A systematic literature search was performed on PubMed and Web of Knowledge, who were last accessed on 1 March 2015, with no restriction for time of publication. The following search strategies were used: ('particular matter' OR 'air pollution' OR 'PM₁₀' OR 'PM_2.5') AND ('intima media thickness' OR 'carotid intima media thickness' OR 'carotid intima media' OR 'artery intima media thickness' OR 'intima media thickness measurement' OR 'intima media thickness cardiovascular' OR 'carotid intima media thickness measurement' OR 'carotid intima media thickness cardiovascular'). We also considered references found in the literature search. Two investigators (EBP and NM) read all the papers and extracted and computerized the relevant information independently (Tables 1 and 2). This meta-analysis complies with the preferred reporting items of the statement for Meta-analysis Of Observational Studies in Epidemiology (MOOSE).[25]

Selection of the studies was based on the research question, inclusion and exclusion criteria (Fig 1). All studies were reviewed by title and abstract and, if eligible for inclusion, by reading the full text. All types of studies and designs were considered for inclusion. Nevertheless, studies needed to report originally collected data. Therefore, reviews, editorials and debates were excluded.

We selected the studies that used particulate matter with an aerodynamic diameter of 10 μm or less (PM₁₀) or 2.5 μm or less (PM_2.5) as indicators of air pollution. Studies using only other air pollution measures or indicators were excluded. Furthermore, only studies measuring carotid IMT were included.

Out of the 42 initially identified articles, 9 reported a cross-sectional association between CIMT and PM₁₀[22] or PM_2.5[19–21, 23, 24, 28] and 3 reported longitudinal associations between CIMT progression and PM_2.5[23, 26, 27]. If a group published two or more papers based on the same study population,[23, 28–30] only the publication that provided the most detailed information was included. We selected the results adjusted for gender, age, and BMI as well as for other known correlates of CIMT such as cholesterol levels and smoking status, if provided.

A meta-analytical combined estimate was derived from the point estimate of each separate study weighted by the inverse of the variance (1/SE²). In the case that only data for PM₁₀ was available (n = 1), we converted the point estimate under the assumption that PM₁₀ consist for 70% of PM_2.5.[31] The combined estimate was computed using a random-effects model and is presented as a percent change in CIMT associated with a 5 μg/m³ higher long-term PM_2.5 exposure for cross-sectional associations. Similarly, we computed a combined estimate based on the studies reporting a longitudinal association between the progression of CIMT and exposure to PM_2.5. Results of this additional meta-analysis are presented as μm change in CIMT per year for a 5 μg/m³ higher long-term PM_2.5 exposure.

The sensitivity of the cross-sectional findings was examined by recalculating the combined estimate while excluding one study at a time in order to evaluate the influence of individual studies on the combined effect size. If the combined estimate, excluding one study, lies outside the confidence interval of the overall estimate, the excluded study has a disproportionate influence on the combined effect size. Further, between-study heterogeneity was examined using the Cochran Q and I² test. We plotted the association size against the SE of the study in order to investigate publication bias. This should result in a funnel shape (funnel plot) if there is no bias. As an additional sensitivity analysis, we recalculated the combined estimate while using the overall estimate of the European Study of Cohorts for Air Pollution Effects (ESCAPE), replacing the separate estimates. As a final sensitivity analysis, we replaced the point estimates of the overall, between-city, association by the within-city estimate, presented by Adar et al.[23]

Of the 42 studies reviewed, 26 were excluded after review of the title and/or abstract; 19 reported associations with other types of exposures than PM_2.5 or PM₁₀, or another outcome measure than CIMT. Three were reviews, one was on study design, one was performed in an animal model, one reported technical aspects of IMT measurements and one on PM_2.5 modeling approaches. After assessment of the full-text, 7 additional studies were excluded; three were based on the same study population[28–30], two were case-control studies of which no relevant association size could be computed[32, 33] and 3 reported associations with other air pollution measurements or indicators than PM_2.5 or PM₁₀.[34–36] We identified a set of six studies which investigated the cross-sectional association between CIMT and PM and three longitudinal studies on CIMT progression in association with PM (Fig 1).

The selection of six cross-sectional studies includes four longitudinal studies investigating baseline cross-sectional associations[20–23], one study reporting baseline cross-sectional analyses of two trials[19] and one reporting cross-sectional results from four different cohort studies within ESCAPE.[24]

The Heinz Nixdorf Recall (HNR) cohort, from the publication by Bauer and colleagues[20], is also one of the cohorts included in ESCAPE.[24] Since different modeling approaches were used for estimating the participants’ exposure to PM_2.5, differences between the reported point estimates of the two publications are found. We included the point estimates of the most recent publication from ESCAPE by Perez et al.[24] in the main meta-analysis and performed a sensitivity analysis using the point estimate from the publication by Bauer et al. Therefore, all six studies are listed in chronological order in Table 1.

The five publications included in the main meta-analysis comprised 18,349 participants from 8 cohort studies. The majority of the study populations had an even gender distribution (range: 34 to 55% women) and an average age of 57 years. The average exposure to PM_2.5 in the different study populations ranged from 7.2 to 20.7 μg/m³ and CIMT averaged (SD) 0.73 (0.14) mm.

All studies used modeled PM concentrations based on the participant’s residence averaged over one year prior to the CIMT measurements. Whenever possible, preference was given to mean IMT measurements of the common carotid artery. In all reports, results were adjusted for gender, age and smoking status. Most studies also considered additional covariates including BMI,[20–22, 24] blood pressure[19, 21] and cholesterol levels.[19, 21, 23]

The combined estimate showed a 1.66% increment (95% CI: 0.86 to 2.46; P<0.0001) in CIMT for each 5 μg/m³ higher long-term PM_2.5 exposure (Fig 2). Cochran Q statistics did not indicate incomparability of the study’s results (P = 0.34). Exclusion of Adar et al.[23] resulted in a drop in the combined estimate to 1.48% (95% CI: 0.35 to 2.62; P = 0.01), whereas it increased to 1.78% (95% CI: 1.07 to 2.49; P<0.0001) when omitting the results of the IMPROVE-Stockholm cohort from ESCAPE.[24] None of the studies moved the combined estimate outside the confidence interval of the overall estimate. Including the point estimate of the study by Bauer et al.[20] as a result from the HNR study instead of the result from ESCAPE, increased the combined estimate to 1.99% (95% CI: 0.95 to 3.04; P = 0.0002). Using the overall point estimate of ESCAPE, instead of the results from the 4 subcohorts, increased the combined estimate to 1.73% (95% CI: 0.92 to 2.54; P<0.0001). In a final sensitivity analysis, we replaced the overall, between-city, associations by the within-city associations as reported by Adar et al.[23] This lowered the combined estimate to 1.34% (95% CI: 0.30 to 2.38; P = 0.01) but the estimate remained within the confidence interval of the overall estimate of the main meta-analysis. The funnel plot did not provide indications of publication bias (Fig 3).

We identified three longitudinal studies, comprising 7,268 participants, which investigated the association between progression of CIMT and PM_2.5 of which the characteristics are listed in Table 2.[23, 26, 27] The study populations had an even gender distribution (range: 51 to 63% women) and an average age of 56 years. The average exposure to PM_2.5 in the different study populations ranged from 4.1 to 20.8 μg/m³, CIMT averaged (SD) 0.71 (0.15) mm and average CIMT progression ranged from 2 to 14 μm per year.

The combined estimate showed a 1.04 μm per year (95% CI: 0.01 to 2.07; P = 0.048) greater CIMT progression for each 5 μg/m³ higher long-term PM_2.5 exposure (Fig 4). Cochran Q statistics did not indicate incomparability of the study’s results (P = 0.90). However, when replacing the between-city associations by the within-city associations as reported by Adar et al.[23], Cochran Q was significant (P = 0.0025) and the random combined estimate changed to 3.64 μm per year (95% CI: -1.21 to 8.50; P = 0.14).