Arterial Blood Pressure and Long-Term Exposure to Traffic-Related Air Pollution: An Analysis in the European Study of Cohorts for Air Pollution Effects (ESCAPE)

Background: Long-term exposure to air pollution has been hypothesized to elevate arterial blood pressure (BP). The existing evidence is scarce and country specific. Objectives: We investigated the cross-sectional association of long-term traffic-related air pollution with BP and prevalent hypertension in European populations. Methods: We analyzed 15 population-based cohorts, participating in the European Study of Cohorts for Air Pollution Effects (ESCAPE). We modeled residential exposure to particulate matter and nitrogen oxides with land use regression using a uniform protocol. We assessed traffic exposure with traffic indicator variables. We analyzed systolic and diastolic BP in participants medicated and nonmedicated with BP-lowering medication (BPLM) separately, adjusting for personal and area-level risk factors and environmental noise. Prevalent hypertension was defined as ≥ 140 mmHg systolic BP, or ≥ 90 mmHg diastolic BP, or intake of BPLM. We combined cohort-specific results using random-effects meta-analysis. Results: In the main meta-analysis of 113,926 participants, traffic load on major roads within 100 m of the residence was associated with increased systolic and diastolic BP in nonmedicated participants [0.35 mmHg (95% CI: 0.02, 0.68) and 0.22 mmHg (95% CI: 0.04, 0.40) per 4,000,000 vehicles × m/day, respectively]. The estimated odds ratio (OR) for prevalent hypertension was 1.05 (95% CI: 0.99, 1.11) per 4,000,000 vehicles × m/day. Modeled air pollutants and BP were not clearly associated. Conclusions: In this first comprehensive meta-analysis of European population-based cohorts, we observed a weak positive association of high residential traffic exposure with BP in nonmedicated participants, and an elevated OR for prevalent hypertension. The relationship of modeled air pollutants with BP was inconsistent. Citation: Fuks KB, Weinmayr G, Foraster M, Dratva J, Hampel R, Houthuijs D, Oftedal B, Oudin A, Panasevich S, Penell J, Sommar JN, Sørensen M, Tittanen P, Wolf K, Xun WW, Aguilera I, Basagaña X, Beelen R, Bots ML, Brunekreef B, Bueno-de-Mesquita HB, Caracciolo B, Cirach M, de Faire U, de Nazelle A, Eeftens M, Elosua R, Erbel R, Forsberg B, Fratiglioni L, Gaspoz JM, Hilding A, Jula A, Korek M, Krämer U, Künzli N, Lanki T, Leander K, Magnusson PK, Marrugat J, Nieuwenhuijsen MJ, Östenson CG, Pedersen NL, Pershagen G, Phuleria HC, Probst-Hensch NM, Raaschou-Nielsen O, Schaffner E, Schikowski T, Schindler C, Schwarze PE, Søgaard AJ, Sugiri D, Swart WJ, Tsai MY, Turunen AW, Vineis P, Peters A, Hoffmann B. 2014. Arterial blood pressure and long-term exposure to traffic-related air pollution: an analysis in the European Study of Cohorts for Air Pollution Effects (ESCAPE). Environ Health Perspect 122:896–905; http://dx.doi.org/10.1289/ehp.1307725

. Adjusted association of traffic-related air pollution (5-µg/m³ increase in PM 2.5 , or 10-µg/m³ increase in NO 2 ) with systolic blood pressure (with or without BPLM intake, change in mmHg), prevalent hypertension (odds ratio), and use of blood pressure lowering medication (odds ratio) in the study population, based on multi-pollutant models (N=91,574) 18 Figure S1. Meta-analysis with traffic load at the major road fragments in categories with systolic and diastolic blood pressure 19 Figure S2. Association of PM 2.5 , NO 2 , and traffic load at the major road fragments with systolic blood pressure in main and extended meta-analysis, and stratified by details of measurement procedure 20 Figure S3. Cohort-specific and meta-analysis estimates of association of PM 2.5 with systolic blood pressure: main and extended meta-analysis, stratified by medication status 22 Figure S4. Association of PM 2.5 with systolic blood pressure: a comparison of adjustment models from crude to main in medicated and non-medicated with blood-pressure lowering medication, and censored regression 23 Figure S5. Extended adjustment sets and analysis with back-extrapolated exposures with systolic blood pressure in participants not medicated with blood pressure lowering medication 24

References 26
Methods

Cohort-specific information, funding and acknowledgments
The National FINRISK Study (FINRISK, Finland) has been established in 1972; randomly selected representative population samples are recruited from different parts of Finland every five years. For this study, three cross-sectional population surveys (1997, 2002, and 2007) and two study areas (cities of Helsinki/Vantaa and Turku/Loimaa region) were used (Peltonen et al. 2008). Funding and acknowledgments. For the Finnish part, additional funding came from the Academy of Finland (project number 129317). Exposure assessment was performed by Tarja Yli-Tuomi, Pekka Taimisto, and Arto Pennanen from the Department of Environmental Health.
Mortality, area-level SES, and building data were provided by Statistics Finland.
The Danish Cancer study (DCH, Denmark) aims to investigate diet and lifestyle in relation to the incidence of cancer and other chronic diseases. n total, 57,053 of 160,725 residents of Copenhagen or Aarhus aged 50 -64 years without a history of cancer (excluding non-melanoma skin cancer) were enrolled into the original cohort between 1993 and 1997 (Tjønneland et al. 2007). Only participants from the Copenhagen area were included in ESCAPE. Funding and acknowledgments. The Danish Cancer Society contributed funding to establish the Diet Cancer and Health study.
The population-based Oslo Health Study (HUBRO, Norway) was designed to identify health needs within Oslo and determine the priorities of the health sector (Søgaard et al. 2004). Another aim was to identify social and geographical differences in health and associated risk factors.
HUBRO was carried out in the city of Oslo in 2001. All men and women born in 1924All men and women born in , 1925All men and women born in , 1940All men and women born in , 1941All men and women born in , 1955All men and women born in , 1960All men and women born in and 1970  The Stockholm 60-year olds cohort (60-year-olds, Sweden) was initiated for identification of biological and socio-economic risk factors and predictors for cardiovascular diseases through a cross-sectional health screening study (Wändell et al. 2007 (Beulens et al. 2010).
• The EPIC-Prospect study (the Netherlands) consists of 17,500 healthy women, living in Utrecht and surroundings were enrolled. Women were recruited from breast cancer screening participants, age 50-70 years at enrolment. Baseline information was collected between 1993-1997 on the basis of two self-administered questionnaires and a medical examination (Beulens et al. 2010).
• Funding and acknowledgments. The EPIC-MORGEN and EPIC-PROSPECT cohorts were supported by the Dutch Ministry of Public Health, Welfare and Sports (VWS),
The EPIC Oxford cohort (EPIC-Oxford, the United Kingdom) was recruited from the nationwide general population throughout the United Kingdom into the Oxford cohort, containing 65,429 men and women through medical general practices or by post between 1993 and 2000 (Danesh et al., 2007).
The Heinz Nixdorf Risk Factors, Evaluation of Coronary Calcification, and Lifestyle (Recall) study (HNR, Germany) is an ongoing population-based, prospective cardiovascular cohort study that started in 2000. The main objective of the study is the assessment of the predictive power of coronary artery calcification for the incidence of hard coronary events such as myocardial infarction and sudden cardiac death in an unselected general population. HNR includes 4814 randomly selected participants aged 45 to 75 years from three large adjacent German cities (Essen, Mülheim, Bochum) of the densely populated and highly industrialized Ruhr Area (Schmermund et al. 2002;Stang et al. 2006 ≤10 µm (PM 10 ), >2.5 to ≤10 µm (PM coarse ; calculated as PM 10 minus PM 2.5 ), PM 2.5 absorbance, and nitrogen oxides (NO 2 -nitrogen dioxide and NO x -nitrogen oxide) at residence were estimated in each study region using Land Use Regression (LUR) models, based on a harmonized approach described elsewhere (Beelen et al. 2013;Eeftens et al. 2012a).
Measurements of both PM and nitrogen oxides were conducted for a one-year period between 2008 and 2011 at sites chosen to represent the anticipated spatial variation of air pollution in the corresponding study area (Cyrys et al. 2012;Eeftens et al. 2012b (Beelen et al. 2013;Eeftens et al. 2012a). Concentrations for each individual's residence were then predicted using the developed study-specific LUR models, and the extreme values were truncated to the minimum and maximum measured values at the monitoring sites.

Extrapolation of exposure values back in t ime
The predicted concentrations for PM 10 and NO 2 were back extrapolated to the time of the BP measurement, if corresponding measured data from routine network sites were available. The

Exposure increments in the analyses
All exposure increments were a priori chosen centrally for the complete ESCAPE study to ensure that all papers in ESCAPE would report results on the same scale. The standardized increments were based on the mean 5 th -95 th percentile difference across all ESCAPE study areas. Results were calculated for fixed increments of exposures: 5 µg/m³ for PM 2.5 and PM coarse , 10 µg/m³ for PM 10 and NO 2 , 20 µg/m³ for NO , and 10 -5 m -1 x for PM 2.5 absorbance. Traffic load on major road fragments in a 100 m buffer was included in the analyses as a continuous variable with an increment of 4,000,000 vehicles × m/day, as well as in categories: ≤500,000 (reference); >500,000 & ≤1,500,000; >1,500,000 & ≤3,000,000; >3,000,000 vehicles × m/day. Traffic intensity at the nearest road was analyzed as a continuous variable with an increment of 5,000 vehicles/day.

Noise assessment
Since 2002 member states of the European Union (EU) are obliged to produce every 5th year noise maps for major roads, major railways and major airports and for larger agglomerations. and German cohorts; the interim method of the EU was applied for REGICOR.

Assessment of blood pressure lowering medication use
Intake of blood pressure (

Cohort-specific analysis
The main model fit and assumptions for linear regression were checked using a centrally developed code: collinearity of predictor variables; distribution of residuals; presence of influential observations. These checks were performed in the model without exposure.
The following model specifications were used in the cohort-specific analyses: • Model 1: crude (only exposure); fixed-effects model.
• Model 3a: same covariates as in model 3; random-effects model (neighborhood as random intercept) • Model 4 (main model): model 3a with neighborhood level SES; random-effects model.
If the random-effects models did not converge, we accounted for clustering by using the robust variance estimator for the cluster (neighborhood), specifying that the standard errors allow for intragroup correlation, relaxing the usual requirement that the observations be independent. In the TwinGene cohort, consisting of twin pairs by about 50%, a random intercept was introduced for twin pair instead (based on the lower AIC).

Sensitivity meta-analysis and meta-regression
To study the effect of the quality of BP measurement on the estimate, we combined studies in groups according to details of the BP measurement procedure and performed the stratified metaanalysis. First, we divided the studies in 3 groups by the number of consequent measurements: • Group 1: ≥3 BP measurements (HUBRO, SNAC-K, HNR, KORA included); • Group 2: ≥2 BP measurements (FINRISK, 60-year-olds, EPIC-Umeå, EPIC-MORGEN, EPIC-Prospect, SAPALDIA, REGICOR added to the studies in Group 1); • Group 3: <2 measurements (SDPP; DCH, EPIC-Oxford added to the studies in Group 2).
Next, we divided studies by participant's position during the BP measurement: We carried meta-regression with the metareg command in STATA 12.0, using as the following parameters independent predictors: • time interval between BP measurement and exposure assessment (year of exposure measurement minus a) first and b) last year of BP measurement, entered as continuous variable, or in 5 year categories, and in 10-year categories; • mean and standard deviation of LUR exposure; • geographical position: North, Center, South; • proportion of participants with road traffic noise levels at the residence > 65 dB, • cohort characteristics: mean and standard deviation of BP, age, BMI, proportion of participants under BP lowering treatment, with prevalent hypertension, current smokers, frequent alcohol drinkers, practicing no sport, unemployed, with low education.
Meta-regression was performed for systolic BP in medicated and non-medicated participants with PM 2.5 and NO 2 as exposures. The significance level was set to p <0.01 to account for the multiple testing.   Table S3. Adjusted association of traffic-related air pollution (5-µg/m³ increase in PM 2.5 , or 10µg/m³ increase in NO 2 ) with systolic blood pressure (with or without BPLM intake, change in mmHg), prevalent hypertension (odds ratio), and use of blood pressure lowering medication (odds ratio) in the study population, based on multi-pollutant models (N=91,574).  Figure S1. Meta-analysis with traffic load at the major road fragments in categories with systolic and diastolic blood pressure. Legend: BP = blood pressure; BPLM = BP lowering medication; change in mmHg with 95% confidence interval is presented. Adjusted for age, sex, body mass index, smoking status, pack-years, passive smoking, alcohol consumption, physical activity, educational level, economic activity, neighborhood SES  In the e xtended meta-analysis, DCH and EPIC-Oxford were added (N=15). • SM (N=4): FINRISK, SDPP, SNAC-K, EPIC-Umeå a included. a EPIC-Umeå was included only in the analyses with NO 2 and traffic load at major road fragments, because PM 2.5 was not modeled in this cohort. Figure S3. Cohort-specific and meta-analysis estimates of association of PM 2.5 with systolic blood pressure: main and extended meta-analysis, stratified by medication status. Legend: BP = blood pressure; BPLM = BP lowering medication; change in mmHg with 95% confidence interval is presented. Figure S4. Association of PM 2.5 with systolic blood pressure: a comparison of adjustment models from crude to main in medicated and non-medicated with blood-pressure lowering medication, and censored regression.

Supplemental Figures
Legend: BP = blood pressure; BPLM = BP lowering medication; RF = risk factors; RE = random effect for neighborhood; nSES = neighborhood socio-economic status; change in mmHg with 95% confidence interval is presented. Figure S5. Extended adjustment sets and analysis with back-extrapolated exposures with systolic blood pressure in participants not medicated with blood pressure lowering medication. Legend: BP = blood pressure; BPLM = BP lowering medication; change in mmHg with 95% confidence interval is presented.