Association of long-term exposure to various ambient air pollutants, lifestyle, and genetic predisposition with incident cognitive impairment and dementia | BMC Public Health

Association of long-term exposure to various ambient air pollutants, lifestyle, and genetic predisposition with incident cognitive impairment and dementia | BMC Public Health

Study population

UK Biobank, as a large population-based, nationwide, and open-access prospective study, recruited over 500,000 individuals in conjunction with 22 assessment centres across the UK. Through self-completed touch-screen questionnaires, computer-assisted interviews, physical and functional measurements, and samples of blood, urine, and saliva, it successfully collected a large variety of health-related information, consisting of sociodemographic characteristics, diseases phenotypic, lifestyle, and genetic variants [8, 9]. More details of the study design and methodology have been thoroughly described by former researchers [8]. UK Biobank is operating under the approval of North-West Multicentre Research Ethics Committee to ensure its ethical robustness. All participants provided their consent for regular blood, urine, and saliva sampling and more accurate data on their lifestyles. Furthermore, the dataset also promised to be anonymized to protect the privacy of participants.

In the present study, the inclusion and exclusion criteria were set as follows. Inclusion: (1) participants who gave written consent to participate; (2) participants who have completed the follow-up. Exclusion: (1) participants who had a history of cognitive impairment or dementia on the basis of self-report or medical records at the baseline visit; (2) participants with missing air pollution exposure information; (3) participants with missing genetic data; (4) participants with missing lifestyle information. For the duration of follow-up, according to the UK Biobank, the recruitment of participants was carried out between 2006 and 2010, while the end date of follow-up was December 31, 2019. As a result, we excluded those with MCI or any form of dementia at the baseline. After the screening, a total of 502,149 participants were included for consideration. However, 41,277 of them were found to lack the necessary information about air pollution exposure and were eventually excluded, leaving a total of 460,872 eligible participants for the final analysis. The flow chart of study participants was displayed in Supplementary Fig. S1.

Assessment of outcomes

All patients were diagnosed in accordance with the criteria of the International Classification of Diseases, 10th revision (ICD-10). The ICD-10 codes of all-cause dementia included G30, F01, G20. The ICD-10 codes of Alzheimer’s dementia, vascular dementia, and MCI were set as G30, F01, and F06.7 respectively. More detailed information was shown in Supplementary Table S1.

Air pollution score (APS)

With its ability to consider different types of land-use variables in the assessment of target pollutants, land-use regression has now become an effective means to depict intra-urban air pollution concentration variation in fine spatial-temporal resolution globally [10]. Therefore, the UK Biobank Study adopted a land-use regression model based on the European Study of Cohorts for Air Pollution Effects (ESCAPE) project [11, 12] to estimate the annual average concentrations of PM2.5, PM10, PM2.5−10, NO2, and NO, which can be found at https://biobank.ndph.ox.ac.uk/showcase/label.cgi?id=114. While the land-use regression models calculate the annual moving average concentrations of air pollutants using the predictor variables retrieved from the GIS variables, including land use, traffic, and topography by a 100 m × 100 m resolution. Participants’ ambient air pollution concentrations were then assigned according to their residential coordinates in the 100 m × 100 m grid cells. The exposure levels of five air pollutants that mentioned above were all collected in 2010.

In order to further assess the combined exposure of five different ambient air pollutants, we calculated an APS [13] by summarizing the concentrations of PM2.5, PM10, PM2.5−10, NO2, and NO, weighted by the multivariable-adjusted risk estimates (β coefficients) on cognitive impairment in the current study. The equation was:

$$\beginarrayl\textAir pollution score\\ = (\beta _PM2.5 \times PM_2.5 + \beta _PM10 \times PM_10 + \beta _PM2.5 – 10 \times PM_2.5 – 10 + \beta _NO_2 \times NO_2\\ \quad + \beta _NO \times NO) \times (5 \div \textsum of the \beta \,\textcoefficients)\endarray$$

The APS ranged from 39.22 to 157.77. A greater score indicated a higher level of combined exposure to various air pollutants. All participants were categorized as five groups according to the quintiles of the APS level.

Evaluation of genetic risk

The genetic association of dementia has been widely confirmed by evidence from mounting publications based on genome-wide association study (GWAS) [14,15,16,17,18,19]. In the UK Biobank project, research team conducted the genotyping, imputation, and quality control of the genetic data, providing a critical route to further investigation into the heredity-related dementia. More specific descriptions were available elsewhere [20].

According to the previous studies [21, 22], the expression of polymorphisms in the apolipoprotein E gene (APOE) was recognized as a strong genetic risk factor for Alzheimer’s dementia and several other neurodegenerative diseases, including vascular dementia and Lewy body dementia. Therefore, in the present study, the APOE status recorded in the genetic database of UK Biobank were fully utilized by researchers to evaluate the genetic risk for cognitive impairment among participants. The population was divided into three groups of low, intermediate, and high genetic risk of cognitive impairment based on their APOE gene carrying status for the convenience of subsequent statistical analysis.

Healthy lifestyle score (HLS)

A healthy lifestyle score (HLS) was generated based on 7 variables: physical activity, body mass index (BMI), alcohol consumption, smoke status, waist-to-hip ratio (WHR), sedentary time (hours/day), and vegetable and fruit intake (servings/day).

The BMI was calculated as weight divided by height squared (kg/m2). On the basis of multiple meta-analyses [23,24,25,26,27] on BMI-related all-cause mortality, healthy weight was defined as the BMI values in a normal range (18.5 ~ 24.9); As regards the physical activity estimation of an individual, the International Physical Activity Questionnaire (IPAQ) guideline [28, 29] was adopted for metabolic equivalent task (MET) calculation. A physical activity guideline [30] was also used to determine whether the MET values were appropriate for the benefits of participants’ health; The WHR was defined as waist circumference (WC) divided by hip circumference (HC), which was a strong indicator of central obesity. According to the recommendation from IPAQ guideline [28], the WHR was identified as healthy when it was < 0.85 for females, and < 0.90 for males; The time spent using the computer and watching television for recreational purposes was calculated as sedentary time. Participants were classified as unhealthy if they failed to keep the daily sedentary time less than 3 h [31]; The participants reported their dietary arrangements at the baseline visit for the calculation of the total fruits and vegetables intake, with ≥ 6 servings/day categorized as healthy diet [32]; In terms of alcohol and tobacco consumption, the population was divided into three groups of never, previous, and current, among which that currently smoking or drinking was identified as unhealthy living habit.

Participants scored 1 point for each of these health-related behaviours once they met the criteria mentioned above. The HLS ranged from 0 to 7 theoretically. After the scoring was completed, participants were split up into three groups as unfavourable (0, 1), intermediate (2, 3), and favourable (≥ 4) in accordance with their HLSs.

Measurements of other potential covariates

Research team collected age, sex, ethnicity, Townsend deprivation index (TDI) [33], blood pressure level, employment status, education background, income bracket, and history of hypertension, diabetes, cardiovascular disease (CVD), coronary artery disease (CAD), and stroke as potential modification factors.

During the initial baseline visit, both systolic blood pressure (SBP) and diastolic blood pressure (DBP) were measured by trained clinical workers to ascertain the blood pressure level. In addition, the income bracket evaluation was based on the average total household income (<£18,000, £18,000~£52,000, £52,000~£100,000, and >£100,000). The data on education duration was used for the assessment of education background (≤ 7 years, 8 ~ 10 years, 11 ~ 15 years, and ≥ 16 years).

Statistical analysis

The follow-up time was measured from the recruitment date to the first diagnosis of any form of cognitive impairment or dementia, lost to follow-up, death, or end of the current follow-up, whichever came first. Our research team adopted Cox proportional hazards models to evaluate the hazard ratio (HR) and 95% confidence interval (CI) for the incident cognitive impairment and dementia related to single air pollutants and the APS. Other potential confounders, including age (continuous), sex (male, female), ethnicity (white, non-white), SBP (continuous), BMI (kg/m2, continuous), employment status (yes, no), physical activity (MET, continuous), education background (≤ 7 years, 8 ~ 10 years, 11 ~ 15 years, and ≥ 16 years), income bracket (<£18,000, £18,000~£52,000, £52,000~£100,000, and >£100,000), alcohol consumption status (never, previous, and current), tobacco consumption status (never, previous, and current), hypertension history (yes, no), diabetes history (yes, no), and CVD history (yes, no) were adjusted in those models that mentioned above.

Several sensitivity analyses were also performed to test the robustness of our outcomes. First, in order to reduce the impact of the missed diagnosis at the baseline visit on the effectiveness of analysis as much as possible, researchers excluded those cases which reported the cognitive impairment diagnosis in the first 2 years of follow-up. Second, an analysis of participants who have lived in the current address for at least 5 years was also conducted by the research team to estimate the long-term effect of air pollution exposure on cognitive impairment. Finally, researchers brought other mixed covariates, for example, age, sex, ethnicity, and so on, into consideration to minimize their potential influences during the follow-up.

Besides, given that not all participants diagnosed with cognitive impairment or dementia were necessarily attributed to air pollution exposure, our research team adopted Levin’s formula [34] to estimate the proportion of patients that could be prevented if the risk factor was eliminated. In the current study, the HRs were used as the risk ratios (RRs) for the calculation [35].

$$ Population\ Attributable\ Fraction \left(PAF\right)=\fracP_e\times \left(RR_e-1\right)\left[P_e\times \left(RR_e-1\right)+1\right]$$

Where Pe was the representation of risk factor prevalence and RRe was the relative risk because of the factor, comparing the incidence of cognitive decline in the exposed and unexposed groups. However, participants may not only face a single risk factor. As a result, it was important to calculate the weighted PAF adjusted for the correlation to further account for the existence of multiple risk factors [36]. The formula [37] was shown as follows, where communality was the sum of the square of all factor loadings and the w was 1 minus communality.

$$ Weighted\ PAF=1-\prod \left[1-\left(w\times PAF\right)\right]$$

Meanwhile, the restricted cubic spline analysis was also used to examine whether there was a dose-response relationship between single air pollutants or the APS with the incidence of cognitive impairment and dementia. We additionally conducted the stratified analysis on potential confounders to further explore the possible relevance of the genetic predisposition, sociodemographic characteristics, lifestyle, and prevalent disease with air pollution-induced cognitive impairment.

All statistical analyses were performed with SAS software. All P-values were based on the two-sided test, and P-values < 0.05 were considered statistically significant. The figures in the current article were generated with R software, GraphPad Prism software (version 9.4.1), and Adobe Illustrator software.