Potential Effects of Long-Term Exposure to Air Pollution on Dementia: A Longitudinal Analysis in American Indians Aged 55 Years and Older
1. Introduction
Air pollution, an important source of premature mortality and morbidity, is unevenly distributed across the United States (US) [1]. In the US, people of color experience greater exposure to air pollution than non-Hispanic Whites [2]. This disparity was found among people at all income levels across states, urban centers, and rural areas [2]. American Indians who live on tribal lands face disproportionate health impacts from air pollution, such as ground-level ozone (O3). Ground-level ozone is not emitted directly into the air but is created from chemical reactions between volatile organic compounds (VOCs) and oxides of nitrogen (NOx). They react with sunlight to form O3 pollution that is harmful to human health. Cars, power plants, refineries, chemical plants, and other sources, including oil and gas industries that are prevalent on tribal lands, emit such pollutants [3,4]. Other air pollutants detrimental to human health include PM2.5 (particles with a diameter of 2.5 μm or less) and nitrogen dioxide (NO2), which are more commonly found in large metropolitan areas. Long-term/chronic exposure to PM2.5, O3, NOx, or nitrogen dioxide (NO2) can trigger local inflammation and oxidative stress in the brain [5,6], which may contribute to neurodegeneration processes and lead to dementia [7,8].
Epidemiological studies have investigated the association between air pollution and dementia in different populations. However, the results are inconsistent, possibly due to variations in study design, population, air pollutants, source of exposure, exposure level, length of follow-up period, outcome assessment, and/or uncontrolled confounding factors. Although most previous studies found positive associations between incident dementia, dementia hospitalization, or cognitive decline and exposure to PM2.5 [9,10,11,12,13,14,15,16], NOx/NO2 [10,11,14,17,18], or O3 [12,19,20,21], some found no or negative associations between such outcomes and PM2.5 [20,22], NOx/NO2 [18], or O3 [10,23]. Oudin et al. (2018) [24] found that the association differed according to the source of PM2.5. For example, PM2.5 from traffic exhaust was associated with higher risk of incident dementia, while PM2.5 from residential wood burning was not. In contrast, Tonne et al. (2014) [15] reported that PM2.5 from traffic was not associated with cognitive change. Moreover, some studies reported different results for different air pollutants [10,12,14,19,20,23]. For instance, Cerza et al. (2019) [19] found positive associations between O3 and dementia hospitalization, yet a negative association between NO2 and dementia hospitalization in Rome. Carey et al.’s London-based study (2018) [10] described positive associations of PM2.5 and NO2 with dementia, yet noted a negative association between O3 and dementia. In addition to these studies, Wang et al. (2022) [25] recently documented reduced dementia risk associated with air quality improvement regarding PM2.5 and NO2 among older women living in the US.
Although numerous studies have examined the association between air pollution and dementia, none have specifically investigated this problem among American Indians. American Indian communities have lower access to and use of health services than non-Hispanic Whites [26] and are disproportionately affected by a higher burden of various chronic diseases [27,28]. To address this gap, we linked individual-level data on dementia diagnosis and related health conditions extracted from the Indian Health Service (IHS) National Data Warehouse with county-level publicly available air pollution and geographic information to investigate the effects of air pollution on dementia risk in American Indians.
3. Results
Our study included 26,871 American Indians who were 55 years or older in FY2007, registered at the same project site and used IHS services at least once each year between FY2007–2009, and were dementia-free at baseline. Descriptive statistics of individual-level characteristics are shown in Table 1. Among all individuals included in this study, about 54% were in the 55–64 age group, and 59% were female. Almost half of the individuals with dementia were 75 years or older, while only 13% of those without dementia were 75+ years old. Those with dementia had higher percentages of all comorbidities investigated, including depression, diabetes, hypertension, and CVD. More individuals with dementia were enrolled in Medicaid than those without dementia (26.01% vs. 15.27%), while fewer individuals with dementia had private insurance than those without dementia (20.28% vs. 27.15%).
Figure 1 shows the yearly average air pollution levels in all 3109 counties in the contiguous US and in the 64 IHS counties where our study participants resided. The average PM2.5 levels in the 64 IHS counties were consistently lower than those in all 3109 counties in the contiguous US during the period from 1999 to 2013, while the NO2 levels in the 64 counties were like those of the US. Mean O3 levels in the 64 counties were higher than those of all the US counties in most years.
In Table 2, we illustrate adjusted associations between dementia and air pollution from a series of Cox regression models. In the models that adjusted for gender, project sites, comorbidities, health coverage, and county-level SES (Model 5), higher NO2 levels were associated with lower risk of dementia (HR = 0.91 and 95% CI: 0.83–0.99), while higher O3 levels were associated with higher risk of dementia (HR = 1.28 and 95% CI: 1.11–1.48). There was no association of dementia with PM2.5 (HR = 0.85 and 95% CI: 0.66–1.10). However, after adding co-exposure to other air pollutants to the regression model (Model 6), dementia and NO2 were not associated with each other (HR = 0.92 and 95% CI: 0.76–1.10), while O3 was still strongly associated with a higher risk of dementia (HR = 1.24, 95% CI: 1.02–1.50 per 1 standardize deviation of O3). Associations of dementia with PM2.5 were not significant in Model 6.
Table 2 also reveals that the relationships between dementia and air pollution differ by gender. While neither PM2.5 nor NO2 levels were associated with the risk of dementia in either female or male American Indians, as reflected by their wide confidence intervals, higher O3 levels were associated with higher risk of dementia among females (HR = 1.39 and 95% CI: 1.04–1.85). However, O3 was not associated with dementia risk among American Indian men (HR = 0.98 and 95% CI: 0.64–1.51).
4. Discussion
In this study, we investigated the longitudinal associations between three common air pollutants (PM2.5, O3, and NO2) and all-cause dementia in a large American Indian population aged 55 years and older who resided in the contiguous US. Overall, we found that a higher level of O3 was associated with higher risk of incident dementia. We also found negative associations between PM2.5 and incident dementia in the unadjusted model, but the associations became positive after adjusting for covariates and other air pollutants. No association between NO2 and incident dementia was found in any of the regression models except Model 5. Furthermore, we found that the association between O3 and dementia risk was substantially higher among female American Indians than that among males. Meanwhile, neither PM2.5 nor NO2 levels were associated with dementia incidence among either female or male American Indians in gender-stratified analyses.
Our results are consistent with the findings from Cleary et al. (2018) [20], who, in a geographically heterogeneous and broadly distributed sample from the National Alzheimer’s Coordinating Center, observed an increased rate of cognitive decline associated with O3, yet not with PM2.5. Similarly, Chen and Schwartz (2009) [45] reported consistent associations between a higher level of O3 and reduced cognitive performance in US adults. In addition, Cerza et al. (2019) [19] also found a positive association between O3 and dementia hospitalization, no association between NOx and dementia hospitalization, but mixed findings between PM2.5 and different dementia subtypes (i.e., a positive association with vascular dementia yet a negative association with Alzheimer’s disease).
The effects of air pollution on dementia have been previously explored in numerous studies in the US [13,14,25,46] and other countries around the world [10,11,12,18,19,21,23]. However, none of them investigated this problem among American Indians, some of whom live on or near tribal lands with high ground-level ozone levels [3,4]. To our knowledge, this is the first longitudinal study to investigate the long-term effects of air pollution on dementia incidence among American Indians. Although multiple epidemiological studies have found associations between PM2.5 and elevated dementia risk [10,11,12,13,14,23,46], the setting of our study is very different from past work, which were mostly conducted in urban settings. IHS Data Project sites are primarily located in rural areas where the PM2.5 levels are much lower than the overall US averages during 1999–2013 (Figure 1). Furthermore, some of the previous studies did not adjust for exposure to other air pollutants, such as O3. Although the three air pollutants are not highly correlated with each other, with Pearson’s correlations of 0.14 or less (Appendix B), the adjustment for other air pollutants in the regression models substantially changed some of the parameter estimates, especially for PM2.5, suggesting the other air pollutants could be important confounders for the association of each pollutant with dementia.
With respect to gender-specific analyses, in previous studies [47,48,49] that investigated gender differences in the association between air pollution and dementia in high-income countries, most of them found that the associations were stronger among females. Consistent with those studies, we also found strong association between O3 and dementia risk among American Indian women, but not in American Indian men. Based on previous studies of AI/ANs, it is not surprising to observe gender differences in the effects of air pollution on dementia risk in this population. For example, it has been reported that AI/AN males were more likely to die before age 65 than AI/AN females, and they were more likely to die due to heart disease and diabetes [50,51]. Thus, early mortality risk might be a competing risk of dementia for American Indian men. In the past, we also found diabetes was cross-sectionally associated with dementia among AI/AN women but not men [37]. Diabetes is another well-established dementia risk factor but was found to be uncorrelated with dementia among AI/AN men, which could also be partially explained by the potential competing risk associated with early mortality in this subgroup of the AI/AN population. Furthermore, substantial biological or physiological distinctions between men and women could also partially explain the gender differences in the association of air pollution and dementia risk observed in the current study.
The strengths of this study include a large, well-characterized cohort of American Indian older patients from geographically diverse sites across the US and the availability of multiple linked data sources with information on diagnosed comorbid conditions, health service utilization, and county-level socioeconomic status, which yield a wide range of potential confounders. Many of the comorbid conditions, such as CVD and diabetes, have been identified as important modifiable risk factors for dementia and thus could be potential confounders for the associations between air pollution and dementia. Some previous studies also suggest that CVD and diabetes could mediate the association between air pollution and dementia, suggesting their inclusion as model covariates might lead to over-adjustment. However, adding these chronic conditions to the models did not materially change the estimates for the parameters of interest (Table 2), indicating the potential robustness of our study findings with respect to chronic conditions. Another strength of this study is the longitudinal study design with long-term air pollution measures available at least 3 years before incident dementia cases, which provides stronger evidence for a causal link than in cross-sectional studies.
This study has several limitations. First, our air pollution exposure was only assessed at the county level because the county code is the smallest standard geographic unit available in the IHS Data Project. County-level exposure estimates may not be the most accurate proxy for long-term environmental exposure for each participant [52]. The exposure aggregation at the county level can lead to effect estimates biased toward the null with less precision (i.e., larger standard errors for effect estimates) compared to fully individual-level studies. This could partially explain the lack of associations between dementia risk and PM2.5 and NO2 found in the current study. However, it may be helpful to inform environmental policies at the county-level. Moreover, studies that assessed air pollution exposure based solely on individual-level residential address may not be very accurate either by ignoring exposure at the workplace, during regular commutes, or during other social/physical activities [53,54]. Using more detailed residential and work addresses for the IHS users and evaluating the air pollution exposure based on this individual-level information would be a valuable future extension to the current study.
Second, we were only able to adjust for a few major risk factors for dementia at the individual-level including age, gender, and baseline comorbid conditions. Other important potential confounders or effect modifiers are not available, such as apolipoprotein (APOE) genotype and level of physical activity [13,55]. However, Wang et al. (2022) [25] reported that the associations between air pollution and dementia did not substantially differ by APOE genotype among US elderly women. Similarly, previous literature suggested that air pollution may prevent people from engaging in physical activity in highly polluted environments [56]. Therefore, instead of being a potential confounder, physical activity may lie on the causal pathway between air pollution and dementia and, thus, should not be adjusted for in the regression model as a covariate when air pollution is the main exposure. Future studies could investigate the joint effect of air pollution and physical inactivity on dementia or the extent to which physical inactivity explains the association between air pollution and dementia using causal mediation analysis in this population.
Third, identifying dementia patients via clinical diagnostic codes likely underestimates the prevalence and incidence of dementia in the population under consideration. Using 3 years of baseline may not be long enough to capture all the baseline prevalent dementia patients, which means some of the prevalent dementia patients may have been misclassified as incident cases and might lead to biased estimates of the association between air pollution and dementia risk. Meanwhile, an average of 3.67 years of follow-up is also relatively short and we are limited in statistical power because relatively few incident cases of dementia are expected during the few follow-up years given dementia is a chronic disease that usually takes many years to develop. Additionally, the providers at different locations and counties likely use somewhat different criteria when making dementia diagnoses. It is unclear if the potential geographic heterogeneity in the practice of dementia diagnosis could lead to biases in our estimates for the parameters of interests and warrants further study on this topic.
Last, as the study only included those who were regular users of the IHS services every year and registered at the same site during the baseline period, the analytical sample could be biased toward those with more chronic conditions who needed to visit their IHS/tribal providers regularly. As shown in Appendix E, when comparing those included in our study sample (n = 26,871) to those who were excluded due to lack of regular use of the IHS/tribal services at the same location but otherwise eligible for this study (n = 25,815), we found that the individuals included in our study sample had a similar mean age as those who were excluded, but the included sample had a higher percentage of females (59.5% vs. 54.7%), a higher percentage of Medicare enrollment (56.5% vs. 46.8%), and a higher proportion of private health insurance coverage (33.6% vs. 25.3%). Meanwhile, the included sample had a lower proportion of individuals enrolled in Medicaid than the excluded sample (15.9% vs. 16.8%). The potential selection bias in our study sample again calls for a future study that includes individual addresses for each IHS user to further evaluate the associations between air pollution exposures and dementia risk among American Indians.
