable to AD [18]. Therefore, it is imperative to know whether further improvement in low-exposure air pollution levels already in compliance with the current standard still translate to slower cognitive decline. Last, our findings may provide the impetus for future research to consider how AQ improvement could potentially benefit brain development, as studies have found that higher levels of air pollution may hinder cognitive development in children [61,62].

The underlying neurobiological processes driving the observed slower cognitive declines associated with improved AQ are unclear. In our sensitivity analyses excluding women with stroke, the slower cognitive declines with improved AQ were largely unchanged, suggesting that reducing clinical stroke in late life may not make significant contributions to the observed associations. Neuroimaging studies have not shown consistent associations between air pollution and MRI-based measurements of subclinical cerebrovascular disease [9,11,63–65]. By contrast, excluding dementia cases resulted in a substantial attenuation in the estimated associations with TICSm declines, suggesting that the putative benefit of AQ improvement on slowing declines in general cognitive status may be operating by slowing the neuropathological processes near the clinical stage. On the other hand, there was a modest attenuation in the estimated associations with CVLT declines after excluding dementia cases, suggesting the observed associations of slower episodic memory decline with AQ improvement were only partly explained by the underlying neuropathological processes leading to clinical dementia. The remaining association observed among women without dementia implied the possibility of neuroprotective mechanisms underlying the putative benefits on brain health associated with improved AQ. Put together, these results point to the possibility that AQ improvement may benefit the continuum of pathological brain aging. Longitudinal studies with high-quality data on PET scan and fluid-based biomarkers are needed to better understand the underlying neuropathological processes amendable to improved AQ in late life, such as preserving brain volume or maintaining function of neural networks.

We acknowledge several study limitations. First, cognition was assessed using telephone-based interviews rather than the “gold standard” method of face-to-face administration. However, telephone-based cognitive assessment has been shown to be both reliable and valid and may improve study validity by increasing retention and data completeness [66]. Second, we estimated the ambient levels of air pollution at residential locations and detailed information on time-activity patterns were not available. However, the use of individual-level estimates of reduced exposures to ambient air pollutants is appropriate for studying the public health benefits [23,25–27] since these pollutants are regulated by the EPA. Third, the errors in predicting the air pollution estimates may have varied throughout the 10-year period and may have contributed uncertainty to our analyses. Fourth, we could not completely rule out unmeasured confounding by other environmental factors (e.g., noise and green space) with longitudinal changes that may be concurrent with improved AQ. However, noise levels have been increasing [67], and green space has been decreasing over time due to increasing urbanization [68], so it is unlikely they would contribute to the health benefits of long-term AQ improvement. Fifth, regression to the mean may be a concern due to measuring changes in AQ across only 2 time periods, which may have attenuated the estimated associations between AQ improvement and cognitive decline, as suggested by the results of our sensitivity analyses further adjusting for either recent or remote exposures (S7 Table). Sixth, because our study sample came from a nonrandom selective process (Fig 1), we could not completely rule out the possibility of selection biases. Last, our findings cannot be generalizable to men or younger women.

Our study had several major methodological strengths. First, the geographic diversity coupled with the extended longitudinal follow-up provided an ideal environmental context to explore the health benefits of AQ improvement across the US. Second, the estimated benefits were based on within-cohort comparisons, instead of cross-cohort comparisons done in previous studies [25–27], greatly reducing possible confounding stemming from between-cohort differences [69]. Third, using individual-level estimates of ambient air pollution to define improved AQ further reduces the spatial confounding arising from using county-/city-/community-level estimates as was done in previous studies [23–27]. Last, our analyses also accounted for different sources of spatiotemporal confounding, such as adjustment of WHIMS-ECHO enrollment year, including a random effect of clinic centers, and further adjusting for temporal changes in relevant covariates that may correlate with AQ improvement.

In conclusion, we found that long-term AQ improvement during late life was robustly associated with slower rates of cognitive declines among older women. The associations were similar for both PM_2.5 and NO₂ and did not differ by age, geographic region, education, or cardiovascular risk factors. Future studies are needed to improve our understanding on the underlying neuropathological processes that may be amendable by reducing exposure to late-life ambient air pollution.

Supporting information