The data used for this analysis were obtained from the UK Biobank (UKB). This prospective population-based cohort study involved over 500,000 individuals from various regions of the UK (Sudlow et al., 2015; Wu et al., 2021). The UKB study received approval from the Northwest Multicenter Research Ethics Committee, and all participants provided written informed consent (Sudlow et al., 2015; Yang et al., 2020). Our research project has been granted access to the UKB data (Application ID: 76636).

Exclusion criteria for our study encompassed missing records of air pollutant data and other baseline information, as well as individuals with a pre-existing diagnosis of all-cause dementia, AD, or vascular dementia. A flowchart illustrating the study population is presented in Figure 1.

A total of 437,932 participants were included in our primary analysis, while our sensitivity analysis comprised 298,995 participants. All participants underwent a complete case analysis.

PM_2.5 and PM₁₀ refer to particulate matter with diameters equal to or less than 2.5 and 10 μm, respectively. Measurements of air pollution levels for PM_2.5, PM₁₀, and NOx (nitrogen oxides) were obtained from the Small Area Health Statistics Unit, ¹ as part of the BioSHARE-EU Environmental Determinants of Health Project, ² in which the Biobank is a partner. The air pollution data were estimated based on the residential address of each participant at the scheduled follow-up date. The average concentration of air pollutants was determined using the 2010 European Study of Cohorts for Air Pollution Effects (ESCAPE) development and land use regression (LUR) model ³ (Eeftens et al., 2012). The LUR model for air pollutants in 2010 was developed by ESCAPE monitoring conducted between January 26, 2010, and January 18, 2011, representing the year 2010 (Huang et al., 2021; Parra et al., 2022).

The determination of outcomes for all-cause dementia, AD, and vascular dementia in the UKB database is based on algorithmically-defined criteria. ⁴ Information on death is obtained from the UKB’s death registration records. The most recent follow-up was conducted on March 23, 2021.

In this study, to minimize false positives, newly discovered UKB loci were not included. We selected 29 independent SNPs that exhibited significant associations with AD from previous GWASs (Marioni et al., 2018; Jansen et al., 2019; Leng et al., 2021). The list of selected SNPs can be found in Supplementary material in Supplementary Table S1. For each individual included in the UKB, we determined an AD genetic risk score (AD-GRS) based on a previously published study (Fan et al., 2020). The effect size (b-coefficient) of each SNP was calculated using the reported GWAS results (Jansen et al., 2019). Each participant was assigned one of three categories of genetic risk for AD: high (quintile 5), intermediate (quartiles 2–4), or low (quintile 1).

Baseline characteristics were compared between the control groups (non-dementia, non-AD, non-vascular dementia, and survivors) and different outcomes (all-cause dementia, AD, and vascular dementia) using appropriate statistical tests. The Chi-square test was used for categorical variables, the Student’s t-test was used for continuous variables with a normal distribution, and the Wilcoxon rank sum test was used for continuous variables without a normal distribution. Continuous variables were reported as either the mean with standard deviation or the median with interquartile range (IQR).

To assess the non-linear relationship between air pollution (PM_2.5, PM₁₀, and NOx) and various disease outcomes (all-cause dementia, AD, and vascular dementia), a restricted cubic spline (RCS) model with 4 knots located at the 5th, 35th, 65th, and 95th percentiles was used. The number and position of the knots were determined based on Harrell’s suggestion for adequate model fit (Li et al., 2022). The RCS model was adjusted for age, Townsend deprivation index (TDI), sex, smoking, ethnicity, education level, and alcohol use. Participants were divided into two groups based on their concentrations of PM_2.5 (PM_2.5 < 10 μg/m³, PM_2.5 ≥ 10 μg/m³), PM₁₀ (PM₁₀ < 15 μg/m³, PM₁₀ ≥ 15 μg/m³), and NOx (NOx < 50 μg/m³, NOx ≥ 50 μg/m³) according to the RCS curve. Kaplan–Meier survival curves were used to visualize the risk of developing disease (all-cause dementia, AD, and vascular dementia) among different air pollution groups (PM_2.5, PM₁₀, and NOx). The log-rank test was employed to assess differences between the air pollution groups.

A Cox proportional hazards model was utilized to estimate the hazard ratios (HR) of PM_2.5, PM₁₀, and NOx in relation to different outcomes (all-cause dementia, AD, and vascular dementia). The proportional hazards assumption was tested using the Schoenfeld residual method, further details of the process can be found in Supplementary material in Supplementary Figure S1. Three models were constructed: Model 1 was unadjusted, Model 2 was adjusted for age, TDI, sex, smoking, ethnicity, education level, and alcohol use. In addition, the IPTW technique harmonizes the distribution of propensity scores across various cohorts, neutralizing the influences of confounding variables (Neumann and Billionnet, 2016; Wen et al., 2022). To minimize the influence of confounding factors on the outcomes, we employed the COX model after IPTW to verify the association between air pollution and AD. The IPTW variables included age, TDI, sex, smoking, ethnicity, education level, and alcohol use.

A Cox proportional hazards model was employed to investigate the association between air pollution (PM_2.5, PM₁₀, and NOx) and AD-GRS with respect to AD risk. The main analysis involved the following steps: Step 1: Participants were categorized into nine groups based on different combinations of AD-GRS (low, intermediate, and high) and air pollution levels (PM_2.5, PM₁₀, and NOx). We selected the low-level air pollution combined with low AD-GRS as the reference group to examine the association of the other groups with AD risk. Step 2: If the results from Step 1 indicated a significant difference in AD risk between low-level and high-level air pollution within a specific AD-GRS subgroup, we designated the low-level air pollution in conjunction with the corresponding AD-GRS as the new reference group to assess the association of the remaining groups with AD risk. The multivariable model was adjusted for age, TDI, sex, smoking, ethnicity, education level, and alcohol use. Additionally, several sensitivity analyses were conducted: (1) adjusting for additional confounders such as marital separation/divorce status (yes/no), body mass index (BMI, kg/m³), healthy diet (yes/no), physical activity (low/moderate/high), and average total household income before tax (categorized as less than £18,000/£18,000 to £30,999/£31,000 to £51,999/£52,000 to £100,000/greater than £100,000); (2) excluding participants with a diagnosis of dementia or all-cause mortality within a two-year period; (3) applying IPTW to all variables to control for confounders. The healthy diet score was based on factors such as beef/mutton intake, vegetable intake, fruit intake, fish intake, cereal intake, and sodium concentration in urine. Assessment methods for a healthy diet have been widely utilized in previous studies (Ma et al., 2019). A Cox proportional was used to estimate the hazard ratios for AD after IPTW, and the IPTW variables included age, BMI, TDI, sex, smoking, ethnicity, education level, marital separation/divorce status, healthy diet, physical activity, household income, and alcohol use.

All statistical analyses were conducted using the R package (version 4.1.0). A value of p below 0.05 was considered statistically significant.

Our study included a total of 437,932 participants, with a median follow-up duration of 12.01 years. Table 1 presents the outcomes observed at the end of the follow-up period: 2088 participants developed all-cause dementia, 778 participants developed AD, 432 participants developed vascular dementia, and 29,132 participants passed away. There were statistically significant differences (p < 0.05) in baseline characteristics between the control group and the various outcome groups, including age, TDI, sex, smoking, ethnicity, education level, and alcohol use.

To investigate the non-linear relationship between air pollution (PM_2.5, PM₁₀, and NOx) and various outcomes (all-cause dementia, AD, and vascular dementia), an RCS analysis was conducted. Based on the RCS curve presented in Figure 2, participants were divided into two groups according to their concentrations of PM_2.5 (PM_2.5 < 10 μg/m³, PM_2.5 ≥ 10 μg/m³), PM₁₀ (PM₁₀ < 15 μg/m³, PM₁₀ ≥ 15 μg/m³), and NOx (NOx < 50 μg/m³, NOx ≥ 50 μg/m³). As illustrated in Figure 3, participants with PM_2.5 concentrations of ≥10 μg/m³ exhibited higher risks of developing all-cause dementia, AD, and vascular dementia compared to those in the PM_2.5 < 10 μg/m³ group. Additionally, participants in the PM₁₀ ≥ 15 μg/m³ group had a higher risk of developing all-cause dementia compared to the PM₁₀ < 15 μg/m³ group. Moreover, individuals exposed to NOx concentrations of ≥50 μg/m³ were associated with elevated risks of all-cause dementia and AD in comparison to those in the NOx < 50 μg/m³ group.

A Cox proportional hazards model was utilized to estimate the HR for all-cause dementia, AD, and vascular dementia. As presented in Table 2, after adjusting for multiple variables (Model 2), the analysis revealed that participants with PM_2.5 concentrations of ≥10 μg/m³ did not exhibit a significantly higher risk for AD or vascular dementia (p > 0.05). However, individuals in the PM_2.5 ≥ 10 μg/m³ group were associated with a higher risk of developing all-cause dementia (HR = 1.1; 95% CI: 1.05–1.28, p < 0.05). In comparison to the PM₁₀ < 15 μg/m³ group, there was no significant association observed between participants in the PM₁₀ ≥ 15 μg/m³ group and a higher risk for all-cause dementia, AD, or vascular dementia. Conversely, participants exposed to NOx concentrations of ≥50 μg/m³ were significantly associated with an increased risk of all-cause dementia (HR = 1.14; 95% CI: 1.02–1.26, p < 0.05) and AD (HR = 1.26; 95% CI: 1.08–1.48, p < 0.05) compared to the NOx < 50 μg/m³ group. Similar results were obtained when conducting an analysis using IPTW with multiple variables (Supplementary material in Supplementary Table S2).

A Cox proportional hazards model was employed to examine the joint association between air pollution (PM_2.5, PM₁₀, and NOx) and AD-GRS in relation to AD risk. The analysis did not reveal a significant interaction between air pollution and AD-GRS for AD risk (p > 0.05). As depicted in Figure 4, participants with PM₁₀ concentrations of ≥15 μg/m³ or PM_2.5 concentrations of ≥10 μg/m³ did not display a significantly higher risk of AD, regardless of their AD-GRS score, when compared to the PM₁₀ < 15 μg/m³ or PM_2.5 < 10 μg/m³ groups (p > 0.05). However, among participants with a high AD-GRS, those exposed to NOx concentrations of ≥50 μg/m³ exhibited a significantly elevated risk of AD in comparison to individuals in the NOx < 50 μg/m³ group (HR = 1.36; 95% CI: 1.03–1.18, p < 0.05; see Supplementary material in Supplementary Figure S2). Furthermore, participants with a high AD-GRS were significantly associated with a higher risk of AD regardless of the levels of air pollution (PM_2.5, PM₁₀, and NOx). Furthermore, the sensitivity analysis (Supplementary material in Supplementary Figure S3) showed similar results to Figure 4.