Ambient Air pollution and the Severity of Alzheimer’s Disease Neuropathology

Abstract

Importance:

Exposure to fine particulate matter air pollution (PM_2.5) may increase risk for dementia. It is unknown whether this relationship is mediated by dementia-related neuropathologic change found at autopsy.

Objective:

To examine relationships between PM_2.5 exposure, dementia severity, and dementia-associated neuropathologic change.

Design:

This cohort study used data associated with autopsy cases collected from 1999 to 2022 at the Center for Neurodegenerative Disease Research brain bank, University of Pennsylvania. Data were analyzed from January to June 2025.

Setting:

Tertiary referral center.

Participants:

602 cases with common forms of dementia and/or movement disorders and elderly controls were included after excluding 429 cases with missing data on neuropathologic measures, demographic factors, apolipoprotein E (APOE) genotype, or residential address.

Exposures:

One-year average PM_2.5 concentration prior to death or prior to last CDR-SB assessment was estimated using a spatiotemporal prediction model at residential addresses.

Main Outcomes and Measures:

Dementia severity was measured by Clinical Dementia Rating-Sum of Boxes (CDR-SB) scores. Ten dementia-associated neuropathologic measures representing Alzheimer’s disease, Lewy body disease, limbic-predominant age-related transactive response DNA-binding protein (TDP)-43 encephalopathy, and cerebrovascular disease were graded or staged. Linear, logistic and structural equation models were used to examine the relationships between PM_2.5, CDR-SB, and neuropathologic measures, adjusting for demographic factors and APOE ε4 allele status.

Results:

In a total of 602 autopsy cases (median [IQR] age at death 78 [71–85] years; 328 [54.5%] male), higher PM_2.5 exposure prior to death was associated with increased odds of more severe Alzheimer’s disease neuropathologic change (ADNC) (OR, 1.19; 95% CI, 1.11 to 1.28). In a subset of 287 cases with CDR-SB records (median [IQR] age at death 79 [72–86] years; 154 [53.7%] male), higher PM_2.5 exposure prior to CDR-SB assessment was associated with greater cognitive and functional impairment (β = 0.48; 95% CI, 0.22 to 0.74). Finally, 63% of the association between higher PM_2.5 exposure and greater cognitive and functional impairment was statistically mediated by ADNC (β = 0.30; 95% CI, 0.04 to 0.53).

Conclusions and Relevance:

PM_2.5 exposure was associated with increased dementia severity and increased ADNC. Population-based studies are needed to better understand this relationship.

Introduction

Air pollution has been suggested to be an environmental risk factor for dementia.^1–3 Studies have shown that exposure to air pollution, or more specifically fine particulate matter with aerodynamic diameter less than 2.5μm (PM_2.5), is associated with an increased incidence of dementia,⁴ impaired cognitive function,^5,6 and accelerated cognitive decline.⁷ Disease-modifying therapies that target the underlying neuropathology of Alzheimer’s disease (AD) show partial clinical benefit, but substantial economic and logistical barriers preclude widespread implementation, and serious side effects occur in some instances.^8,9 Given the social inequities with regard to air pollution exposure around the globe, better understanding of the relationship between air pollution exposure and risk for different subtypes of dementia may provide evidence that curbing PM_2.5 levels has the potential to benefit health outcomes and prevent dementia risk at the population level.

Despite growing evidence of the adverse effect of PM_2.5 on cognition,^10,11 the underlying biological mechanisms for this relationship are largely unknown. PM_2.5 exposure has been associated with brain volume loss,¹² brain atrophy,^13–15 worsening of cerebrospinal fluid amyloid-β₄₂ biomarker levels^16,17, increased amyloid PET positivity,¹⁸ and accelerated epigenetic aging,¹⁹ raising the possibility that PM_2.5 might contribute to dementia risk by enhancing AD neurodegenerative disease pathways.^20–22 Post-mortem autopsy examination remains the gold standard for the neuropathologic diagnosis of AD and related dementias. However, only two post-mortem studies have investigated the association between PM_2.5 exposure and AD neuropathology, but not other neurodegenerative disease pathology^23,24, and the tripartite relationship between PM_2.5, neurodegenerative disease pathology, and cognitive and functional outcomes has not been studied.

To address this knowledge gap, we studied here the relationships between PM_2.5 exposure, neuropathologic change, and cognitive and functional impairment in a large, well-characterized autopsy cohort. Using high spatial resolution estimates of PM_2.5 obtained from a validated prediction model,²⁵ we determined whether PM_2.5 exposure prior to death was associated with differences in the burden of the most common dementia-related neuropathologies including Alzheimer’s disease neuropathologic change (ADNC), Lewy body disease (LBD), limbic-predominant age-related transactive response DNA-binding protein (TDP)-43 encephalopathy neuropathologic change (LATE-NC), and cerebrovascular disease.^26,27 Next, we evaluated the relationship between PM_2.5 exposure, dementia severity measured by Clinical Dementia Rating Scale Sum of Boxes (CDR-SB) score, and dementia-related neuropathologic changes in our autopsy cohort. For this purpose, we explored whether PM_2.5 exposure prior to CDR-SB assessment was associated with worse CDR-SB scores and whether this relationship was mediated by neuropathologic change observed at autopsy.

Methods

Written informed consent was obtained from participants of neurodegenerative research programs with approval from the University of Pennsylvania Institutional Review Board, in addition to informed consent at time of death from next-of-kin prior to autopsy. This cohort study conforms to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.

Study participants

We obtained data associated with autopsy cases from the Integrated Neurodegenerative Disease (INDD) database at the Center for Neurodegeneration Disease Research (CNDR) at the University of Pennsylvania.²⁸ Cases were recruited mainly through the Penn Alzheimer’s Disease Research Center which focuses on dementia of the Alzheimer’s type (as opposed to vascular or other dementias) and the Penn Parkinson’s Disease and Movement Disorder Center which includes a research cohort focusing on cognitive dysfunction in the setting of LBD including dementia with Lewy bodies and Parkinson’s disease dementia. Details on the study cohort are provided in eMethods and eFigure 1 in Supplement. In brief, we restricted our autopsy cases to individuals older than the age of 40 with common neuropathological forms of dementia from 1999 through 2022. We excluded 429 cases due to missing data on neuropathology (n=9, 2.1%), APOE ε4 allele status (n=3, 0.7%), race (n=2, 0.5%), years of education (n=226, 52.8%), and/or residential address (n=179, 41.7%) resulting in a full cohort of 602 cases.

Air pollution exposure assessment

We used estimates of annual PM_2.5 exposure at a 0.01°×0.01° (approximately 1.1km × 1.1km) grid cell resolution from 1998 to 2022, from a publically available prediction model (V5.GL.04, Washington University, St. Louis).²⁵ We then spatially matched each case’s geocoded residential address before death to grid cells to obtain 1-year average PM_2.5 concentration either prior to death or prior to last CDR-SB assessment proximate to death as the primary exposure (eMethods in Supplement).

Neuropathological assessment

Ten-neuropathologic measures, representing four proteinopathies (tau, β-amyloid, α-synuclein, and TDP-43) and three cerebrovascular lesions (infarcts, amyloid angiopathy, arteriolosclerosis), were obtained from the INDD database and used to assess neuropathologic outcomes as follows: Thal amyloid phase, Braak stage, CERAD score, level of ADNC,²⁹ absence/presence of LBD,³⁰ LATE-NC stage,²⁷ absence/presence of large infarcts anywhere in the brain based on gross and microscopic examination, absence/presence of moderate to high occipital cerebral amyloid angiopathy burden, absence/presence of moderate to high arteriolosclerosis in occipital white matter, and likelihood that cerebrovascular pathology contributed to cognitive impairment (VCING).³¹. All assessments were based on consensus criteria.^27,29–31 Further details are described in eMethods in Supplement.

Clinical assessment

The INDD records on clinician-reported Clinical Dementia Rating-Sum of Boxes (CDR-SB) were used as the clinical outcome. The CDR-SB is a comprehensive measure of dementia severity across six cognitive and functional domains including memory, orientation, judgment, community affairs, home hobbies, and personal care, and is scored ranging from 0 to 18, with higher scores indicating more severe dementia.³²

Statistical analysis

Analyses were performed with R version 4.2.3. All statistical tests were two-sided. Diagnostic and assumption tests were performed for fitted models, if appropriate (eMethods in Supplement).

For the full autopsy cohort, logistic regression was used to estimate the association between 1-year average PM_2.5 exposure before death and the odds of ordered categorical neuropathologic outcomes that included Thal amyloid phase, Braak stage, CERAD score, ADNC level, LATE-NC stage, and VCING likelihood level, and the odds of dichotomous neuropathologic outcomes that included the presence/absence of LBD, large infarcts, occipital lobe cerebral amyloid angiopathy, and occipital white matter arteriolosclerosis. All models were adjusted for sex, age at death, APOE ε4 allele status, race, and years of education.

For the subset of cases where their last CDR-SB was evaluated within 5 years of death (CDR cohort), we estimated the association between 1-year average PM_2.5 exposure before last CDR assessment and last CDR-SB score prior to death representing dementia severity at death using linear regression and neuropathologic measures using the same statistical models and neuropathologic outcomes described for the full autopsy cohort. All models included the same covariates used for the full autopsy cohort.

For the subset of cases where multiple CDR-SB records were available (longitudinal CDR cohort), we used a linear mixed-effects model to estimate the association between 1-year average PM_2.5 concentration before each CDR assessment and longitudinal change in CDR-SB scores, with an interaction term between PM_2.5 exposure and interval years from initial CDR-SB assessment to each follow-up assessment as the main effect using case-specific intercepts and slopes as random factors. Covariates were the same as those described for other cohorts.

To explore the adjusted tripartite relationship between PM_2.5 exposure, CDR-SB, and neuropathologic change, we performed a mediation analysis with structural equation modelling (SEM) using the R “lavaan” package.³³ In the SEM, we estimated direct and indirect effects of 1-year average PM_2.5 exposure before last CDR-SB assessment on last CDR-SB score prior to death, accounting for ADNC as a potential endogenous mediator, using a bias-corrected, accelerated bootstrap confidence interval method with 2000 simulations.

Post-hoc analyses

In a secondary analysis exploring effect modification, we added an interaction term for APOE ε4 allele status in main regression models. In a series of sensitivity analyses, we determined the effect of uncertainty in the PM_2.5 prediction model (eMethods in Supplement). Furthermore, we applied additional exposure time windows corresponding to 2-year, 3-year, and 4-year average PM_2.5 exposure before death or before last CDR assessment prior to death, included an additional covariate of area-level socioeconomic status (SES) (eMethods in Supplement), and also applied different estimated PM_2.5 values from another well-validated prediction model (eMethods in Supplement).

Results

Cohort characteristics

Demographic, clinical, and genetic features of the cohort are provided in Table 1. Within the full autopsy cohort of 602 cases, 328 (54.5%) were male, the median (IQR) age of death was 78 (71–85) years, most of the subjects were white (94.4%) and non-Hispanic or Latino (99%), and 53.2% carried at least one ε4 allele of the APOE which is the strongest genetic risk factor for sporadic AD. Overall, our cohort was highly educated with a median (IQR) of 16 (12–18) years of education, with a moderate level of dementia corresponding to a median (IQR) CDR-SB score of 12 (5–17) at last assessment. The most common clinical diagnoses were AD (47.3%) and Parkinson’s disease with dementia (14.1%).

Table 1.

Demographic and clinical characteristics of study participants.

Characteristic	Overall, No. (%)
No. of participants	602
Sex
Female	274 (43.5)
Male	328 (54.5)
Age at death, median (IQR), y	78 (71–85)
Race
Non-White	34 (5.6)
White	568 (94.4)
Ethnicitya
Hispanic or Latino	589 (99.0)
Not Hispanic or Latino	6 (1.0)
APOE ε4 allele status
APOE ε4 (+)	320 (53.2)
ε2/ε4	11 (1.8)
ε3/ε4	228 (37.9)
ε4/ε4	81 (13.5)
APOE ε4 (−)	282 (46.8)
ε2/ε2	2 (0.3)
ε2/ε3	39 (6.5)
ε3/ε3	241 (40.0)
Education, median (IQR), y	16 (12–18)
CDR-SB at last assessment within 5 years of death, median (IQR), yb	12 (5–17)
Age at last CDR-SB assessment, median (IQR), yb	78 (71–85)
Interval between last CDR-SB assessment and death, median (IQR), yb	1 (1–3)
Annual change in CDR-SB, median (IQR), yc	1.4 (0.4–2.2)
Interval between initial and last CDR-SB assessment, median (IQR), yc	4 (2–7)
Primary clinical diagnosis
Behavioral variant frontotemporal dementia	16 (2.7)
Corticobasal syndrome	26 (4.3)
Dementia, uncertain etiology	7 (1.2)
Dementia with Lewy bodies	51 (8.5)
Frontotemporal dementia, not otherwise specified	16 (2.7)
Hydrocephalus	1 (0.2)
Limbic-predominant age-related TDP-43 encephalopathy	1 (0.2)
Logopenic primary progressive aphasia	14 (2.3)
Mild cognitive impairment	12 (2.0)
Neurologically normal	28 (4.7)
Parkinson’s disease with dementia	85 (14.1)
Parkinson’s disease without dementia	37 (6.1)
Possible/Probable Alzheimer’s disease	285 (47.3)
Posterior cortical atrophy	6 (1.0)
Progressive non-fluent aphasia	4 (0.7)
Progressive supranuclear palsy	1 (0.2)
Semantic variant primary progressive aphasia	4 (0.7)
Vascular dementia	8 (1.3)

Abbreviations: APOE, Apolipoprotein E; CDR-SB, Clinical Dementia Rating Scale-Sum of Boxes, TDP-43, transactive response DNA-binding protein 43.

^aA total of 7 (1.2%) participants had missing data.

^bScores on the Clinical Dementia Rating–Sum of Boxes (CDR-SB) at last assessment prior to death range from 0 to 18, with higher scores indicating greater cognitive and functional impairment. A total of 287 participants were included after excluding 315 (52.3%) cases with missing data on CDR-SB scores within 5 years of death.

^cAnnual change in CDR-SB score was assessed for a period between initial to last CDR assessment, with higher values indicating faster cognitive and functional impairment. A total of 261 participants were included after excluding 341 (56.6%) cases who did not have multiple, annual records on CDR-SB scores.

Neuropathological features of the cohort are presented in Table 2. More than a half of the cohort exhibited severe AD neuropathologic measures including Thal amyloid phase 4/5 (71.3%), Braak stage V/VI (63.8%), CERAD score 3 (64.3%), and high overall ADNC (62.3%). 38.2% exhibited transitional/limbic or diffuse/neocortical LBD, and 34.4% exhibited LATE-NC. Notably, very few cases exhibited large infarcts (5.3%), moderate to severe occipital lobe cerebral amyloid angiopathy (31.1%), moderate to severe occipital white matter arteriolosclerosis burden (16.8%), or overall high likelihood that cerebrovascular pathology contributed to cognitive impairment (2.7%).

Table 2.

Neuropathologic characteristics of study participants

Characteristica	Overall, No. (%)
No. of Participants	602
Thai amyloid phase
0	41 (6.8)
1/2	63 (10.5)
3	69 (11.5)
4/5	429 (71.3)
Braak NFT stage
0	14 (2.3)
I/II	101 (16.8)
III/IV	103 (17.1)
V/VI	384 (63.8)
CERAD score
0	99 (16.4)
1	42 (7.0)
2	74 (12.3)
3	387 (64.3)
ADNC level
Not	40 (6.6)
Low	101 (16.8)
Intermediate	86 (14.3)
High	375 (62.3)
Lewy pathology
No	209 (37.4)
Brainstem predominant	54 (9.0)
Transitional or limbic predominant	106 (17.6)
Diffuse or neocortical predominant	124 (20.6)
Amygdala	109 (18.1)
LATE
No	395 (65.6)
LATE-NC stage 1	35 (5.8)
LATE-NC stage 2	147 (24.4)
LATE-NC stage 3	25 (4.2)
Large infarct	NA
No	570 (94.7)
Yes	32 (5.3)
Cerebral amyloid angiopathy burden in occipital lobe
Absent to mild	415 (68.9)
Moderate to severe	187 (31.1)
Arteriolosclerosis burden in occipital white matter
Absent to mild	501 (83.2)
Moderate to severe	101 (16.8)
VCING level
Low	535 (88.9)
Intermediate	51 (8.5)
High	16 (2.7)

Abbreviation: NFT, neurofibrillary tangle; CERAD, Consortium to Establish a Registry for Alzheimer’s Disease; ADNC, Alzheimer’s disease neuropathologic change; LATE, limbic-predominant age-related transactive response DNA-binding protein (TDP)-43 encephalopathy; VCING, vascular cognitive impairment neuropathology guidelines.

^aAll pathological measures were assessed using consensus criteria.

Air pollution and neuropathologic change in autopsy cases

Our cases were geographically distributed across eleven states (Pennsylvania, New Jersey, Delaware, New York, Maryland, Virginia, West Virginia, Connecticut, Ohio, Colorado, California) with the majority of cases from Pennsylvania (72.9%) (eFigure 2 in Supplement). The median (IQR) 1-year average PM_2.5 concentration prior to death was 9.4 (8.1–12.4) μg/m³.

In adjusted models, higher PM_2.5 exposure was associated with increased odds of more severe AD neuropathology (Figure 1 and eTable 1 in Supplement). Specifically, for every 1 μg/m³ increase in 1-year average PM_2.5 exposure before death, there were 17%, 20%, 20%, and 19% increases in the odds of higher Thal amyloid phase (odds ratio [OR],1.17; 95% confidence interval [CI], 1.08 to 1.27; P<.001), Braak stage (OR, 1.20; 95% CI, 1.11 to 1.29; P<.001), CERAD score (OR, 1.20; 95% CI, 1.11 to 1.30; P<.001), and overall level of ADNC (OR, 1.19; 95% CI, 1.11 to 1.28; P<.001), respectively. Furthermore, for every 1 μg/m³ increase in 1-year average PM_2.5 exposure, there was a 16% increase in the odds of having large infarcts (OR,1.16; 95% CI, 1.01 to 1.32; P=.04). In contrast, higher PM_2.5 was not associated with increased odds of LBD stage (OR, 0.98; 95% CI, 0.91 to 1.04; P=.47), LATE-NC stage (OR, 1.03; 95% CI, 0.96 to 1.10; P=.46), and other cerebrovascular lesions including occipital lobe cerebral amyloid angiopathy burden (OR, 1.00; 95% CI, 0.93 to 1.07; P>.99), occipital white matter arteriolosclerosis burden (OR, 1.01; 95% CI, 0.92 to 1.10; P=.89), and overall VCING level (OR, 1.08; 95% CI, 0.97 to 1.19; P=.15).

Figure 1. PM_2.5 exposure and dementia-related neuropathologies in the full autopsy cohort.

Abbreviations: OR, odds ratio; CI, confidence interval; CERAD, Consortium to Establish a Registry for Alzheimer’s Disease; ADNC, Alzheimer’s disease neuropathologic change; LBD, Lewy body disease; LATE, limbic-predominant age-related transactive response DNA-binding protein (TDP)-43 encephalopathy; VCING, vascular cognitive impairment neuropathology guidelines.

Odds ratios (ORs) and 95% confidence intervals (CIs) are reported for each neuropathologic outcome corresponding to every 1 μg/m³ increase in 1-year average PM_2.5 exposure before death from a series of ordinal logistic regression models with ordinal Thal amyloid phase, ordinal Braak stage, ordinal CERAD score, ordinal ADNC level, ordinal LATE-NC stage, and ordinal VCING score as outcome variables, and binary logistic regression models with dichotomously treated LBD stage, presence of large infarcts, presence of occipital cerebral amyloid angiopathy, and presence of arteriolosclerosis as outcome variables in the full autopsy cohort (n=602). All models were controlled for sex, age at death, race, APOE ε4 status, and years of education. ORs and 95% CIs greater than 1 indicate worse neuropathologic outcomes. P values less than 0.05 were considered statistically significant.

Air pollution, neuropathologic change, and dementia in the subset of autopsy cases with Clinical Dementia Recording-Sum of Boxes (CDR-SB) scores

To explore the relationship between PM_2.5 exposure, neuropathologic change, and clinical dementia severity, we restricted the study cohort to 278 autopsy cases with CDR-SB scores within 5 years of death. The median (IQR) age of death was 79 (72–86) years and 154 (53.7%) were male. Overall, this CDR cohort was not different from the full cohort in terms of demographic and neuropathologic characteristics (eTables 2 and 3 in Supplement).

In adjusted models, each 1 μg/m³ increase in 1-year average PM_2.5 concentration before last CDR-SB assessment prior to death was associated with a 0.48 point increase in the last CDR-SB score (β = 0.48; 95% CI, 0.22 to 0.74; P<.001; Table 3) indicating greater cognitive and functional impairment. Although the requirement for CDR-SB clinical assessment resulted in a smaller cohort, the odds ratios for AD neuropathologic outcomes were similar to those observed in the full autopsy cohort with an overall increase in the odds of more severe AD neuropathology (Thal amyloid phase; OR, 1.23; 95% CI, 1.09 to 1.40; P=.001, Braak stage; OR, 1.25; 95% CI, 1.12 to 1.41; P<.001, CERAD score; OR, 1.26; 95% CI, 1.13 to 1.42; P<.001, ADNC level; OR, 1.25; 95% CI, 1.12 to 1.40; P<.001; eFigure3 and eTable 4 in Supplement).

Table 3.

PM_2.5 exposure and Clinical Dementia Rating-Sum of Boxes

Variablea	β (95% CI)b	P valuec
PM2.5	0.48 (0.22, 0.74)	<.001
Sex, Male	−0.72 (−2.13, 0.68)	.31
Age at CDR	−0.08 (−0.14, −0.02)	.01
APOE ε4 (+)	2.94 (1.57, 4.31)	<.001
Race, White	−1.71 (−4.53, 1.10)	.23
Education	−0.31 (−0.54, −0.08)	.009

Abbreviations: CI, confidence interval; CDR, Clinical Dementia Rating Scale; APOE, Apolipoprotein E.

^aAssociations were evaluated by a linear regression model in which the outcome variable was Clinical Dementia Rating-Sum of Boxes (CDR-SB) score at last assessment with higher values indicating greater cognitive and functional impairment, the exposure variable was 1-year average PM2.5 exposure before death, and covariates were sex, age at last CDR-SB assessment, APOE ε4 allele status, race, and years of education.

^bEstimated effects (β) and 95% confidence intervals (CIs) are shown as adjusted CDR-SB score at last assessment per 1 μg/m³ increase in 1-year average PM_2.5 exposure before last CDR-SB assessment, with positive values indicating greater cognitive and functional impairment. The CDR cohort of 287 cases who had their last CDR evaluation within 5 years of death was analyzed.

^cP values less than 0.05 were considered statistically significant.

In an adjusted longitudinal model with 261 cases who had multiple CDR-SB records, every 1 μg/m³ increase in 1-year average exposure to PM_2.5 before each CDR assessment was associated with a 0.07 point increase in annual change of CDR-SB score (β = 0.07; 95% CI, 0.04 to 0.09; P<.001; eTable 5 in Supplement) indicative of faster cognitive and functional decline.

Furthermore, we examined whether the adjusted association between higher PM_2.5 exposure and more severe dementia indicated by higher CDR-SB was mediated by ADNC (Figure 2). Using SEM mediation analysis, the estimated total effect of 1-year average PM_2.5 concentration before last CDR-SB was 0.48 (β = 0.48; 95% CI, 0.23 to 0.72) and the estimated indirect effect through increasing the odds of higher ADNC levels was 0.30 (β = 0.30; 95% CI, 0.04 to 0.53), where 63% of the estimated association between PM_2.5 exposure and dementia severity was mediated by ADNC.

Figure 2. Alzheimer’s disease neuropathologic change mediates the relationship between PM_2.5 exposure and Clinical Dementia Rating–Sum of Boxes.

Abbreviations: CI, confidence interval; ADNC, Alzheimer’s disease neuropathologic change; CDR-SB, Clinical Dementia Rating Scale-Sum of Boxes.

A path diagram of the structural equation modelling shows relationships among PM_2.5 exposure, ADNC, and CDR-SB with 1-year average PM_2.5 exposure before last CDR assessment prior to death as the exposure variable, ordinal ADNC level as the potential mediator, and CDR-SB at last assessment prior to death as the outcome variable in the CDR cohort (n=287), after adjusting for sex, age at CDR, race, years of education, and APOE ε4 status as confounding variables. In this model, the indirect effect is the effect of PM_2.5 exposure on CDR-SB through ADNC and the direct effect is the effect of PM_2.5 exposure on CDR-SB that is independent from ADNC. β estimates of indirect, direct, and total effects with 95% confidence intervals (CIs) were calculated with SEM using a bias-corrected and accelerated bootstrap confidence interval method with 2000 simulations. Positive values indicate greater cognitive and functional impairment. This model exhibited good fit indices as follows: comparative fit index [CFI], 0.98; root mean square error of approximation [RMSEA], 0.02; Tucker-Lewis index [TLI], >0.99).

Diagnostic and assumption tests confirmed both linearity and no spatial autocorrelation for the observed associations in regression models (eTable 6 in Supplement) and the SEM model exhibited good model fit indices (Figure 2).

Post-hoc analyses

Secondary analyses revealed that APOE ε4 allele status was not an effect modifier for both neuropathologic (eTables 7 and 9 in Supplement) and cognitive outcomes (eTable 8 in Supplement). Sensitivity analyses confirmed the robustness of the association between higher PM_2.5 and worse AD neuropathologic outcomes in the face of uncertainty in predicted PM_2.5 values (eFigure 4 in Supplement). Furthermore, the observed associations remained when applying longer exposure time frames (eFigures 5 and 6 and eTables 10 and 11 in Supplement), another well-validated PM_2.5 prediction model (eTable 12 in Supplement), and an additional covariate of median household income as a measure of socioeconomic status (eTables 13, 14, 15, 16, and 17 in Supplement).

Discussion

In this relatively large autopsy series, we found that higher PM_2.5 concentrations were strongly associated with more severe amyloid and tau pathologies, culminating in more advanced overall ADNC. In addition, exposure to higher PM_2.5 levels was also associated with significantly greater and faster cognitive and functional impairment. These associations were not modified by APOE ε4 allele status. Finally, the impact of PM_2.5 on dementia severity was largely mediated by an increase in ADNC. This suggests that PM_2.5 may directly affect brain vulnerability where increased ADNC appears to mediate PM_2.5-induced cognitive dysfunction.

The relationship between PM_2.5 exposure and neurodegenerative pathology is largely unknown. Two prior studies examining the effects of PM_2.5 on AD pathology have been reported, one of which demonstrated an association between PM_2.5 exposure and higher CERAD score but no other AD pathology based on a study of an Alzheimer’s Disease Research Center autopsy cohort.²⁴ In contrast, no significant effect of PM_2.5 exposure and amyloid, tau or overall ADNC was observed in another study of a community-based autopsy cohort study.²³ In contrast with these previous studies, we observed that PM_2.5 exposure before death was significantly associated with an overall increased odds ratio for more severe AD neuropathology including higher Thal amyloid phase, Braak stage, CERAD score, and ADNC level. Inconsistencies between studies may be related to differences in autopsy cohort characteristics and exposure measurements. While the study of the community-based cohort is likely to have more heterogenous causes of cognitive dysfunction,²³ our cohort consisted mainly of symptomatic dementia cases that were enriched for AD dementia, perhaps allowing for more homogeneity and therefore statistical power to detect Alzheimer’s disease specific effects. In addition, when compared with the prior study of the research-based autopsy cohort,¹⁶ our study included a relatively large sample size and estimated total PM_2.5 derived from different emission sources, perhaps enhancing the statistical power to detect associations between PM_2.5 and neuropathologic outcomes.

Our study builds upon prior studies in that cognitive and functional data was included for a subset of individuals in order to evaluate the impact of PM_2.5 exposure on dementia severity at death in relation to neuropathologic changes. Considering the temporal ordering between exposure and outcomes variables, when evaluating dementia severity, we applied 1-year average PM_2.5 concentrations before last CDR-SB assessment proximate to death. In doing so, our findings are consistent with prior population-based cohort studies which have demonstrated an association between PM_2.5 exposure and worse cognitive outcomes.^5,13 Furthermore, the association of PM_2.5 exposure with AD pathology but not other pathologies suggests that there may be some specificity with regards to the effects of PM_2.5 exposure on CNS pathology. This is consistent with antemortem human studies which tend to show that PM_2.5 exposure is associated with increased AD biomarkers including structural MRI changes, cerebrospinal fluid measures,¹⁶ and amyloid PET outcomes.^18,34–36 Notably, we found that the relationship between PM_2.5 exposure and cognitive and functional dysfunction was largely mediated by increased ADNC. This supports the hypothesis that PM_2.5 exposure is deleterious for brain maintenance pathways, resulting in worsening of ADNC.^37,38

One strength of this study is the comprehensive study of a large autopsy series with well-defined demographical, clinical, and neuropathological profiles based on consensus criteria. Furthermore, we demonstrated the robustness of our findings with several sensitivity analyses accounting for PM_2.5 exposure uncertainty estimates, different exposure time frames, and area-level socioeconomic status.

Despite these strengths, we note the following limitations. First, our clinical research-oriented autopsy cohort is skewed demographically. Indeed, the vast majority of subjects were white, non-Hispanic or Latino, highly educated with a higher than college degree, and/or from less disadvantaged neighborhoods. This biased, unrepresentative sample may limit the generalizability of our findings. Moreover, compared to a population-based study cohort where a diverse spectrum of cognitive and neuropathologic conditions are found,³⁹ our autopsy cohort included a large proportion of cases through research programs that were enriched for AD dementia and that did not enroll individuals with vascular dementia, which may induce selection bias. Due to the rarity of cerebrovascular pathology in our cohort, we likely underestimated the true associations between PM_2.5 and cerebrovascular disease which have been strongly supported by epidemiologic studies^40,41. Moreover, our study included a small number of cases when compared to most epidemiologic studies, precluding a more robust examination of relationships among variables with strong statistical power. Thus, studies of large population-based autopsy cohorts are required in order to extrapolate these findings to the general population.

Second, we excluded many cases with missing values due to incomplete assessment. While overall characteristics did not differ between the full versus CDR cohorts, there was a small difference in the mean age at death for those excluded from this study. However, the mean difference of only one or two years is unlikely to be clinically or biologically meaningful. In addition, PM_2.5 exposure is associated with increased mortality, and so there is a potential for survivorship bias in this study. Methods such as inverse probability weighting (IPW) may be considered to correct for selection bias such as the impact of conditioning on death and consent to autopsy.

Third, PM_2.5 was measured only at each case’s last residential address. Moving to a new address was not common in our cohort with less than 1% of individuals changing their address within the last year of life. However, longitudinal PM_2.5 exposure values including the use of more accurate PM_2.5 self-monitors would add more precision to future studies.

Finally, we have not comprehensively evaluated other potential confounding factors such as physical and leisure activity, smoking and alcohol history, medical treatment history, neighborhood greenness, differences between urban and rural exposures (as this cohort was 100% urban), and other air pollutants such as NO₂ or ozone that may affect downstream neuropathologic change and cognition. Furthermore, due to the nature of an observational cohort study, we were not able to uncover the mechanisms underlying our findings which may require experimental and/or interventional studies to uncover. Considering these limitations, replication in large population-based cohorts and mechanistic studies are warranted.

Conclusions

This autopsy cohort study reinforces the finding that PM_2.5 exposure appears to negatively affect cognitive function and suggests that this relationship may be mediated by ADNC. Our findings suggest that PM_2.5 exposure may exacerbate AD pathogenesis. Population-based autopsy studies are further needed to replicate our findings and better understand relationships among PM_2.5 exposure, cognition, and neuropathology.

Key points.

Question:

What are the relationships between air pollution, neuropathology, and dementia?

Findings:

In this cohort study of 602 autopsy-confirmed individuals, exposure to higher levels of fine particulate matter air pollution (PM_2.5) was associated with more advanced Alzheimer’s disease neuropathologic change (ADNC) and more advanced clinical measures of dementia. The association between PM_2.5 exposure and clinical dementia severity appeared to be statistically mediated by ADNC.

Meaning:

Higher PM_2.5 exposure may exacerbate Alzheimer’s disease neuropathologic change and cognitive dysfunction in the setting of dementia. Population-based autopsy studies are further needed to generalize these findings.