Long-Term Exposure to Ambient Air Pollution and Incidence of Cerebrovascular Events: Results from 11 European Cohorts within the ESCAPE Project

Background: Few studies have investigated effects of air pollution on the incidence of cerebrovascular events. Objectives: We assessed the association between long-term exposure to multiple air pollutants and the incidence of stroke in European cohorts. Methods: Data from 11 cohorts were collected, and occurrence of a first stroke was evaluated. Individual air pollution exposures were predicted from land-use regression models developed within the European Study of Cohorts for Air Pollution Effects (ESCAPE). The exposures were: PM2.5 [particulate matter (PM) ≤ 2.5 μm in diameter], coarse PM (PM between 2.5 and 10 μm), PM10 (PM ≤ 10 μm), PM2.5 absorbance, nitrogen oxides, and two traffic indicators. Cohort-specific analyses were conducted using Cox proportional hazards models. Random-effects meta-analysis was used for pooled effect estimation. Results: A total of 99,446 study participants were included, 3,086 of whom developed stroke. A 5-μg/m3 increase in annual PM2.5 exposure was associated with 19% increased risk of incident stroke [hazard ratio (HR) = 1.19, 95% CI: 0.88, 1.62]. Similar findings were obtained for PM10. The results were robust to adjustment for an extensive list of cardiovascular risk factors and noise coexposure. The association with PM2.5 was apparent among those ≥ 60 years of age (HR = 1.40, 95% CI: 1.05, 1.87), among never-smokers (HR = 1.74, 95% CI: 1.06, 2.88), and among participants with PM2.5 exposure < 25 μg/m3 (HR = 1.33, 95% CI: 1.01, 1.77). Conclusions: We found suggestive evidence of an association between fine particles and incidence of cerebrovascular events in Europe, even at lower concentrations than set by the current air quality limit value. Citation: Stafoggia M, Cesaroni G, Peters A, Andersen ZJ, Badaloni C, Beelen R, Caracciolo B, Cyrys J, de Faire U, de Hoogh K, Eriksen KT, Fratiglioni L, Galassi C, Gigante B, Havulinna AS, Hennig F, Hilding A, Hoek G, Hoffmann B, Houthuijs D, Korek M, Lanki T, Leander K, Magnusson PK, Meisinger C, Migliore E, Overvad K, Östenson CG, Pedersen NL, Pekkanen J, Penell J, Pershagen G, Pundt N, Pyko A, Raaschou-Nielsen O, Ranzi A, Ricceri F, Sacerdote C, Swart WJ, Turunen AW, Vineis P, Weimar C, Weinmayr G, Wolf K, Brunekreef B, Forastiere F. 2014. Long-term exposure to ambient air pollution and incidence of cerebrovascular events: results from 11 European cohorts within the ESCAPE project. Environ Health Perspect 122:919–925; http://dx.doi.org/10.1289/ehp.1307301

. R 2 of LUR models a : model R 2 and leave-one-out cross-validation R 2 in the 11 cohorts 19 µ ): distribution in the 11 cohorts 20 Figure S1. Map of the study area 21 Figure S2. Forest plots of the cohort-specific results from the "base" model, with regard to the association between PM 2.5 and PM 10 with stroke incidence (also, pooled estimates from random-effects meta-analysis are reported) 22

Cohort details
• FINRISK: The National "FINRISK" surveys (Finalnd) have been conducted every 5 years since 1972 to monitor the risk factor trends of chronic diseases, including cardiovascular diseases, diabetes, cancer, asthma, and allergy (Peltonen et al. 2008). For each survey, a stratified random sample has been selected from the 25-64 (74 since 1997) year old inhabitants in different regions of Finland. The ESCAPE study used FINRISK data from four surveys (1992,1997,2002,2007) and two study regions (the cities of Helsinki and Vantaa, and Turku city with its nearby municipalities). The FINRISK study protocol has been described elsewhere (Vartiainen et al. 2010). The surveys included a self-administered questionnaire (the questions focus mainly on socioeconomic factors, medical history, health behavior, and psychosocial factors) and a clinical examination including measurements of height, weight and blood pressure and blood sampling. The participants have been annually followed up through 31 Dec 2008 (up to 16 years) for fatal and nonfatal coronary and stroke events and total mortality. The National Hospital Discharge Register and the National Causes of Death Register were used to identify these events. These registers cover every hospitalization in Finland and every death of permanent residents in Finland, yielding in practice 100% coverage of the follow-up events (Pajunen at el. 2005;Tolonen et al. 2007). In addition, we used the drug reimbursement records from the Social Insurance Institution of Finland to identify subjects who had developed diabetes or hypertension during the follow-up period.
• SNAC-K: The Swedish National study of Aging and Care in Kungsholmen (SNAC-K) is an ongoing longitudinal study aiming to investigate the ageing process and identify possible preventive strategies to improve health and care in elderly adults (Lagergren et al. 2004 including blood pressure and blood samples were also obtained. The DCH cohort is followed up regularly by use of complete nationwide registers hence the loss to follow-up is virtually nil. • HNR: The Heinz Nixdorf Recall (HNR) study is an on-going population-based prospective cohort study, including 4,814 randomly selected participants 45 to 75 years of age at baseline (2000)(2001)(2002)(2003) from three large adjacent cities (Essen, Mülheim and Bochum) of the densely and highly industrialized Ruhr area in western Germany (Schmermund et al. 2002). The HNR study was initiated to evaluate predictive value of EBCT compared to traditional and new risk factors in order to develop more effective methods of predicting cardiovascular disease in the general population. The study was approved by the institutional ethics committees and follows strict internal and external quality assurance protocols. Examination assessment included a self-administered questionnaire, face-to-face interviews for personal risk factor assessment, clinical examinations, and comprehensive laboratory tests according to standard protocols. With regard to the ascertainment of stroke incident and prevalent cases, the HNR cohort relied on death certificates, and conducted an expert review of medical records by an independent endpoint committee (adjudication of cases) to ascertain cases during follow-up.
Study design has been described in detail elsewhere (Schmermund et al. 2002). Follow-up started in 1998, including collaboration with the local cancer registry, the demographic computerized archives of the Torino area and the discharge report database for hospital patients.

Prevalent cases
Prevalent cases of either coronary or cerebrovascular disease at baseline were excluded. Methods to define and ascertain prevalent cases differed between the cohorts. For most of the cohorts, prevalent cases were defined by information collected at baseline by a questionnaire on acute myocardial infarction, stroke history and medication use. When only hospitalization data were available, at least three years of data before enrolment were considered.

Exposure assessment
Long-term exposure to ambient air pollutants at the residential address of each individual was estimated within the ESCAPE project following a three-step procedure. predictors, such as population density, traffic intensity and altitude. Measurements were taken at each site over three 14-day periods in different seasons. Results for each location were averaged after correcting for temporal variation using data obtained from a reference site, which was in operation year-round. Coarse PM was calculated as the difference between PM 10 and PM 2.5 .
Second, land-use regression models were developed for each study area and pollutant, with sitespecific annual average concentrations as the dependent variables, and an extensive list of spatial attributes as predictors. These included: roads characteristics, land-use data, population-density data, altitude, etc. The indicators of model performance were generally good for all pollutants, with leave-one-out cross-validation (LOOCV) R-squared coefficients for PM 2.5 ranging from 0.53 (FINRISK, Finland) to 0.79 (HNR, Ruhr area).
Third, once the models were optimized, their coefficients were applied to predict annual air pollution averages at the residence address of each cohort member. If values of predictor variables for the cohort addresses were outside the range of values for the monitoring sites, values were truncated to the minimum and maximum values at the monitoring sites. Truncation was performed to prevent unrealistic predictions (e.g. related to too small distance to roads in GIS) and because we did not want to extrapolate the derived model beyond the range for which it was developed. Truncation has been shown to improve predictions at independent sites. The estimates so obtained were considered as the reference long-term exposures in the present study.
Pollution measurements were performed in 2008-2011, but follow-up from baseline addresses was in all cohorts covering earlier time periods. We therefore extrapolated predicted concentrations back in time using the absolute difference and the ratio between the two periods, based on data from routine background monitoring network site(s) in the study areas. Details on this procedure can be found on the website http://www.escapeproject.eu/manuals/.
In addition, two traffic variables were used as relevant exposures: the traffic intensity on the nearest road (measured as the number of vehicles/day), and the traffic load on major roads within a 100m buffer (measured as the product of the intensity and the length of the roads intersecting the buffer). These variables were evaluated alone, and in combination with predicted background NO 2 concentrations. The noise exposure assessment was carried out locally. More specifically, the noise level (Lden: day-evening-night equivalent level) was calculated for the most exposed   reported: the first value is the model R 2 , the second value is the leave-one-out cross-validation R 2 obtained by deleting one monitor at a time and computing the R 2 coefficient on the remaining ones. a For each cohort, the distribution of the PM 2.5 concentrations predicted from the land-use regression models are reported: number of cohort participants, mean and standard deviation of the PM 2.5 concentrations predicted at the residence address, minimum, percentiles and maximum values, interquartile range (IQR, obtained as the difference between the 3 rd and the 1 st quartiles), and 5-95 range (obtained as the difference between the 95 th and the 5 th percentiles). Figure S1. Map of the study area. Figure S2. Forest plots of the cohort-specific results from the "base" model, with regard to the association between PM 2.5 and PM 10 with stroke incidence (also, pooled estimates from random-effects meta-analysis are reported).