Association between Long-Term Exposure to Traffic-Related Air Pollution and Subclinical Atherosclerosis: The REGICOR Study

Background: Epidemiological evidence of the effects of long-term exposure to air pollution on the chronic processes of atherogenesis is limited. Objective: We investigated the association of long-term exposure to traffic-related air pollution with subclinical atherosclerosis, measured by carotid intima media thickness (IMT) and ankle–brachial index (ABI). Methods: We performed a cross-sectional analysis using data collected during the reexamination (2007–2010) of 2,780 participants in the REGICOR (Registre Gironí del Cor: the Gerona Heart Register) study, a population-based prospective cohort in Girona, Spain. Long-term exposure across residences was calculated as the last 10 years’ time-weighted average of residential nitrogen dioxide (NO2) estimates (based on a local-scale land-use regression model), traffic intensity in the nearest street, and traffic intensity in a 100 m buffer. Associations with IMT and ABI were estimated using linear regression and multinomial logistic regression, respectively, controlling for sex, age, smoking status, education, marital status, and several other potential confounders or intermediates. Results: Exposure contrasts between the 5th and 95th percentiles for NO2 (25 µg/m3), traffic intensity in the nearest street (15,000 vehicles/day), and traffic load within 100 m (7,200,000 vehicle-m/day) were associated with differences of 0.56% (95% CI: –1.5, 2.6%), 2.32% (95% CI: 0.48, 4.17%), and 1.91% (95% CI: –0.24, 4.06) percent difference in IMT, respectively. Exposures were positively associated with an ABI of > 1.3, but not an ABI of < 0.9. Stronger associations were observed among those with a high level of education and in men ≥ 60 years of age. Conclusions: Long-term traffic-related exposures were associated with subclinical markers of atherosclerosis. Prospective studies are needed to confirm associations and further examine differences among population subgroups.

Air pollu tion from traffic and other sources is an established cause of premature mortality (Brook et al. 2010). A relevant part of this environmental burden of disease relates to cardio vascular diseases (CVDs), which were responsible for 10% of total disability-adjusted life years in 2004 and the leading causes of death worldwide in 2008 (World Health Organization 2008). A common feature of this group of diseases is athero sclerosis, a chronic and degenerative process that mainly occurs in large and medium-sized arteries and is characterized by asymmetric focal thickenings of the innermost layer of the artery, the intima. The development of athero sclerosis is the result of the total cumulative exposure to atherogenic risk factors such as hypertension, high cholesterol, diabetes, obesity, smoking, physical inactivity, and other lifestyle factors and their interactions with genetic susceptibility (Lusis 2000). Acute events such as myocardial infarction or stroke can be triggered by shortterm exposure to air pollu tion (Brook et al. 2010). However, whether and how ambient air pollu tion contributes to athero genesis is subject to debate. Although experimental studies on animals provide strong evidence for a causal athero genic role of air pollu tion, particularly in obese mice (Sun et al. 2005), evidence from epidemiological studies is limited.
The long subclinical phase of atherosclerosis makes it possible to investigate the determinants of the vascular pathology long before its clinical manifestation. The association between air pollu tion and intima media thickness (IMT), an established marker of subclinical athero sclerosis, was reported for the first time in volunteers participating in two clinical trials in California (Künzli et al. 2005). Two population-based cross-sectional analyses, namely the Heinz Nixdorf Recall study in Germany (Bauer et al. 2010) and the Multi-Ethnic Study of Atherosclerosis (MESA) cohort in the United States (Diez Roux et al. 2008), also reported associations between air pollu tion and IMT, although a study of young adults in the Netherlands found no association (Lenters et al. 2010). So far, only one longitudinal study has been published, and it was based on heterogeneous samples of volunteers participating in five clinical trials (Künzli et al. 2010), including the two trials of the first cross-sectional study published on this topic (Künzli et al. 2005). Findings from the study suggested a possible role of ambient air pollut ants, indicated by particulate matter (PM) ≤ 2.5 µm in diameter (PM 2.5 ) and living close to busy highways, in the progression of IMT (Künzli et al. 2010).
Most previous studies have reported that associations between air pollu tion and IMT differ among population subgroups (Künzli et al. 2011). However, subgroup patterns have not been consistent; thus a clear understanding of susceptibility factors is still lacking. Human studies have not investi gated whether diet modifies the effects of air pollu tion-a plausible hypothesis given evidence from animal studies (Sun et al. 2005) and the effects of diet on oxidative stress and systemic inflammation that are likely to contribute to the systemic effects of ambient air pollu tion (Brook et al. 2010).
volume 121 | number 2 | February 2013 • Environmental Health Perspectives The Mediterranean region of Girona, and Spain in general, has one of the lowest cardiovascular mortality rates in Europe, despite very high prevalence of conventional cardio vascular risk factors (Masiá et al. 1998), a paradox that may be explained in part by protective effects of the Mediterranean diet (Martínez-González et al. 2011). We aimed to investigate the association between long-term exposure to traffic-related air pollu tion and subclinical athero sclerosis in Spain, and modification of this association by diet and other factors identified in previous studies (Künzli et al. 2011). We investigated this in the follow-up of participants of three population-based cohorts of the REGICOR (Registre Gironí del Cor: the Gerona Heart Register) study (Grau et al. 2007). Subclinical athero sclerosis was measured by carotid IMT, a validated marker of athero sclerosis (Coll and Feinstein 2008). In addition, we measured the ankle-brachial index (ABI), a marker of the presence and severity of peripheral artery disease. Both IMT and ABI are associated with cardio vascular events and mortality (Ankle Brachial Index Collaboration 2008;Lorenz et al. 2007). In the absence of PM measurements in this region, we used estimates of 10-year average home outdoor nitrogen dioxide (NO 2 ) concentrations and residential traffic intensity as markers of exposure to local traffic-related air pollut ants.

Methods
REGICOR-Air is a cross-sectional study nested in the REGICOR cohort study (Grau et al. 2007) evaluating the association between air pollu tion and cardiovascular diseases. We used data from the follow-up of three populationbased cohorts originally enrolled in 1995, 2000, and 2005 (with response rates of 72.4, 70.0, and 73.8%, respectively) from 12 towns that represent the geographic diversity of the Girona Province and have large contrasts in ambient air pollu tion levels [see map in Supplemental Material, Figure S1 (http:// dx.doi.org/10.1289/ehp.1205146)]. During 2007-2010, the participants residing in these towns, who were alive and not institutionalized, were invited to participate in REGICOR-Air, and the response rate was approximately 82%. In addition to extensive health status reassessment, IMT and ABI were measured for the first time in this population. Address histories for the past 10 years were collected by questionnaire, and each address was geocoded at the front-door level. This study was approved by the Hospital del Mar Research Institute ethics committee, and participants gave written informed consent.
IMT was measured by three trained, certified sonographers using ultrasound examinations of the left and right carotid arteries. Standardized scan and image analyses protocols were used. The scan protocol entailed the right and left carotid arteries. In each artery, the following predefined longitudinal segments were imaged: a) the common carotid 1 cm proximal to the dilation, b) the carotid bulb, and c) the internal carotid 1 cm distal to the flow divider. For imaging, an Acuson Aspen ultrasound instrument (Acuson-Siemens, Erlangen, Germany) equipped with an L7 5-12MHz transducer and dedicated REGICOR application scan protocol (AMC IMAGELAB, University of Amsterdam, Amsterdam, and Technical University Eindhoven, Eindhoven, the Netherlands) were used. A still image of each arterial segment was saved as a DICOM (digital information and communication in medicine) file. These source files were locally stored and securely transferred to IMAGELAB, where trained and certified sonographers analyzed the images using validated "eTrack REGICOR" software (Department of Physiology and Vascular Medicine Academic Medical Centre, Amsterdam, the Netherlands) (de Groot et al. 2008). IMT was defined as the average distance between the lumen-intima and media-adventitia interfaces in a given 1 cm segment of the artery far wall. The main outcomes were a) the mean of the IMTs for the left and right common carotid arteries (IMTcca), and b) the mean of the IMTs for the left and right common carotid arteries, internal carotid arteries, and carotid bulbs (IMT6seg). These two outcomes were assessed separately because of differences in the segments' cellu lar constituents-with more foam cells in the common carotid artery and more cholesterol-rich plaques in the carotid bulb and the internal carotid artery (Gállego Pérez-Larraya et al. 2012)and in risk factors for IMT (Polak et al. 2010), which suggest that etiology may differ according to location. Between-sonographer and between-visit variability were evaluated based on repeated IMT measurements conducted at two visits, 2 weeks apart, by up to three sonographers, in 42 participants.
Systolic blood pressure was measured in a supine position after a 5-min rest in the brachial artery of both arms and the posterior tibial and dorsalis pedis arteries of both legs, using a continuous Doppler device. Right and left ABI were calculated as the ratio of the highest leg pressure to the highest brachial pressure in the corresponding arm, and the lowest of the two ABI ratios were cate gorized as low (< 0.9), normal (0.9-1.3), or high (> 1.3) for analysis (Ankle Brachial Index Collaboration 2008;McDermott et al. 2005).
Adherence to the Mediterranean diet was measured by a 10-point index based on sex-specific intake tertiles of seven beneficial (cereals, fruits, vegetables, legumes, seafood, nuts, moderate red wine) and two detrimental (meats, dairy products) food groups, and categorized in quartiles. Modifications such as excluding low-fat dairy products or white meats from the detrimental foods, or incorporating additional unhealthy food groups (e.g., soft drinks, salty snacks, pastries) as detrimental components, had no meaningful impact on findings in sensitivity analyses (data not shown). A detailed description of the methodol ogy used to measure adherence to the Mediterranean diet in the REGICOR study is given by Schröder et al. (2004). The plausibility of reported dietary intakes was assessed based on disparities between reported energy intakes and estimated energy requirements (Mendez et al. 2011).
Participants with a clinical history of CVD (myocardial infarction, stroke, angina, catheterization, angioplasty, bypass surgery, or amputation due to circulatory problems) were excluded (n = 227) because medication use or altered health behaviors among these participants may have influenced IMT/ABI measures obtained for this study.
Exposure assessment. We estimated the 10-year time-weighted average of the home outdoor concentrations of NO 2 for each participant using land use regression (LUR) models. In the absence of air quality data in most of these towns, we conducted an extensive monitoring campaign using NO 2 passive samplers to validate model estimates. We measured NO 2 in the balcony of 562 participants' homes for 1 month in the spring and again in the fall between June 2007 and July 2009. Homes were selected to cover a broad range of traffic-related pollu tion and urban settings (e.g., low-and high-building-density areas), to be representative of the residential locations of the cohort participants and to be well distributed across the towns. We adjusted for temporal variability using monthly mean NO 2 concentrations collected at a fixed location in each town for at least 1 year. NO 2 annual means were derived by multiplying the monthly means at each location by the ratio of the annual mean to the same month mean NO 2 at the town's fixed location.
To predict NO 2 at each participant's residence we used LUR models based on NO 2 annual means and data on traffic intensity, bus routes and stops, distance to traffic, land cover, building density, and other geographic information system (GIS)-derived variables. Given the geographic diversity of the study area and differences in the availability of GIS data among towns, we divided the study area into seven subareas, i.e., groups of adjacent towns. LUR models were derived for each group by supervised forward linear regression according to the methodology described by Rivera et al. (2012). The models explained between 33% and 63% of measured NO 2 (crossvalidation R 2 0.32 and 0.61, respectively) [see Supplemental Material, Table S1 and Figure S2 (http://dx.doi.org/10.1289/ehp.1205146)].
We estimated the outdoor annual mean NO 2 at each residential location by applying the LUR models to the address geocode. NO 2 concentrations were back-extrapolated to the time period of residence in each address using daily means at a fixed urban background monitoring location. Finally, for each participant we calculated the time-weighted average of NO 2 estimates across all residences in the 10 years preceding the IMT measurement (10-year NO 2 exposure). Time periods when participants lived at addresses that were geocoded with low precision or were outside of the study area were excluded when deriving the 10-year NO 2 exposure. Participants who lived in the study towns for < 6 years or had high-precision geocodes (i.e., at the front-door level) for < 6 years (n = 365) were excluded from the main analyses.
We also used traffic proximity markers as surrogates of air pollu tion exposure in independent analyses. Traffic intensity, available from local registries and the Urban Mobility Plan for Girona (2007), was assigned to the central road network used within ESCAPE (European Study of Cohorts for Air Pollution Effects) (Hoek et al. 2012). Traffic counts were conducted at approximately 670 streets with missing traffic information to complete the traffic inventory. The traffic intensity assessment is described in detail elsewhere . For each address, we calculated the traffic intensity at the nearest street and the traffic load (sum of traffic intensity multiplied by length of road segment in all segments) in a 100 m buffer and derived 10-year average values for each participant.
Statistical analysis. Crude and adjusted associations of IMT (which were natural logtransformed to reduce heteroscedasticity) with individually assigned air pollu tion exposures were estimated using linear regression models. Linear models were compared with additive models with smoothing splines using the generalized additive model (GAM) function in the MGCV R package (http://www.R-project.org) and linearity was confirmed. We analyzed ABI using multinomial logistic regression (Hosmer and Lemeshow 2004). We initially adjusted our models by age and sex. Age was a main determinant of both outcome and exposure. Thus, in a second step, we analyzed the association of the potential confounders with both outcome and exposure variables, adjusting by age and sex. Variables associated with both, namely, smoking status, education level (as a proxy of socioeconomic status), and marital status were included in a minimal adjustment set. The association between exposure and IMT was strongly confounded by age. To determine whether age was an effect modifier, the interaction term age × exposure was introduced in the model, and it was not statistically significant (p > 0.05). We also evaluated whether the effect of age differed for men and women. This was the case, so we added the sex × age interaction to our minimal adjustment set (model 1). We further adjusted by other potential confounders {physical activity [weekly energy expenditure in leisure time based on the Minnesota questionnaire (Elosua et al. 2000) expressed in tertiles of metabolic equivalents], diet [adherence to Mediterranean diet and plausibility of reported diet], SES at the census-tract level [percentage of residents with < secondary school education in the tract where participants resided the longest, based on the 2001 census], and occupational status}, and results (not shown) did not differ from those of model 1. Finally, in addition to the variables listed above, we also included potential intermediates [body mass index (BMI), high-density lipoprotein level (HDL), waist circumference, systolic and diastolic blood pressure, and medication treatment (lipid-lowering, antihypertensives, antiplatelets, or anticoagulants, based on self-report and medical records)] to explore the effects of air pollu tion that do not follow these pathways (model 2). Adjustment variables were entered in the models either as continuous variables or dummy variables for all categories according to We did not adjust the exposure-outcome associations for area of residence in the main analyses because the study area was relatively small (65 × 70 km), data collection in every town was done by the same team using the exact same procedures and because doing so would have partially removed the exposure contrast corresponding to between-town variability. However, we explored the sensitivity of the results to the inclusion of area of residence (corresponding to the address of longest residence) as a random effect variable. The trend of the associations remained the same, but the estimates were closer to the null and less precise (data not shown). This is consistent with a decrease in the exposure contrast, which makes the detection of an effect more difficult. Residual confounding by area is less likely because the estimates were adjusted by a large array of confounders, including area-level confounders such as education at the census-tract level.
Because living near busy roads is also associated with traffic-related noise, we adjusted for road traffic noise in a subsample of participants with available noise data (n = 1,084) (Foraster et al. 2011). However, results were not sensitive to adjustment for noise (data not shown). Results are expressed as the cross-sectional percent difference in IMT, and the relative risk ratio (RRR) of high and low ABI, associated with a 10-year exposure contrast corresponding to the difference between the 5th and 95th percentiles in the study population. The RRR of high ABI is the prevalence of high ABI relative to normal ABI in exposed participants divided by the prevalence of high ABI relative to normal ABI in nonexposed participants.
We tested effect modification by diet, and factors identified as potential effect modifiers in previous studies, namely, age, sex, education level, smoking, diabetes, menopause, and medication treatment, using stratified analysis. We report the effect estimates of percent differences in IMT by strata of the potential effect modifiers.
To assess the sensitivity of the results to the NO 2 LUR models, we used the annual mean NO 2 at the closest monitor both within 100 m and within 200 m of the residence, instead of the 10-year modeled NO 2, in two separate sensitivity analyses. Median distance to the closest NO 2 monitoring site was 90 m. The sample size was restricted in these analyses because only those whose address of longest residence was within 100 m and within 200 m of a monitor were kept (e.g., 2,265 and 1,778 persons lived within 200 m of a monitor with data available on IMTcca and IMT6seg, respectively).
Analyses were performed using Stata version 10.1 (StataCorp, College Station, TX, USA) and R version 2.12 (http://www.R-project.org). The alpha level for statistical significance was set at 0.05.

Results
Information on IMTcca, IMT6seg, and ABI was available for 2,780, 2,188, and 2,738 partici pants, respectively. Fewer participants had IMT6seg measurements because of difficulties in analyzing the images to obtain measurements from the internal carotid artery and carotid bulb (Gállego Pérez-Larraya et al. 2012). The characteristics of the study populations are summarized in Table 1. Participants were 32-86 years of age. Percentages of participants with low, medium, and high education levels are consistent with Spain as a whole (Instituto Nacional de Estadística 2012). Participants included in the analyses did not differ from those excluded (participants who lived in the study towns for < 6 years or with clinical history of CVD) in terms of exposure levels (data not shown). The median IMTcca was 0.68 mm (range, 0.40-2.05 mm). The repeatability study showed intraclass correlation coefficients for sonographers and visits of 0.83 for the IMTcca and of 0.77 for the IMT6seg. ABI was on average 1.10 (range, 0.5-1.75) with 2.0% of the study population classified as having low ABI (< 0.9), and 4.2% with high ABI (> 1.3) ( Table 1). Participants with low ABI were on average 9 years older than participants with normal ABI, whereas those with high ABI did not differ in age from participants with normal ABI (data not shown). Inter-and intra operator variability of ABI measurements were low, with intra class correlation coefficients of 0.92 and 0.94, respectively.
The 10 Table S4) and 0.96 among those who moved at least once during the 10 years (data not shown). Ranges of 10-year time-weighted average values for traffic in the nearest street and traffic within 100 m indicated substantial variability among participants overall (see Supplemental Material, Table S3) and according to town of residence (see Supplemental Material, Table S5). Traffic intensity showed higher spatial variability than NO 2 [(75th percentile -25th percentile)/50th percentile was 2.51 for 10-year traffic in the nearest street, 1.56 for 10-year traffic load in 100 m, and 0.73 for 10-year NO 2 ]. NO 2 and traffic exposure variables were moderately correlated (Pearson correlation coefficients, r = 0.52-0.72), and the two traffic variables had a correlation of 0.58 (see Supplemental Material, Table S4).
In unadjusted models, NO 2 , traffic in the nearest street, and traffic in the 100 m buffer were strongly associated with the intima media thickness (both IMTcca and IMT6seg) ( Table 2). Associations decreased after adjusting for age, and in general, further adjustment (model 1 and model 2) provided similar estimates. Associations between a 25 µg/m 3 increase in NO 2 (corresponding Table 2. Estimated percent difference in IMT associated with a 10-year average exposure contrast between the 5th and 95th percentiles.
Ten-year NO 2 and both residential traffic indicators were associated with higher prevalence of high ABI (Table 3). Adjusted RRRs for high versus normal ABI were 1.98 (95% CI: 1.09, 3.60) for a 25 µg/m 3 increase in NO 2 ; 1.89 (95% CI: 1.07, 3.34) for a 7,200,000 vehicle-m/day increase in traffic load within 100 m; and 1.70 (95% CI: 1.13, 2.57) for an increase of 15,000 vehicles/day on the nearest street. The RRR of low ABI compared to normal ABI were consistent with the null hypothesis for all exposures.

Discussion
Our multivariate models revealed positive associations between three markers of longterm exposure to traffic-related air pollu tion, Figure 1. Subgroup analyses: estimates of percent differences (95% CIs) in IMTcca (left) and IMTseg (right) associated with exposure contrasts between the 5th and 95th percentiles for 10-year weighted average values of NO 2 (25 µg/m 3 ; A), traffic load within 100 m (7,200,000 vehicle-m/day; B), and traffic intensity on the nearest street (15,000 vehicles/day; C) according to sex, sex × age ≥ 60 years, education, medication treatment, and smoking status. Estimates adjusted by sex, age, sex × age interaction, smoking status, education, marital status, BMI, HDL, waist circumference, systolic and diastolic blood pressure, weekly energy expenditure in physical activity during leisure time (tertiles), adherence to Mediterranean diet, plausibility of reported diet, medication (med) treatment, and percentage of people with low education (educ) at the census-tract level. Models for traffic load were additionally adjusted for occupational status.
volume 121 | number 2 | February 2013 • Environmental Health Perspectives namely traffic load within 100 m of the residence, traffic intensity in the nearest street, and modeled concentration of home outdoor NO 2 and carotid subclinical athero sclerosis in a random healthy (no history or current signs of CVD) population sample of the Spanish Mediterranean region. However, associations with NO 2 were weak and reached statistical significance only in the subgroups of persons with a high education level (IMTcca and IMT6seg) and men ≥ 60 years of age (IMTcca). Both indicators of residential traffic were associated with approximately 2% thicker carotid intima-media. High exposures to NO 2 and traffic were also associated with higher prevalence of high ABI [RRR for high vs. normal ABI between 1.70 (95% CI: 1.13, 2.57) for traffic intensity in nearest street and 1.98 (95% CI: 1.09, 3.60) for NO 2 ]. To put these results in context with common risk factors for athero sclerosis, in the same study population included in main analyses, a 10-year difference in age, adjusted by confounders in model 2, was associated with an 8% (95% CI: 7.3, 9.0) difference in IMTcca and an RRR of 10.3 (95% CI: 10.0, 10.5) for high versus normal ABI. Interestingly, age strongly confounded the association of pollu tion with IMT. Removing age adjustment from the main multivariate models resulted in substantially larger estimates, very similar to the crude associations. Atherosclerosis is a life time process, thus age is a strong correlate of the cumulated lifetime exposure to air pollu tion as well as the main predictor of athero sclerosis. Therefore, adjustment for age may in part reflect overadjustment for exposure.
Our results suggest that long-term exposure to traffic-emitted pollu tion may contribute to carotid artery athero sclerosis. Given widespread exposure to high traffic intensity, the potential association of long-term pollution with athero sclerosis has important public health implications. Ninety-five percent of our study population was exposed to traffic intensity levels in the nearest street of < 15,000 vehicles/day, which is very low compared with other European cities. For example, in larger cities such as Barcelona, many people live within very short distances of streets with traffic intensities of 50,000-100,000 vehicles/day (Amato et al. 2009). Likewise, levels of NO 2 can exceed 100 µg/m 3 near streets with high traffic intensity. Expressing our results in terms of the 5th-95th percentile exposure contrast (25 µg/m 3 NO 2 , 15,000 vehicles/day on the nearest street, or 7,200,000 vehicle-m/day within 100 m) thus reflects ranges that are frequently observed in European cities.
Our findings are consistent with previous reports of association of medium-to long-term exposure with subclinical athero sclerosis (Bauer et al. 2010;Diez Roux et al. 2008;Hoffmann et al. 2007;Künzli et al. 2005) and systemic inflammatory markers (Hoffmann et al. 2009a) although direct comparisons are limited given our use of NO 2 instead of PM mass as a marker of pollu tion. However, our effect estimates are within the same order of magnitude as seen in previous studies.
In contrast, the association between pollution and ABI was less clear. Low ABI (< 0.9) is more clearly associated with mortality and cardio vascular disease than high ABI, but only high ABI was associated with traffic-related exposure in our study. Only 56 participants (2%) were classified as having low ABI [compared with 116 (4.2%) with high ABI], resulting in low power to detect an association. In the MESA study, 20-year exposures to PM 2.5 and PM 10 (PM ≤ 10 µm in diameter) were not associated with ABI as a continuous outcome, whereas 1-year mean PM 10 exposures were associated with higher ABI, "indicating less subclinical disease among persons with greater exposure" (Diez Roux et al. 2008). On the other hand, a study in Germany reported that living 50 m from a major road compared to living > 200 m away was associated with an odds ratio (OR) of 1.77 (95% CI: 1.01, 2.1) for peripheral arterial disease (ABI of < 0.9 or history of treatment for peripheral artery disease), whereas no associations were found with annual residential PM 2.5 (Hoffmann et al. 2009b).
The predictive power of an ABI of > 1.3 for cardio vascular risk has been less studied. It has been associated with calcification of the arterial wall, higher levels of many CVD risk factors (McDermott et al. 2005), and higher risk of all-cause mortality and foot ulcers; it has been weakly associated with heart failure and stroke (Allison et al. 2008). Thus, there is evidence of higher cardio vascular risk both at low and high levels of ABI. We found that traffic-related exposure was associated with high ABI. To our knowledge, the association of air pollution with high ABI has not been investigated previously. ABI is considered to be a marker for medial sclerosis, which is a specific form of arterial disease distinct from athero sclerosis (Alzamora et al. 2012). Because ABI is a ratio of systolic blood pressures, high ABI could result from low brachial pressure, or high ankle pressure, or both (Allison et al. 2008). Thus, if air pollu tion affects the upper and lower vascular beds differently, for example, by affecting brachial pressure at an earlier stage than ankle pressure, it could be associated with both high and low ABI. This hypothesis is consistent with our finding that high (vs. normal) ABI was not associated with IMT (adjusted RRR of 1.4; 95% CI: 0.3, 6.3 per 1 mm increase in IMTcca), whereas low (vs. normal) ABI was associated with thicker artery walls (adjusted RRR of 7.7; 95% CI: 1.6, 36.8, p = 0.01 per 1 mm increase in IMTcca).
In contrast to our findings, 20-year exposures to PM 2.5 and PM 10 , in the MESA study were weakly associated with carotid IMT (1-3% differences in IMT for 12.5 and 21 µg/m 3 increases in 20-year average PM 2.5 and PM 10 , respectively) but not with ABI (Diez Roux 2008). In Los Angeles, a 10 µg/m 3 increase in PM 2.5 at the current address was associated with a 4.2% (95% CI: -0.2, 8.9) increase in IMT (Künzli et al. 2005). In a longitudinal study that included the population of Künzli et al. (2005), a 10 µg/m 3 increase in PM 2.5 and living within 100 m of a highway were associated with 2.5 µm/year (95% CI; -0.31, 5.38) and 5.5 µm/year (95% CI; 0.13, 10.8) increases in IMT, respectively (Künzli et al. 2010). In the Heinz Nixdorf Recall study, a 4.2 µg/m 3 increase in 1-year average PM 2.5 and a 1,939 m increase in distance from a major road were associated with a 4.3% (95% CI: 1.9, 6.7%) and a 1.2% (95% CI: -0.2, 2.6%) difference in IMT, respectively (Bauer et al. 2010). Distance to traffic was used as a marker of traffic-related exposure; thus a negative association with IMT was expected. A 5.2 µg/m 3 increase in 1-year average PM 10 was significantly associated with IMT in a study of > 2,000 civil servants in London (Tonne et al. 2012). Positive, though not significant, associations of IMT with measures of traffic intensity at the current address, but no association with a 25 µg/m 3 increase in NO 2 , were observed in a population-based study of 745 young adults in Utrecht (Lenters et al. 2010).
Although associations with traffic as a surro gate of exposure appear to be more consistent across studies than associations with pollut ants, more research is needed to clarify the role of specific constituents of air pollu tion on athero genesis. In the study of young adults in Utrecht, associations of NO 2 , black smoke, PM 2.5 , and sulfur dioxide (SO 2 ) with IMT, pulse wave velocity, and augmentation index were investigated, but significant associations were found only for NO 2 with pulse wave velocity and augmentation index and for SO 2 with augmentation index (Lenters et al. 2010). The evidence on the health effects of NO 2 varies across studies. It has been reported to be more strongly associated with cardio vascular mortality than PM 2.5, black smoke, and SO 2 (Beelen et al. 2008) and with cardio pulmonary mortality than PM 10 , total suspended particles, black smoke, and SO 2 (Gehring et al. 2006), but Pope et al. (2002) found no association between NO 2 and cardio pulmonary mortality. The different associations observed between NO 2 and cardio pulmonary and cardiovascular outcomes across studies might reflect the suitability of using NO 2 as a marker of traffic-related pollu tion. Although the general suitability of NO 2 has been questioned before, the possibility of it being site dependent has not been studied. It has been reported that the correlation of NO 2 with traffic intensity varies across locations (Raaschou-Nielsen et al. 2000). The correlation of NO 2 with the components of the traffic emissions cocktail that promote (assuming the association was causal) or are truly associated with cardio vascular outcomes might also vary across locations.
Our exposure estimates (both NO 2 and traffic) were based on data collected at participants' residences, instead of data from air quality monitoring stations as far away as 10 km or more (Diez Roux et al. 2008). In addition, our traffic markers were based not only on proximity and road type classi fi cation (Allen et al. 2009;Hoffmann et al. 2009a), but also on actual traffic intensity derived from a dense traffic-count network. However, R 2 s for our NO 2 models indicate some exposure measurement error . To evaluate the potential influence of measurement error (both bias and misclassification) introduced by the exposure models, we compared associations with our model-based estimates of 10-year average NO 2 with associations based on annual mean NO 2 at the closest monitor. Results (not shown) remained unchanged, suggesting that measurement error introduced by the model was negligible or comparable to the error associated with measures taken at the closest monitor.
Stronger positive associations in persons with a high education level were consistent across IMT measurements and across all markers of pollu tion. Larger estimated effects of traffic-related exposure on IMT were observed in men ≥ 60 years of age in all models. We did not find evidence of effect modification by Mediterranean diet, established risk factors for CVDs, or subclinical disease (indicated by medication treatment). Other studies have found heterogeneous associations across subgroups of age, sex, BMI, smoking status, socioeconomic status, town of residence, and other cardio vascular risks factors (Bauer et al. 2010;Künzli et al. 2005;Lenters et al. 2010). The detection of interactions in epidemiologic studies is often underpowered, and testing many interactions can lead to a multiple comparison problem. However, our subgroup analysis was designed a priori, and although our results may be subject to the aforementioned problems, the accumulated evidence over different studies will help to identify susceptible subgroups.
In Girona, exposure to traffic-related pollutants was higher for persons with a high education level. Higher NO 2 concentrations were also found at the most privileged census-tract locations [see Supplemental Material, Figure S4 (http://dx.doi.org/10.1289/ehp.1205146)]. This has been reported before for southern European cities (Cesaroni et al. 2010), where wealthy persons live in downtown areas that are more polluted, in sharp contrast to what has been observed in North America (Gunier et al. 2003) and northern Europe (Chaix et al. 2006), where the most deprived bear the highest air pollu tion concentrations.
The persons with higher levels of education were also younger, had lower blood pressure, BMI, glucose levels, and low-density lipoprotein levels; a higher percent had quit smoking; and a lower percent had never smoked compared to persons with low and medium education levels. Stronger associations between air pollu tion and IMT among those with high versus low education levels were not explained by age interactions. This may indicate higher susceptibility but, more likely, a better detectability among persons with fewer competing risks for athero sclerosis (due to less confounding). It is difficult to evaluate whether this is supported by previous studies given that results are very heterogeneous. No effect modification by education level has been observed before for the association of PM 2.5 with IMT, ABI, or coronary artery calcification (Diez Roux et al. 2008), nor for the association of NO 2 or PM 2.5 with IMT and arterial stiffness (Lenters et al. 2010). Whereas increasing effects of roadway proximity on aortic artery calcification have been reported for increasing income (p trend < 0.01) (Allen et al. 2009).
Associations of PM 2.5 with systemic inflammation markers have been reported to be stronger in men, and more specifically in highly educated men, in Germany (Hoffmann 2009a). Slightly stronger associations with annual PM 10 were estimated among men compared to women in the study of civil servants in London (Tonne et al. 2012). Contrary to our results for men ≥ 60 years of age, stronger associations between PM 2.5 and IMT were estimated for women ≥ 60 years of age in the study in Los Angeles (Künzli et al. 2005). The difference in augmentation index associated with increased levels of NO 2 and PM 2.5 was also higher for young women in Utrecht (Lenters et al. 2010).
Associations with NO 2 or traffic load or intensity at the current address (at the time of examination) were weaker and less precise (data not shown) than associations with 10-year weighted averages across addresses, indicating exposure misclassification. Similar results were observed in the London study in which exposure at current address gave less precise estimates compared to exposure averaged during 1 year before examination (Tonne et al. 2012). This highlights the relevance of using long-term exposure when studying effects on the IMT. On the other hand, our results remained unchanged when using exposure at the address of longest residence compared with 10-year averaged exposure (data not shown). Thus, for settings similar to the Spanish Mediterranean region, in terms of patterns of spatial distribution of NO 2 and low residential mobility (i.e., 80% of persons did not change address in the 10 years), the exposure at the address of longest residence may be a good proxy for long-term exposure.
Limitations of our study include the crosssectional design and the possibility of unmeasured confounding, including confounding related to environmental tobacco smoke, and the lack of information on daily activity patterns to assess time spent at home. Strengths include the large population-based sample size and the availability of data on health and potential confounders, the 10-year address histories used to estimate long-term NO 2 and traffic exposure markers, and the dense NO 2 and traffic monitoring network data used to estimate exposures at each participant's residences.

Conclusions
We found evidence supporting an association between long-term exposure to traffic and subclinical carotid athero sclerosis and high ABI levels in our study population, with stronger associations in persons with a high level of education and in men ≥ 60 years of age. Longitudinal studies are needed to confirm whether air pollu tion contributes to the chronic processes of athero genesis.