The main results regarding air pollution and CIMT progression are presented in Table 3, Figure 1 and Figure 2. In general, both proximity markers of exposure and P2.5 were positively associated with CIMT progression. In the total sample, associations were statistically significant for proximity to highways and of borderline significance for both, PM2.5 (P-value =0.8) and living within 100 m of a highway or 50 m of a main road (p-value: 0.07; see Table 4). For those living within 100 m of a highway (N=23), the CIMT progression was 5.62 µm/year (95% CI: 0.31–10.94; unadjusted model) and 5.46 µm/year (95%CI: 0.13–10.79; main model) faster than among those living further away (Figure 2). As expected, defining a less extremely polluted group as “exposed” (i.e. those living within 50 m of a main road or within 100 m of a highway; N=238), the main model coefficient was lower (1.64; 95% CI: −0.15 to 3.42) (see Table 4). An increase of 10 µg/m3 in the home outdoor PM2.5 concentration was associated with a statistically insignificant 1.74 and 2.53 µm/year faster progression in CIMT in the unadjusted and the main adjusted model, respectively (Figure 1).

Main effect estimates were not appreciably altered by inclusion or exclusion of a range of covariates or adjustment for baseline CIMT, and adjusted coefficients were without exception slightly larger than the crude associations. Baseline CIMT and progression were mildly negatively correlated (r=−0.09; p<0.01). Adding both, PM2.5 and proximity to a highway to the main model had little impact on the coefficients although confidence intervals slightly increased (Table 3).

Table 4 (and Table S3 in the supplement) present the investigation of possible interactions between air pollution and other factors. The association of CIMT with ‘proximity to highways’ as well as with ‘proximity to highways (100 m) or a main road (50 m)’ was substantially larger and statistically significant among the socially deprived (as measured by low income). These effect modifications were statistically significant (p for interaction <0.05). A similar pattern emerged for PM2.5 but associations and interactions were not statistically significant (Table 4). Stratification by randomization status (active intervention vs. placebo) revealed larger and statistically significant coefficients for both, proximity to highways (100 m) and PM2.5, among the subjects randomized to active treatments. For the highway proximity, this interaction was statistically significant. The coefficients in the placebo group were not statistically significant and were five to ten times smaller relative to the treatment group (see Table 4 and Figure 3 & 4).

While our progression findings have not been previously reported, the results agree with cross-sectional observations in humans [8], [9] and with several prospective animal experiments where rats, mice, and rabbits exposed to ambient particulate matter developed atherosclerotic plaques and calcification [5], [6], [7], [30]. These animal studies – one placing mice adjacent to a Southern California highway [6] – reported a range of effects relevant to atherogenesis and significant progression of atherosclerosis, which was possibly modified by endogenous or exogenous factors such as genetics and diet [7].

Despite pooling five trials, the major weakness of the study remains the limited sample size. The percent of people living very close to highways is small, in general, and was only 1.6% in our more affluent population. Thus, the strongest results relied on only 23 ‘exposed’ subjects (i.e. within 100 m of a highway). In contrast, results of the PM2.5 analyses were not affected by this limitation as this exposure metric is defined on a continuous scale. The PM2.5 estimates did not reach statistical significance in the full data, but were clearly significant among the pool of four instead of five studies (excluding BVAIT) as well as among the on-trial treatment group. In fact, the most intriguing and challenging aspect of our results are the sub-group findings. All subgroups tested were identified a priori, based on prior studies and hypotheses, and the confirmation of modifiers of effects underscores the plausible notion that there exist groups who are more susceptible to oxidative and inflammatory air pollutants. In line with other U.S.-based air pollution studies, the effects were stronger among the socioeconomically disadvantaged, a possible marker for concomitant adverse environmental exposures, poor diet, and a more stressful life [19].

In line with our previous cross-sectional findings, proximity to highways appeared to be more strongly related to CIMT progression in women but the difference did not reach statistical significance (Table S3 in the supplement). With the limited sample size, it was however difficult to evaluate this further. Similarly, to evaluate whether the apparent but not significant differences in the trial-specific estimates reflect differences in the distribution of susceptibilities among the populations, larger sample size would be needed. The trials had high levels of quality control and identical standards that yield a high degree of comparability in the CIMT data across all five trials [17], [26]. However, the different objectives and consequent recruitment procedures of each trial – all based on volunteers – increases the inherent heterogeneity in our data and limits our ability to generalize results.

The interaction between randomized ‘treatment on trial’ and the atherogenic effect of air pollution was apparently present across heterogeneous interventions ranging from vitamins (VEAPS and BVAIT) to the regulation of diabetic metabolism (TART) and hormones (EPAT and WELL-HART) (Figures 3 and ​and4).4). Given the double-blind randomized design applied in all trials, and treatment status indeed not being associated with pollution it is not very plausible to explain the interaction finding as chance alone.

With the exception of WELLHART, treatments affected CIMT progression in all trials although not necessarily in the entire populations [11], [12], [13], [14], [15]. This means that the treatments did affect vascular metabolism and homeostasis in a way that resulted in changes in the artery wall structure, measured with CIMT progression. Periods of metabolic change may be a vulnerable window in the presence of other pro-atherogenic factors [31], such as hypothesized for ambient particulate matter [2], [4], [32]. All interventions used in these trials have both pro- and anti-atherogenic properties, and the postulated oxidative and inflammatory effects of ambient air pollutants may interact with the metabolic effects of these interventions in a complex manner. Congruent with this hypothesis are studies in mice where effects of ambient particulate matter on atherogenesis were stronger among those fed with a high fat chow as compared to those under a normal diet [7]. One would need larger studies to corroborate these interactions for the various on-trial treatments. Moreover, the interaction by treatment status was only apparent for PM2.5 and ‘living close to highways’ (with 23 subjects ‘exposed’), but not so for ‘living close to highways or main roads’. In light of this inconsistency, one may question a causal interpretation of the observed effect modification due to treatments. However, as mentioned above in the Methods, living within 50 m of main roads includes 215 subjects in the ‘exposed’ group with lower traffic-related pollution than those 23 participants living within 100 m of a highway. Thus, effect stimates would be expected to be smaller. Statistical power to evaluate interactions for this marker of exposure is not sufficient.

Similarly, due to limited sample sizes, the statistically significant interaction between on-trial treatment and the effects of traffic-related pollution cannot be evaluated further in each trial separately. Thus, it is not clear whether the same interactions would prevail in larger studies or be restricted to a few on-trial treatments. Below, we conjecture about explanations of this interaction trial by trial.

Air pollution associations were strongest in TART with only slightly larger associations in the treatment group. All TART subjects were patients with type 2 diabetes – a pathology with a well known atherogenic propensity and with some preliminary evidence of higher susceptibility to adverse cardiovascular effects of air pollution [4], [11]. The trial tested a peroxisome proliferator-activated receptor agonist (PPARγ), namely thiazolidinedione (TZD). Activation of PPARγ inhibits inflammatory cytokines, cellular proliferation and migration, and expression of cellular adhesion molecules. However, the stimulation of macrophages into foam cells by promoting lipid uptake is a known pro-atherogenic property of PPARγ activation. These effects may dominate if exposure to air pollution is increased, e.g., recruitment of precursors of macrophages as observed in rabbits [5] or the oxidation of lipids [6], [33]. As shown in TART, the balance between pro- and anti-atherogenic effects of TZD depends on host factors and this may hold true for the interaction with air pollution as well. In fact, TZD had protective effects only among those with advanced atherosclerosis (i.e. thicker CIMT) at baseline, and the considerable heterogeneity of responses to TZD supports the notion of complex interactions between exogenous factors (e.g. TZD or air pollution) and reactions in the vascular bed [34].

VEAPS tested the atherogenic effects of vitamin E supplementation. While known as an anti-oxidant, vitamin E also has pro-oxidant and pro-atherogenic effects, and those appeared to dominate in the treatment group where CIMT progression was 4.0 versus only 2.3 micrometers per year in the placebo group (p=0.08) [15]. Oxidized vitamin E may act as a pro-oxidant in the artery wall and these proatherogenic properties may be amplified by pro-inflammatory effects of air pollution. Similarly, interactions between treatment (vitamin B), other endogenous factors and environmental effects may apply to BVAIT; effects of vitamin B were limited to individuals with elevated baseline homocysteine levels. Homocysteine is linked to atherogenic pathways, and interactions between ambient air pollution and homocysteine have been reported previously [35].

EPAT tested the effects of unopposed estrogen replacement therapy (ERT) with 17-beta-estradiol among postmenopausal women [12]. ERT significantly reduced CIMT progression by 5 micrometers per year, but this beneficial effect was seen only in women who did not receive lipid-lowering treatment, namely with statins whereas among this latter group, CIMT changed far less during the EPAT follow-up. Whether the somewhat larger PM2.5 coefficient in the treatment group was a chance finding, and how it interacted with lipid-lowering treatment cannot be determined with the available sample size. However, it is well known that the atherogenic effects of hormones underlie rather complex mechanisms which depend on a range of host factors including time since menopause, concomitant diseases, or type, dose and route of drug administration [36].

The WELLHART trial tested hormones in women with established coronary artery disease, at a later stage of menopause [13]. While treatment had no effect on coronary artery atherosclerosis, the interaction of air pollution with on-trial treatment status was also substantial in this population. The interaction between hormonal factors, ambient air pollutants, and atherogenesis needs further investigation.

An interesting finding is the similarity of coefficients from the one-pollutant and two-pollutant models. As mentioned in Methods, the derivation of the PM2.5 surface did not include any markers of local traffic while the proximity measures capture this local exposure space. Thus, the two-pollutant model results suggest independent associations of ambient PM2.5 (albeit not statistically significant in all groups) and proximity to traffic with atherosclerosis (Table 3). This is in line with two different interpretations. First, it supports the idea that different components of the complex mixture contained in urban air pollution act, in part, independently through complementary mechanisms [4], [32]. Proximity to highways may be a marker for exposure to high loads of ultrafine particles and other highly redox-active pollutants, in particular diesel particles [37], and constituents affecting the airways and alveoli leading to systemic inflammatory responses [38] and atherogenesis [6]. The larger particles (i.e. PM2.5) may result in inflammatory reactions in the small and upper airways alike, and both types of pollution may independently enhance systemic inflammation [38]. Second, the independent effects may indicate that both exposure assessment approaches capture similar types of pollution but on different spatial scales and concentration levels; ‘proximity’ would characterize the most extreme local conditions (hot spots) while PM2.5 captures the additional contrast occurring between geographic areas [37]. In contrast to other areas, e.g. the U.S. East coast, PM2.5 in the Los Angeles area, is to a large extent the consequence of primary and secondary pollutants from traffic, and the proximity measure is, by definition, a marker for traffic-related exhaust emissions and/or re-suspended local pollutants. The results, therefore, point in the direction of traffic-related pollution being an atherogenic health hazard. This is in line with the Southern Californian animal studies where experimental chambers were placed adjacent to a highway. The traffic-related particles enhanced atherogenesis in mice [6].

Longitudinal analyses of change in biologic markers raise some contentious statistical issues with regard to the adjustment for the CIMT levels observed at baseline [39], [40]. Our model estimates were not appreciably altered by adjustment for baseline CIMT and results were very stable across a range of modelling assumptions; thus, the findings are unlikely the result of modelling artefacts. In general, we used baseline covariate information to adjust the multivariate models rather than the change in some of these variables. However, individual follow-up time was fairly short (2–3 years); thus, change in lifestyles or other factors are not expected to be substantial. We have some indirect evidence that a potential change in an important lifestyle, namely giving up smoking during follow-up, was probably not relevant in our analyses; associations of pollution with CIMT progression was not different in smokers, and only 6% of participants were smokers at baseline.

Our analyses do also not take into account possible changes in exposure due to change in pollution levels over time. Exposure was assigned based on the year 2000 information. While levels of pollution fluctuate year by year, with general downward trends, over decades, the key question relevant for our analyses is whether the spatial contrasts of the year 2000 would reflect those of other years. Fortunately, spatial contrasts are more stable over time and many long-term air pollution studies capitalize on the fact that spatial contrasts and traffic data of a specific year will properly approximate exposure contrasts for several years. We are unable to investigate the possible impact of this type of exposure misclassification because departure of assessment from the year 2000 is, in our case, a correlate of trial as each trial took place in different periods (shown in Table 1). As discussed, trials are inherently different in many other respects.

We had no noise data available, thus, a full assessment of the correlations between noise and pollutants – in some studies reported to be lower than expected [41] – was not possible, nor could we evaluate the potential confounding of traffic-relate noise. Future studies should investigate this in more detail.

We were also not able to integrate in the exposure assignment changes in residence that may have occurred during the trial follow-up. This information was not available nor could we approach all former participants as some trials happened several years ago. However, six years of residential history was available among a subsample of 245 subjects. The vast majority (N=210) had not moved for six years; we therefore conclude that exposure misclassification due to ignored moving patterns were not a major limitation. Moreover, it is unlikely that moving patterns were jointly correlated with CIMT progression and air pollution at the address used in our exposure assignment, thus, biases from this exposure misclassification would be expected to go toward null.

CIMT is the earliest detectable anatomic change in the development of atherosclerosis and progression of CIMT is predictive of clinical cardiovascular events [28], [29]. The magnitude of our effect estimates was similar to the effects of on-trial treatments on CIMT progression. Comparison is, however, hampered by the differences in absolute progression between trials due to the differences in inclusion criteria. The mean effect of a 5.4 micrometer greater progression per year among those living within 100 m of busy roads or per 23 ug/m³ contrast in PM2.5 is approximately twice the average progression as observed across these trials. Our results for a 10 ug/m³ contrast in PM2.5 being associated with 2.5micrometer (total data) to 4.4micrometer (treatment groups) increase in annual CIMT progression would translate into 10–30% higher rates in coronary events [26]. This is well within the range of the results reported by the Women's Health Initiative where cardiovascular events increased by 24% (95%CI: 9–41%) for the same contrast of 10 ug/m³ ambient PM2.5 [42]. While these studies and settings are different, our observed effect appears to be of biologically plausible size [13].

Interactions between anti-atherogenic treatments such as lipid-lowering prescriptions and ambient air pollution may be a major concern and research challenge as some treatments may influence mechanisms relevant in the effects of air pollution. As we used existing data we were unable to fully investigate whether such prescriptions taken at baseline or during follow-up or both interacted with the main effects. As shown in the supplement (Table S3), subjects reporting lipid-lowering treatments at study entry showed substantially stronger associations between CIMT progression and pollution than those without such treatments or those who newly received lipid lowering prescriptions during the follow-up period of the trial. Again, statistical power was insufficient to further elaborate on these patterns which may also be a chance finding.

Text S1

Table S1

Table S2

Table S3