Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 May 23.
Published in final edited form as: Environ Res. 2026 Apr 16;301:124533. doi: 10.1016/j.envres.2026.124533

The association between PM2.5 components and cognitive decline: the impact of measurement error correction

Boya Zhang 1, Jaime E Hart 1,2, Francine Laden 1,2,3, Molin Wang 2,3,4, Zuoheng Wang 5, Gang Xu 5, Xin Zhou 5, Donna Spiegelman 5,6, Helen Suh 7, Peter James 1,8,9
PMCID: PMC13196913  NIHMSID: NIHMS2169618  PMID: 42000105

Abstract

Growing evidence suggests that exposure to fine particulate matter (PM2.5) may accelerate cognitive decline and increase dementia risk, but the roles of individual PM2.5 components, particularly when accounting for exposure measurement error, remain unclear. We followed 12,824 participants in the Nurses’ Health Study cognition substudy (2000–2008). One-year average PM2.5 component concentrations preceding each participant’s first cognitive assessment were assigned based on residential address. A global cognitive score was calculated as the mean of z-scores from six tests, and individual rates of cognitive decline were estimated using linear mixed models with random intercepts and slopes. Using information on the relationships of personal exposures and nearest-monitor PM2.5 components from a prior validation study, we employed double/debiased machine learning (DML) method to estimate associations between PM2.5 components and cognitive decline while accommodating high-dimensional multi-pollutant exposures and accounting for exposure measurement error. Higher exposure to several PM2.5 components were associated with changes in rates of cognitive decline. In multi-pollutant models without measurement error correction, each doubling of chloride (Cl) concentration was associated with a 0.0003 (95% CI: 0.0000, 0.0005) standard unit slower annual rate of cognitive decline, while each doubling in silicon (Si) was associated with a −0.0025 (95% CI: −0.0050, 0.0000) faster rate of decline. After correction for measurement error, associations generally strengthened: higher bromine (Br), manganese (Mn), and lead (Pb) were significantly associated with faster cognitive decline. In contrast, higher strontium (Sr) and vanadium (V) were associated with slower decline. Our findings suggest that long-term exposure to certain PM2.5 components, and their sources, may be more detrimental to cognitive aging than others, and highlight the critical need to correct for measurement error when assessing the health impacts of PM2.5 and its components. More research is needed to confirm these findings, ideally with access to larger validation studies of component exposures.

Keywords: PM2.5 components, Cognitive decline, Double/debiased machine learning, Exposure measurement error correction

Introduction

Long-term exposure to fine particulate matter (PM2.5) has been identified as a risk factor for cognitive function and rates of cognitive decline (Weuve et al., 2021). PM2.5 may impair cognitive function by triggering neuroinflammation through systemic inflammation or oxidative stress (X.-q. Liu et al., 2023). Some of the smallest particles, adsorbed with neurotoxic chemicals (Lai et al., 2021; Vanbrabant et al., 2024), may also reach the brain via the olfactory bulb or cross the blood-brain barrier (González-Maciel et al., 2017; Lucchini et al., 2012). Impaired cognitive function is a strong predictor of subsequent dementia, which can result in heavy emotional and economic burdens on the aging person’s family and society (Matthews et al., 2019). Compared with many other identifiable risk factors, PM2.5 exposure can be modified at the population level, making it an ideal focus for large-scale prevention.

As a complex mixture, different PM2.5 components may have varying neurotoxic effects. However, most studies only quantify associations with total PM2.5 mass, so it remains unclear whether specific components, such as trace elements, are more harmful to cognitive function than others. Assessing the impact of exposure to PM2.5 components would help clarify underlying biological mechanisms and also identify target components for intervention.

In addition, PM2.5 component exposure assessment varies across prior studies, including the use of the nearest monitor (Matthews et al., 2019), spatial exposure models (Li et al., 2022; Ye et al., 2022), and machine learning algorithms (Feng et al., 2024). All of these methods have potential to introduce substantial measurement error, especially for components with high spatial variability (Bell et al., 2007). Such measurement errors may involve both classical error, which underestimates the exposure-response association, and Berkson error, which broadens confidence intervals, potentially biasing the overall findings (Bergen et al., 2013; Szpiro et al., 2011; Zhang et al., 2025). However, to date, no studies have incorporated measurement error corrections when evaluating the effect of PM2.5 components on cognitive decline or related outcomes.

In our previous study, we developed a method that applies the double/debiased machine learning machine learning (DML) approach to estimate the effects of individual PM2.5 components in the presence of other PM2.5 components, while correcting for measurement error (Xu et al., 2024). This approach takes a semi-non-parametric approach to confounder control by multiple components and other covariates simultaneously, thereby potentially reducing bias with little loss of precision. Bias due to measurement error was corrected using regression calibration, which is typically used when surrogate exposure (ambient) data are available for most participants, but “true” exposure measurements (personal monitoring measurements) are only obtained for a subset or from an external validation study (Xu et al., 2024). In this study, we applied this method to evaluate the associations of 17 PM2.5 components: bromine (Br), calcium (Ca), chlorine (Cl), copper (Cu), iron (Fe), potassium (K), magnesium (Mg), manganese (Mn), nickel (Ni), lead (Pb), sulfur (S), selenium (Se), silicon (Si), strontium (Sr), titanium (Ti), vanadium (V), and zinc (Zn), with cognitive decline in a large, prospective, nationwide US-based cohort of women, while further correcting for exposure measurement error.

Methods

Study Population

We used data from the Nurses’ Health Study (NHS), a prospective cohort of female registered nurses in the US. In 1976, NHS enrolled 121,700 women aged 30-50 years across 11 US states. Once recruited, participants completed mailed questionnaires every two years asking about demographic characteristics, lifestyles, and history of chronic diseases. The residential addresses of NHS participants, updated every two years, are distributed throughout the US during the follow-up and have been geocoded to obtain latitude and longitude (Jaime E. Hart, Robin C. Puett, et al., 2015). We have included a map of the eligible NHS participants’ addresses in the Supplementary Figure S1.

During 1995-2001, NHS participants who were >70 years old and free of stroke were invited to participate in a telephone-based study of cognitive function. A total of 19,415 eligible women completed the first assessment (93%). Follow-up assessments were performed up to four times at two-year intervals until 2008, with a participation rate >90% among those who remained alive and could be contacted at each follow-up point.

For the present analyses, we included participants with at least two cognitive assessments since 1999 with no missing data on exposure measures or other key covariates. The study protocol was approved by the Institutional Review Boards of the Brigham and Women’s Hospital and the Harvard T.H. Chan School of Public Health. Consent was implied through return of the questionnaires or participation in the phone interviews.

Exposure Assessment

We estimated the monthly average ambient exposure of PM2.5 and 17 PM2.5 components, Br, Ca, Cl, Cu, Fe, K, Mg, Mn, Ni, Pb, S, Se, Si, Sr, Ti, V, and Zn, using concentrations measured at the monitors from the U.S. EPA database (Air Quality System, https://aqs.epa.gov/aqsweb/airdata/download_files.html) during 1999 to 2008. Exposures were assigned based on the nearest monitors located within 60 miles of participants’ residential addresses, updated every two years over follow-up. From the monthly measurements, we calculated the 1-year moving average ambient exposure preceding the date of the first cognitive assessment included in our analyses for each participant. We used this as the surrogate for true exposure of total PM2.5 and its components.

Cognitive Assessment

The assessment of cognitive function in NHS has been previously described (Devore et al., 2014; Fassier et al., 2022). Briefly, the cognitive assessments included the Telephone Interview of Cognitive Status (TICS) and five additional cognitive tests, including a delayed recall test of the 10-word list from the TICS, the East Boston Memory Test (EBMT) immediate and delayed recalls, a category fluency (animal naming) test, and a digital span backward test (Albert et al., 1991; Baddeley et al., 1991). We calculated the global cognitive score as the average of the z-scores of all tests to evaluate overall cognitive functioning, which simplified interpretation. This global cognitive score was also consistent with the majority of NHS papers using this subcohort (Liu et al., 2025; Pescador Jimenez et al., 2024).

Covariates

We obtained information on potential confounders from the nearest NHS questionnaire before the first cognitive assessment included in our analyses. We selected potential cofounders based on previous studies (Duchesne et al., 2022; Grande et al., 2021; Kulick et al., 2020; Weuve et al., 2012). The Directed Acyclic Graph (DAG) illustrating the potential causal structure of these selected confounders was presented in Supplementary Figure S2. Demographic variables included age (in years) at baseline cognitive assessment and follow-up years. Individual socioeconomic status (SES) measures included participants’ highest education level (Registered Nurse, Bachelor’s Degree, Master’s or Doctorate Degree), marital status (married, yes/no), and partner’s highest education level (less than high school, some high school, high school graduate, college graduate, graduate school). We also further adjusted for neighborhood SES (nSES), and the Census region (Northeast, Midwest, West, South) of participants’ residential addresses. nSES was assessed with a series of variables at the Census tract level, including educational attainment (% over 25 with college or higher education), income (median household income), wealth (median household home value, % families receiving interest dividends or rent income, % occupied housing units), employment status (% population 16 + years old unemployed), and racial composition (% White, % Black, % foreign-born) from the US Decennial Census or the American Community Survey closest to the NHS questionnaire year. We assigned a composite nSES score by z-scaling each component measure and summing across the indicators (DeVille et al., 2023).

Statistical Analysis

Compared to the traditional linear regression models, DML models may reduce the bias introduced by correlations among different PM2.5 components in a regularized non-parametric manner. However, the DML approach can only accommodate one observation per participant. Therefore, we first fit linear mixed models with follow-up years as both a fixed and random effect to estimate individual slopes of the global cognitive score, accounting for differences in cognitive change across participants. These slopes represented each participant’s annual rate of cognitive decline.

With this estimated cognitive decline slope as the outcome, we used the DML model to estimate the association of total PM2.5 and its components with the annual rate of cognitive decline. Our analysis first focused on single-pollutant models, which include PM2.5 components one at a time. Then, we fitted the multi-pollutant model that included all 17 PM2.5 components in one model simultaneously. We applied the logarithmic transformation to the PM2.5 component concentrations, given their skewed distribution, and to be consistent with the external validation study described below (Xu et al., 2024). After evaluating the association between log-transformed concentrations of PM2.5 components and the rates of cognitive decline, we reported the differences in the annual rates of cognitive decline with 95% confidence intervals (CI) per doubling increase in the concentrations of each PM2.5 component. To help interpret these mean differences, we compared the associations to age-related differences in cognitive decline rates (fitted in DML models only adjusted for age).

Finally, we extended the DML models to correct for measurement error when using ambient concentrations of total PM2.5 and its components as a surrogate exposure. This correction is implemented by fitting linear mixed models in an external validation study with data from the Multi-Ethnic Study of Atherosclerosis (MESA). The MESA is a community-based, prospective, observational cohort study among adults aged 45 to 84 years. The initial baseline cohort was recruited from 6 US communities (Baltimore City and Baltimore County, MD; Chicago, IL; Forsyth County, NC; Los Angeles County, CA; Northern Manhattan and the Bronx, NY; and St. Paul, MN) (Bild et al., 2002). A subset of participants was recruited for personal monitoring and completed the two-week personal sampling during the summer and winter (Cohen et al., 2009). We assumed that personal exposure to PM2.5 and its components were the true exposure. These exposures were measured over 5 consecutive days during both winter and summer across 1,092 days. The paired measurements of ambient PM2.5 samples were collected through active sampling on Teflon filters and then measured using X-ray fluorescence (Xu et al., 2024). To stabilize the correlations between true and surrogate exposures, log-transformations were applied to both true and surrogate measurements. We used the multi-pollutant DML model with measurement error correction for the main results. A brief, stepwise summary of our DML analysis with measurement error correction is provided in Supplementary Text S1, and further methodological details are available in our previous publication (Xu et al., 2024).

Both single and multi-pollutant DML models above were adjusted for individual-level age at first cognitive assessment included in our analyses, follow-up years since first cognitive assessment, education level, marital status, and partner’s education level. Area-level adjustments included Census region and nSES. We also adjusted for the total PM2.5 to control for overall PM2.5 exposure, allowing us to isolate the associations of PM2.5 components independent of absolute exposure.

In sensitivity analyses, we further adjusted the corrected multi-pollutant models for lifestyle behaviors, including smoking status (never, past, current), alcohol consumption (non-drinkers, 1-14g/day, ≥15g/day), and physical activity (total metabolic equivalent hours per week). We also examined whether the distribution of estimated cognitive decline slopes varied according to the number of cognitive assessments (n = 2-4) used in their calculation. When variation in distribution was observed, we further stratified our analyses based on the number of assessments used to derive cognitive decline. All analyses were conducted using R (version 3.6.3; R Foundation for Statistical Computing, Vienna, Austria) and the reg_dml_tuning function implemented in Python (version 3.10.12).

Results

Table 1 presents the characteristics of our study population overall and by quartile for the annual rate of cognitive decline. Among the 12,824 participants, the mean age at baseline was 76.1 with a standard deviation (SD) of 2.8 years. Compared with participants with the slowest rates of cognitive decline (Q4), those experiencing faster decline tended to be older, have lower levels of education, and live in areas with lower nSES. Participants with the fastest rate of cognitive decline (Q1) were more likely to have higher concentrations of total PM2.5, Ca, Fe, K, and Si, compared to the other three quartiles. We observed moderate correlations among certain PM2.5 components, which may indicate similar origins. For example, we found relatively high correlations among Cu, Fe, Mn, Pb, and Zn, which may be used as indicators for industrial and traffic-related sources (Figure 1).

Table 1.

Characteristics of Women by Quartile of Annual Rate of Cognitive Decline during the Cognitive Function Sub-study of the US-based Nationwide Nurses’ Health Study.

Quartile of Annual Rate of Cognitive Decline
Total Sample
(N=12,824)		 Q1
(−0.30, −0.06)		 Q2
(−0.06, −0.04)		 Q3
(−0.04, −0.01)		 Q4
(−0.01, 0.14)
Age, mean(sd), year 76.1 (2.8) 76.7 (2.7) 76.3 (2.8) 75.8 (2.7) 75.5 (2.7)
Participant’s Education, %
 Bachelor, %(n) 16.8 (2154) 15.5 (496) 16.2 (520) 16.6 (532) 18.9 (606)
 Master or Doctorate, %(n) 6.2 (794) 5.2 165) 5.1 (164) 6.3 (200) 8.3 (265)
Husband’s Education, %
 Some High School 5.1 (651) 5.3 (169) 5.8 (187) 4.9 (156) 4.3 (139)
 High School Graduate 34.6 (4440) 33.7 (1079) 35.1 (1126) 35.2 (1125) 34.6 (1110)
 College Graduate, %(n) 22.7 (2911) 22.2 (710) 21.4 (686) 22.5 (720) 24.8 (795)
 Graduate School, %(n) 16.3 (2095) 15.4 (494) 14.7 (473) 16.4 (523) 18.8 (605)
 Missing, %(n) 19.2 (2465) 21.4 (687) 20.2 (650) 19.3 (616) 15.9 (512)
 Population Density (num per sq km) nSES 1364 (2886) 1510 (3391) 1321 (2690) 1290 (2238) 1335 (3088)
0.04 (3.75) 0.03 (3.73) −0.08(3.74) 0.01 (3.73) 0.22 (3.79)
Census Region, %(n)
 Northeast 53 (6843) 53 (1692) 53 (1708) 53 (1707) 54 (1736)
 Midwest 16 (2038) 16 (526) 16 (507) 16 (508) 16 (497)
 South 17 (2169) 16 (511) 18 (569) 17 (548) 17 (541)
 West 14 (1774) 15 (476) 13 (426) 14 (435) 14 (437)
Total PM2.5 (μg/m3) 13.3 (3.0) 13.4 (3.0) 13.3 (3.0) 13.3 (3.0) 13.3 (2.9)
PM2.5 Components (ng/m3)
 Br 3.2 (1.0) 3.2 (1.0) 3.2 (1.0) 3.2 (1.0) 3.2 (1.0)
 Ca 50.7 (35.4) 51.1 (37.5) 50.3 (34.4) 51.1 (35.7) 50.3 (33.8)
 Cl 44.7 (98.0) 42.8 (93.3) 44.2 (100.3) 47.5 (101.8) 44.3 (96.3)
 Cu 2.9 (1.8) 2.9 (1.8) 2.9 (1.9) 2.9 (1.8) 2.9 (1.8)
 Fe 82.5 (51.2) 83.4 (53.0) 81.2 (47.8) 82.9 (52.7) 82.7 (51.2)
 K 56.4 (1.0) 56.8 (25.3) 56.3 (24.8) 56.3 (24.7) 56.2 (28.2)
 Mg 8.0 (7.4) 8.0 (7.9) 7.9 (6.8) 8.2 (7.7) 7.9 (7.2)
 Mn 2.6 (2.2) 2.6 (2.2) 2.5 (1.9) 2.6 (2.6) 2.6 (2.1)
 Ni 3.7 (6.5) 3.8 (6.5) 3.7 (6.5) 3.6 (6.2) 3.9 (6.6)
 Pb 4.6 (2.6) 4.6 (2.5) 4.5 (2.4) 4.6 (2.4) 4.6 (2.9)
 S 1158.2 (421.3) 1157.7 (427.8) 1161.7 (417.8) 1149.5 (419.4) 1163.9 (420.0)
 Se 1.4 (1.5) 1.4 (1.5) 1.4 (1.5) 1.4 (1.5) 1.4 (1.5)
 Si 113.1 (69.4) 114.9 (73.9) 113.3 (70.1) 112.4 (67.6) 111.7 (65.4)
 Sr 0.7 (0.4) 0.7 (0.5) 0.6 (0.4) 0.7 (0.4) 0.7 (0.4)
 Ti 6.3 (3.0) 6.3 (3.1) 6.3 (3.0) 6.2 (3.0) 6.3 (3.0)
 V 2.4 (1.9) 2.4 (1.9) 2.4 (1.9) 2.4 (1.9) 2.4 (2.0)
 Zn 14.2 (11.5) 14.4 (11.9) 13.8 (10.5) 14.3 (11.8) 14.5 (11.9)
Open in a new tab
*

Values are means(SD) or medians(Q25, Q75) for continuous variables; percentages or ns or both for categorical variables.

Figure 1.

Figure 1.

Open in a new tab

Spearman correlations between PM2.5 and its components exposures in the Cognitive Function Sub-study of the US-based Nationwide Nurses’ Health Study (N=12,824).

In the single-pollutant DML model without measurement error correction (Supplementary Table S1), we found that higher Cl concentrations were associated with a slower rate of cognitive decline, whereas higher Si concentrations were associated with a faster rate of cognitive decline. Both associations remained in the multi-pollutant DML model (Figure 2). Specifically, a doubling increase in the Cl concentration was associated with a 0.0003 (95%CI 0.0000, 0.0005) standard unit slower annual rate of cognitive decline. In contrast, a doubling increase in Si concentration was associated with a −0.0025 (95%CI −0.0050, 0.0000) standard unit faster annual rate of cognitive decline. To help interpret our estimates, we found that 1 year of age is linked to a −0.0040 standard unit annual difference for cognition. Therefore, higher exposure to Si was equivalent to accelerating cognitive aging by approximately 7.5 months.

Figure 2.

Figure 2.

Open in a new tab

The change in the rate of cognitive decline for a doubling increase in the concentrations of PM2.5 components using multi-pollutant DML models without and with measurement error correction in the Cognitive Function Sub-study of the US-based Nationwide Nurses’ Health Study (N=12,824

After correcting for measurement error, the estimated associations became substantially stronger, although with wider confidence intervals, compared to those obtained without measurement error correction (Figure 2). In addition to the association previously observed for Si, we newly identified statistically significant associations between higher concentrations of Br, Mn, and Pb and faster cognitive decline. Specifically, per doubling increase in concentration was associated with a −0.0451 (95% CI: −0.0881, −0.0020) standard unit faster annual cognitive decline rate for Si, −0.1830 (95% CI: −0.3181, −0.0479) for Br, −0.0256 (95% CI: −0.0416, −0.0096) for Mn, and −0.0259 (95% CI: −0.0366, −0.0152) for Pb in our final multi-pollutant DML models adjusted for measurement error, which is equivalent to cognitive aging ranging from 8.7 years for Mn and 62.5 years for Br. In contrast, higher concentrations of Sr and V were associated with a slower rate of cognitive decline.

In sensitivity analyses, the results remained similar after additional adjustment for lifestyle behavior variables, supporting the robustness of our main findings (Supplementary Table S2). The distribution of the derived rate of cognitive decline did not vary by the number of cognitive assessments included (Supplementary Figure S3).

Discussion

Using a nationwide US prospective cohort, we found that long-term exposure to Br, Mn, Pb, and Si was associated with an accelerated rate of cognitive decline. Adjusting for measurement error increased the magnitude of the associations while widening the confidence intervals, suggesting that measurement error is present when relying on ambient exposures as surrogates for personal exposure of ambient origin. As the first study that we know of to assess PM2.5 components and cognitive decline while adjusting for measurement error, we identify key harmful components and highlight the importance of correcting for measurement error when using ambient exposure data.

Although not statistically significant, our findings on the association between higher total PM2.5 and faster rates of cognitive decline are consistent with the findings from previous studies (Duchesne et al., 2022; Grande et al., 2021; Kulick et al., 2020). The inconsistencies could result from the discrepancies in the inclusion of follow-up duration, differing methods and degrees of measurement error in the exposure assessment, or the choice of statistical methods. For example, unlike the earlier NHS study that reported a significant association between total PM2.5 exposure and cognitive decline, our analysis included an additional cognitive assessment and estimated PM2.5 exposure using the nearest monitor method, rather than a spatiotemporal model (Weuve et al., 2012). Also, compared with the growing body of evidence on the association between total PM2.5 and cognitive decline (Duchesne et al., 2022; Grande et al., 2021; Kulick et al., 2020; Weuve et al., 2012), only a limited number of studies have evaluated the association between PM2.5 components and cognitive-related outcomes, such as cognitive function (J. Liu et al., 2023), incident cognitive impairment (Zhang et al., 2024), and dementia (Li et al., 2022; Shi et al., 2023; Zheng et al., 2025). However, the majority of them only focused on major chemical components, including black carbon (BC), organic matter (OM), nitrate (NO3−), sulfate (SO42−), ammonium (NH4+), and soil dust, instead of the trace elements evaluated in our study. Therefore, it is hard to directly compare our findings with previous results. Nevertheless, we can compare the potential major contributing sources to the observed associations between PM2.5 and adverse cognitive outcomes, as both major components and trace elements may serve as tracers of similar emission sources.

As a crystal element, Si is commonly identified as an indicator for soil, which can be formed through natural processes like eolian erosion or industrial activities like mining and cement production. It is often emitted alongside other elements, such as Al, Ca, and Fe (Zhu et al., 2024). In our study, Si was found to be most consistently associated with a faster rate of cognitive decline, even before correcting for bias due to measurement error. This finding aligns with the positive association observed between soil particles and dementia in a US Medicare cohort in the northeastern US (Li et al., 2022).

After correcting for measurement error, Pb and Mn were also found to be associated with faster rates of cognitive decline, suggesting a possible contribution from industrial sources (Hsu et al., 2021; Ma et al., 2024). This association may be explained by the harmful effect of metal accumulation, particularly heavy metals like lead, in the blood and bone on the central nervous system (Gilani et al., 2015).

Previous studies have identified SO42−, BC, and OM as the PM2.5 components most strongly and consistently associated with adverse cognitive outcomes (Li et al., 2022; J. Liu et al., 2023; Shi et al., 2023; Zhang et al., 2024; Zheng et al., 2025), indicating that fossil fuel combustion and traffic sources are likely major contributors to PM2.5-related cognitive risks. In our study, we found an association between Br, which could be an indicator of coal and biomass combustion (Ma et al., 2024), and a faster cognitive rate after correction of measurement error. However, we did not find an association between other related elements (e.g., S) with faster cognitive decline. This inconsistency might be due to sulfur being predominantly found in the form of SO42− rather than the element (Schlesinger, 2007).

In contrast to our hypothesis, we found a slower rate of cognitive decline associated with higher levels of Sr and V concentrations. These findings may be partly explained by exposure measurement error that was not fully addressed by our methods, particularly for Sr and V, which have the lowest mean concentrations among the exposures assessed in our study. In addition, the Sr in PM2.5 is often attributed to traffic-related sources such as brake and tire wear (Jose & Srimuruganandam, 2021), and V may originate from shipping emissions (Hennigan et al., 2019). These sources tend to be more localized, and our nearest-monitor method may be unable to capture the spatial variation, potentially leading to exposure misclassification.

The measurement error correction approach we used is tailored for long-term multi-pollutant studies, where exposure estimates for PM2.5 components often differ from individuals’ true personal exposures (Xu et al., 2024). Measurement error was first characterized with regression calibration using linear mixed models in MESA as the external validation study with detailed individual-level personal exposure in a subset of participants (Kaufman et al., 2012). Then, to minimize the risk of bias due to correlation between PM2.5 components and the desire for robustness in modeling assumptions about them, this approach applied an extended DML framework that allows for estimation of each component’s effects in the presence of both confounding and correlated measurement errors of differing degrees (Xu et al., 2024). Although the personal exposures measured in the external validation study may also be subject to error, as long as these errors are unlikely to be correlated with those from the errors in the nearest monitor data, as is likely to be the case, our corrected estimates and associated CIs will be valid for assessing the effects of individual PM2.5 components (J. E. Hart et al., 2015).

A variety of other methods addressing measurement error in air pollution studies have been proposed before, including simulation extrapolation, bootstrapping, two-stage hierarchical modeling, and PCA-based corrections. While effective in some situations, most of these approaches are limited to handle only a small number of mismeasured exposures, rely on strong assumptions, and/or only handle two-pollutant scenarios (Alexeeff et al., 2016; Bergen et al., 2013; Jaime E. Hart, Donna Spiegelman, et al., 2015; Liao et al., 2018; Wentzell et al., 1997). Furthermore, only a few studies have combined the validation study and main study frameworks, where validation data on personal exposure are used to correct bias in the larger analytic sample (Cai et al., 2023; Rosner et al., 1990; Spiegelman et al., 2000). We found no other method that corrects estimates of the effects of individual PM2.5 components when the components are subject to correlated measurement error. Overall, our findings suggested that the estimated effect of PM2.5 components is characterized by complex measurement error structures.

Strengths and Limitations

Our study had several limitations that should be noted. First, we estimated exposures in our main analyses using the nearest monitor method, which may not capture all of the spatial variability in PM2.5 concentrations and is more likely to introduce nondifferential exposure misclassification (Pinto et al., 2004). In this case, the differences in the estimated associations, before and after measurement error correction, may partly reflect the improved spatial resolution of personal exposure compared with monitor-based ambient exposure. While similar gains could, in principle, be achieved using high-resolution ambient spatiotemporal models for PM2.5 components, as provided by the recently developed spatiotemporal models for PM2.5 components (Amini et al., 2023a, 2023b), the nearest-monitor method partially captures the influence of indoor exposures, whereas spatiotemporal models generally characterize only outdoor concentrations. It also allows inclusion of a wider range of PM2.5 components, as some of the elements examined in our study are not yet included in existing spatiotemporal prediction models (e.g., Cl, Mg, Mn, Se, Sr, Ti). More importantly, the measurement error–correction method used in this study is, at present, only applicable to nearest-monitor–based exposure estimates. We are in the process of extending our measurement error–correction methods to spatiotemporal models, and once these methods are fully developed, we plan to conduct corresponding analyses in future work. Additionally, the external validation study for measurement error correction was based solely on personal exposures from MESA. The limited geographic coverage of the external validation study may have introduced error into our measurement error corrections. Also, in the external validation studies, MESA participants wore personal sampling devices for five consecutive days during both the summer and winter seasons. These relatively short-term measurements were then used to estimate monthly average “true” exposures, under the assumption that the relationship between personal measurements and surrogate ambient exposures during the sampled days was representative of that for the entire month, even though this may not always hold. However, according to the MESA protocol, the selected 5 days include both weekdays and weekends, allowing capture of both weekday and weekend activity patterns. Another limitation of our current approach is that the DML-based methods are unable to incorporate repeated measurements as available in longitudinal studies. As a result, we calculated the rate of cognitive decline as the dependent variable in our analyses. This derived slope may not fully capture variation in cognitive trajectories over the follow-up period. However, we found that the estimated slope remained consistent across different follow-up durations, and the use of such slopes to quantify cognitive decline has been widely adopted in previous studies, including those involving the NHS cohort (Wagner et al., 2020; Zammit et al., 2025).

Lastly, the generalizability of our findings may be limited by the characteristics of the study population, which was comprised of predominantly White, older, females with nursing degrees, which may restrict the applicability of the results to younger, more racially, socioeconomically, and educationally diverse populations. However, there is no evidence suggesting that the distribution of health status or health behaviors among nurses significantly differs from that of the general population (Stanulewicz et al., 2019). Furthermore, their nursing degrees may enable them to provide more accurate self-reported information, potentially reducing information bias compared to other cohorts (Bao et al., 2016). In addition, cognitive performance was assessed using telephone-based instruments, which, although practical in large cohort studies, may be less sensitive than in-person neuropsychological batteries and may be more susceptible to differential misclassification. However, a previous validation study found that this telephone-based cognitive battery performed well compared with a detailed, in-person neuropsychological assessment (Rapp et al., 2012). The focus on early changes in cognitive aging trajectories rather than clinical ADRD outcomes means that we cannot directly assess incident ADRD or clinically defined cognitive impairment. Future research is needed to directly relate specific PM2.5 components to these clinical endpoints. However, our cognitive outcomes do enable the detection of subtle changes at earlier stages of cognitive decline.

Our study also has several major strengths that are worth mentioning. To the best of our knowledge, we are the first to estimate the associations between PM2.5 components and the rate of cognitive decline in the US. While this alone is important, an additional innovative aspect is the use of a novel DML-based approach. This method enables valid estimation of effects for multiple mismeasured and correlated PM2.5 components, while simultaneously correcting for the exposure measurement errors when using ambient PM2.5 concentrations as the surrogate exposure when personal exposure is the object of interest. Additionally, we leveraged comprehensive data from the nationwide cognitive substudy of the NHS, which includes over 10,000 participants from diverse geographic regions with a wide range of PM2.5 and component exposures. Each participant completed up to four objective cognitive assessments, allowing for precise and repeated measurement of cognitive decline. The availability of detailed longitudinal data and extensive covariate information further reduces residual confounding and strengthens the validity of our findings.

Conclusions

In this large nationwide cohort of middle-aged and elderly women, long-term exposures to specific PM2.5 components were associated with an accelerated rate of cognitive decline when measurement error was addressed using a novel statistical approach. Specifically, exposures to Br, Mn, Pb, and Si, which may originate from soil and industrial emissions sources, were significantly associated with faster cognitive decline after correction for exposure measurement error. There was clear evidence that measurement error correction not only increased the magnitude of these associations but also identified additional harmful components. Therefore, our study provides evidence that long-term exposures to certain PM2.5 components may contribute to cognitive decline, and highlights the necessity of applying measurement error correction when analyzing health effects of ambient air pollutants.

Supplementary Material

1

Highlights.

  • Assessed associations between PM2.5 components and cognitive decline.

  • Used double/debiased machine learning for high-dimensional multi-pollutant models.

  • Used external validation data to correct exposure measurement error.

  • Higher Br, Mn, Pb, and Si were linked to faster cognitive decline.

  • Measurement error correction strengthened associations, highlighting its importance.

Acknowledgment

This study was supported by the National Institutes of Health (NIH) National Cancer Institute (NCI) NHS cohort infrastructure grant (UM1 CA186107), NCI grant (R01 CA279175), National Institute of Environmental Health Sciences (NIEHS) grants (R01 ES026246; R01 ES035501), and an Alzheimer’s Association Research Fellowship (25AARF-1372330). We would like to thank the participants and staff of the NHS for their valuable contributions.

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.

Declaration of interests

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

Reference

  1. Albert M, Smith LA, Scherr PA, Taylor JO, Evans DA, & Funkenstein HH (1991). Use of brief cognitive tests to identify individuals in the community with clinically diagnosed Alzheimer’s disease. International journal of Neuroscience, 57(3-4), 167–178. [DOI] [PubMed] [Google Scholar]
  2. Alexeeff SE, Carroll RJ, & Coull B (2016). Spatial measurement error and correction by spatial SIMEX in linear regression models when using predicted air pollution exposures. Biostatistics, 17(2), 377–389. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Amini H, Danesh-Yazdi M, Di Q, Requia W, Wei Y, AbuAwad Y, Shi L, Franklin M, Kang C, & Wolfson MJ (2023a). Annual mean PM2. 5 components (EC, NH4, NO3, OC, SO4) 50m urban and 1km non-urban area grids for contiguous US, 2000-2019 v1. NASA Socioeconomic Data and Applications Center (SEDAC) data set, 7WJ3. [Google Scholar]
  4. Amini H, Danesh-Yazdi M, Di Q, Requia W, Wei Y, AbuAwad Y, Shi L, Franklin M, Kang C, & Wolfson MJ (2023b). Annual Mean PM2. 5 Trace Elements 50m Grids in Urban Areas and 1km Grids in Non-Urban Areas for Contiguous US, 2000-2019, v1. NASA Socioeconomic Data and Applications Center (SEDAC) data set, 1X94. [Google Scholar]
  5. Baddeley AD, Bressi S, Della Sala S, Logie R, & Spinnler H (1991). The decline of working memory in Alzheimer’s disease: A longitudinal study. Brain, 114(6), 2521–2542. [DOI] [PubMed] [Google Scholar]
  6. Bao Y, Bertoia ML, Lenart EB, Stampfer MJ, Willett WC, Speizer FE, & Chavarro JE (2016). Origin, methods, and evolution of the three Nurses’ Health Studies. American journal of public health, 106(9), 1573–1581. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Bell ML, Dominici F, Ebisu K, Zeger SL, & Samet JM (2007). Spatial and temporal variation in PM2. 5 chemical composition in the United States for health effects studies. Environmental health perspectives, 115(7), 989–995. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Bergen S, Sheppard L, Sampson PD, Kim S-Y, Richards M, Vedal S, Kaufman JD, & Szpiro AA (2013). A national prediction model for PM2. 5 component exposures and measurement error–corrected health effect inference. Environmental health perspectives, 121(9), 1017–1025. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Bild DE, Bluemke DA, Burke GL, Detrano R, Diez Roux AV, Folsom AR, Greenland P, JacobsJr DR, Kronmal R, & Liu K (2002). Multi-ethnic study of atherosclerosis: objectives and design. American journal of epidemiology, 156(9), 871881. [DOI] [PubMed] [Google Scholar]
  10. Cai J, Zhang N, Zhou X, Spiegelman D, & Wang M (2023). Correcting for Bias Due to Mismeasured Exposure History in Longitudinal Studies with Continuous Outcomes. Biometrics, 79(4), 3739–3751. 10.1111/biom.13877 [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Cohen MA, Adar SD, Allen RW, Avol E, Curl CL, Gould T, Hardie D, Ho A, Kinney P, & Larson TV (2009). Approach to estimating participant pollutant exposures in the Multi-Ethnic Study of Atherosclerosis and Air Pollution (MESA Air). Environmental science & technology, 43(13), 4687–4693. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. DeVille NV, Iyer HS, Holland I, Bhupathiraju SN, Chai B, James P, Kawachi I, Laden F, & Hart JE (2023). Neighborhood socioeconomic status and mortality in the Nurses’ Health Study (NHS) and the Nurses’ Health Study II (NHSII). Environmental Epidemiology, 7(1), e235. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Devore EE, Grodstein F, Duffy JF, Stampfer MJ, Czeisler CA, & Schernhammer ES (2014). Sleep duration in midlife and later life in relation to cognition. Journal of the American geriatrics society, 62(6), 1073–1081. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Duchesne J, Gutierrez L-A, Carrière I, Mura T, Chen J, Vienneau D, de Hoogh K, Helmer C, Jacquemin B, & Berr C (2022). Exposure to ambient air pollution and cognitive decline: Results of the prospective Three-City cohort study. Environment International, 161, 107118. [DOI] [PubMed] [Google Scholar]
  15. Fassier P, Kang JH, Lee IM, Grodstein F, & Vercambre M-N (2022). Vigorous physical activity and cognitive trajectory later in life: prospective association and interaction by apolipoprotein E e4 in the Nurses’ Health Study. The Journals of Gerontology: Series A, 77(4), 817–825. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Feng Y, Castro E, Wei Y, Jin T, Qiu X, Dominici F, & Schwartz J (2024). Long-term exposure to ambient PM2. 5, particulate constituents and hospital admissions from non-respiratory infection. Nature Communications, 15(1), 1518. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Gilani SR, Batool M, Ali Zaidi SR, Mahmood Z, Bhatti AA, & Durrani AI (2015). Central nervous system (CNS) toxicity caused by metal poisoning: Brain as a target organ. Pakistan journal of pharmaceutical sciences, 28(4). [PubMed] [Google Scholar]
  18. González-Maciel A, Reynoso-Robles R, Torres-Jardon R, Mukherjee PS, & Calderón-Garcidueñas L (2017). Combustion-derived nanoparticles in key brain target cells and organelles in young urbanites: culprit hidden in plain sight in Alzheimer’s disease development. Journal of Alzheimer’s Disease, 59(1), 189–208. [DOI] [PubMed] [Google Scholar]
  19. Grande G, Wu J, Ljungman PLS, Stafoggia M, Bellander T, & Rizzuto D (2021). Long-Term Exposure to PM 2.5 and Cognitive Decline: A Longitudinal Population-Based Study. Journal of Alzheimer’s Disease(Preprint), 1–9. [DOI] [PubMed] [Google Scholar]
  20. Hart JE, Liao X, Hong B, Puett RC, Yanosky JD, Suh H, Kioumourtzoglou MA, Spiegelman D, & Laden F (2015). 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, 14, 38. 10.1186/s12940-015-0027-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Hart JE, Puett RC, Rexrode KM, Albert CM, & Laden F (2015). Effect modification of long-term air pollution exposures and the risk of incident cardiovascular disease in US women. Journal of the American Heart Association, 4(12), e002301. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Hart JE, Spiegelman D, Beelen R, Hoek G, Brunekreef B, Schouten LJ, & van den Brandt P (2015). Long-term ambient residential traffic–related exposures and measurement error–adjusted risk of incident lung cancer in the Netherlands cohort study on diet and cancer. Environmental health perspectives, 123(9), 860–866. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Hennigan CJ, Mucci A, & Reed BE (2019). Trends in PM2. 5 transition metals in urban areas across the United States. Environmental Research Letters, 14(10), 104006. [Google Scholar]
  24. Hsu C-Y, Chi K-H, Wu C-D, Lin S-L, Hsu W-C, Tseng C-C, Chen M-J, & Chen YC (2021). Integrated analysis of source-specific risks for PM2.5-bound metals in urban, suburban, rural, and industrial areas. Environmental Pollution, 275, 116652. 10.1016/j.envpol.2021.116652 [DOI] [PubMed] [Google Scholar]
  25. Jose J, & Srimuruganandam B (2021). Source apportionment of urban road dust using four multivariate receptor models. Environmental Earth Sciences, 80(19), 666. [Google Scholar]
  26. Kaufman JD, Adar SD, Allen RW, Barr RG, Budoff MJ, Burke GL, Casillas AM, Cohen MA, Curl CL, Daviglus ML, Roux AVD, Jacobs DR Jr., Kronmal RA, Larson TV, Liu SL-J, Lumley T, Navas-Acien A, O’Leary DH, Rotter JI, … Tracy RP (2012). Prospective Study of Particulate Air Pollution Exposures, Subclinical Atherosclerosis, and Clinical Cardiovascular Disease: The Multi-Ethnic Study of Atherosclerosis and Air Pollution (MESA Air). American journal of epidemiology, 176(9), 825–837. 10.1093/aje/kws169 [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Kulick ER, Wellenius GA, Boehme AK, Joyce NR, Schupf N, Kaufman JD, Mayeux R, Sacco RL, Manly JJ, & Elkind MSV (2020). Long-term exposure to air pollution and trajectories of cognitive decline among older adults. Neurology, 94(17), e1782–e1792. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7274848/pdf/NEUROLOGY2018943654.pdf [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Lai W, Li S, Li Y, & Tian X (2021). Air Pollution and Cognitive Functions: Evidence from Straw Burning in China. American Journal of Agricultural Economics. [Google Scholar]
  29. Li J, Wang Y, Steenland K, Liu P, van Donkelaar A, Martin RV, Chang HH, Caudle WM, Schwartz J, & Koutrakis P (2022). Long-term effects of PM2. 5 components on incident dementia in the Northeastern United States. The Innovation, 100208. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Liao X, Zhou X, Wang M, Hart JE, Laden F, & Spiegelman D (2018). Survival analysis with functions of mismeasured covariate histories: the case of chronic air pollution exposure in relation to mortality in the nurses’ health study. J R Stat Soc Ser C Appl Stat, 67(2), 307–327. 10.1111/rssc.12229 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Liu J, Liu R, Zhang Y, Lao X, Mandeville KL, Ma X, & Di Q (2023). Leisure-time physical activity mitigated the cognitive effect of PM2. 5 and PM2. 5 components exposure: Evidence from a nationwide longitudinal study. Environment International, 179, 108143. [DOI] [PubMed] [Google Scholar]
  32. Liu X.-q., Huang J, Song C, Zhang T.-l., Liu Y.-p., & Yu L (2023). Neurodevelopmental toxicity induced by PM2. 5 Exposure and its possible role in Neurodegenerative and mental disorders. Human & Experimental Toxicology, 42, 09603271231191436. [DOI] [PubMed] [Google Scholar]
  33. Liu Y, Gu X, Li Y, Wang F, Vyas CM, Peng C, Dong D, Li Y, Zhang Y, & Zhang Y (2025). Interplay of genetic predisposition, plasma metabolome and Mediterranean diet in dementia risk and cognitive function. Nature Medicine, 31(11), 3790–3800. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Lucchini RG, Dorman DC, Elder A, & Veronesi B (2012). Neurological impacts from inhalation of pollutants and the nose–brain connection. Neurotoxicology, 33(4), 838–841. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Ma T, Knobel P, Hadley M, Colicino E, Amini H, Federman A, Schwartz J, Steenland K, & Sade MY (2024). Source-specific PM2. 5 and atherosclerotic cardiovascular disease mortality. NEJM evidence, 3(12), EVIDoa2400182. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Matthews KA, Xu W, Gaglioti AH, Holt JB, Croft JB, Mack D, & McGuire LC (2019). Racial and ethnic estimates of Alzheimer’s disease and related dementias in the United States (2015–2060) in adults aged≥ 65 years. Alzheimer’s & Dementia, 15(1), 17–24. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Pescador Jimenez M, Wagner M, Laden F, Hart JE, Grodstein F, & James P (2024). Midlife residential greenness and late-life cognitive decline among Nurses’ Health Study participants. Environmental health perspectives, 132(7), 077003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Pinto JP, Lefohn AS, & Shadwick DS (2004). Spatial variability of PM2. 5 in urban areas in the United States. Journal of the Air & Waste Management Association, 54(4), 440449. [DOI] [PubMed] [Google Scholar]
  39. Rapp SR, Legault C, Espeland MA, Resnick SM, Hogan PE, Coker LH, Dailey M, Shumaker SA, & Group CATS (2012). Validation of a cognitive assessment battery administered over the telephone. Journal of the American geriatrics society, 60(9), 1616–1623. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Rosner B, Spiegelman D, & Willett WC (1990). Correction of logistic regression relative risk estimates and confidence intervals for measurement error: the case of multiple covariates measured with error. American journal of epidemiology, 132(4), 734–745. [DOI] [PubMed] [Google Scholar]
  41. Schlesinger RB (2007). The health impact of common inorganic components of fine particulate matter (PM2. 5) in ambient air: a critical review. Inhalation toxicology, 19(10), 811–832. [DOI] [PubMed] [Google Scholar]
  42. Shi L, Zhu Q, Wang Y, Hao H, Zhang H, Schwartz J, Amini H, van Donkelaar A, Martin RV, & Steenland K (2023). Incident dementia and long-term exposure to constituents of fine particle air pollution: A national cohort study in the United States. Proceedings of the National Academy of Sciences, 120(1), e2211282119. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Spiegelman D, Rosner B, & Logan R (2000). Estimation and inference for logistic regression with covariate misclassification and measurement error in main study/validation study designs. Journal of the American Statistical Association, 95(449), 51–61. [Google Scholar]
  44. Stanulewicz N, Knox E, Narayanasamy M, Shivji N, Khunti K, & Blake H (2019). Effectiveness of lifestyle health promotion interventions for nurses: a systematic review. International journal of environmental research and public health, 17(01), 17. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Szpiro AA, Sheppard L, & Lumley T (2011). Efficient measurement error correction with spatially misaligned data. Biostatistics, 12(4), 610–623. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Vanbrabant K, Van Dam D, Bongaerts E, Vermeiren Y, Bové H, Hellings N, Ameloot M, Plusquin M, De Deyn PP, & Nawrot TS (2024). Accumulation of ambient black carbon particles within key memory-related brain regions. JAMA Network Open, 7(4), e245678–e245678. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Wagner M, Grodstein F, Proust-Lima C, & Samieri C (2020). Long-term trajectories of body weight, diet, and physical activity from midlife through late life and subsequent cognitive decline in women. American journal of epidemiology, 189(4), 305–313. [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Wentzell PD, Andrews DT, & Kowalski BR (1997). Maximum likelihood multivariate calibration. Analytical chemistry, 69(13), 2299–2311. [DOI] [PubMed] [Google Scholar]
  49. Weuve J, Bennett EE, Ranker L, Gianattasio KZ, Pedde M, Adar SD, Yanosky JD, & Power MC (2021). Exposure to Air Pollution in Relation to Risk of Dementia and Related Outcomes: An Updated Systematic Review of the Epidemiological Literature. Environmental health perspectives, 129(9), 096001. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Weuve J, Puett RC, Schwartz J, Yanosky JD, Laden F, & Grodstein F (2012). Exposure to particulate air pollution and cognitive decline in older women. Archives of internal medicine, 172(3), 219–227. [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Xu G, Zhou X, Wang M, Zhang B, Jiang W, Laden F, Suh HH, Szpiro AA, Spiegelman D, & Wang Z (2024). Causal inference with double/debiased machine learning for evaluating the health effects of multiple mismeasured pollutants. arXiv preprint arXiv:2410.07135. [Google Scholar]
  52. Ye Z, Li X, Han Y, Wu Y, & Fang Y (2022). Association of long-term exposure to PM2. 5 with hypertension and diabetes among the middle-aged and elderly people in Chinese mainland: a spatial study. BMC Public Health, 22(1), 569. [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Zammit AR, Capuano AW, Barnes LL, Schneider JA, Sperling RA, Bennett DA, & Grodstein F (2025). Cognitive decline across five cognitive batteries: Sample size implications for clinical trials. Alzheimer’s & Dementia: Translational Research & Clinical Interventions, 11(3), e70137. [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Zhang B, Eum K-D, Szpiro AA, Zhang N, Hernández-Ramírez RU, Spiegelman D, Wang M, & Suh H (2025). Exposure measurement error in air pollution health effect studies: a pooled analysis of personal exposure validation studies in 17 communities across the United States. International Journal of Environmental Health Research, 1–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Zhang T, Liu W, Yang T, Zhai Y, Gu X, Xu L, Li F, Wu M, & Lin J (2024). Association between ambient fine particular matter components and subsequent cognitive impairment in community-dwelling older people: a prospective cohort study from eastern China. Aging clinical and experimental research, 36(1), 150. [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Zheng L, Su B, Cui F-P, Li D, Ma Y, Xing M, Tang L, Wang J, Tian Y, & Zheng X (2025). Long-Term Exposure to PM2. 5 Constituents, Genetic Susceptibility, and Incident Dementia: A Prospective Cohort Study among 0.2 Million Older Adults. Environmental science & technology, 59(9), 4493–4504. [DOI] [PubMed] [Google Scholar]
  57. Zhu Q, Liu Y, & Hasheminassab S (2024). Long-term source apportionment of PM2. 5 across the contiguous United States (2000-2019) using a multilinear engine model. Journal of Hazardous Materials, 472, 134550. [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
NIHMS2169618-supplement-1.docx (1.7MB, docx)

RESOURCES