Cardiovascular health, traffic-related air pollution and noise: are associations mutually confounded? A systematic review

Studies have shown that exposure to road traffic noise and air pollutant emissions can lead to adverse health effects such as annoyance (World Health Organisation (2011), sleep disturbance (Pirrera et al. 2010), respiratory problems (Health Effects Institute Panel on the Health Effects of Traffic-Related Air Pollution 2010) and cancer (Beelen et al. 2008). A few studies (Finkelstein et al. 2004; Gan et al. 2010; Hoffmann et al. 2006, 2009; Maheswaran and Elliott 2003) have also reported associations with cardiovascular (CV) outcomes. CV health effects could be explained either by noise or air pollution associated with traffic.

On the one hand, there is evidence linking traffic noise to ischemic heart diseases (World Health Organisation 2011; Babisch and Kamp 2009; Babisch 2006) and hypertension (World Health Organisation Regional Office for Europe 2011; van Kempen and Babisch 2012). On the other hand, there is also evidence linking traffic-generated air pollution to CV diseases (Hoek et al. 2002; Brook et al. 2010). However, in a recent report of the Health Effects Institute Panel on the Health Effects of Traffic-Related Air Pollution 2010), the evidence of the effects of traffic-related air pollution on cardiovascular mortality was considered suggestive, but not sufficient. Nonetheless, there is increasing evidence connecting air pollution to overall cardiovascular mortality (Chen et al. 2013), myocardial infarction (Nuvolone et al. 2011; Rosenlund et al. 2006), atherosclerosis (Adar et al. 2013) and atrial fibrillation (Liao et al. 2010). Although noise or air pollution can confound the effect of each other, the underlying physiological mechanisms are likely different.

A few mechanisms are postulated to explain the effect of noise on CV diseases. Noise can act as a general stressor disturbing the body homeostasis through the “stress syndrome” (Babisch and Kamp 2009; Westman and Walters 1981; World Health Organisation Regional Office for Europe 2011; Amato et al. 2010; Gan et al. 2012). Noise can induce stress by two different pathways. The stress response generated by the direct pathway consists of a neural activation by the noise. In the indirect pathway, the activation of the stress response is created by a cognitive interpretation of the noise (Westman and Walters 1981; Gan et al. 2012). The activation of this pathway can be influenced by the perception of the noise, the perceived control over the sound and noise sensitivity of the subjects (European Environment Agency (EEA) 2010). There are also pathways by which noise could increase the risk of cardiovascular diseases. One of those is the disruption of sleeping patterns. Studies have associated the lack of reduction of blood pressure (BP) during the night (“BP dipping”) caused by noise with an increased risk of cardiovascular outcomes (Vardeny et al. 2011; Haralabidis et al. 2011). It is also suggested that short sleep durations may result in higher ghrelin and lower leptin concentrations (Taheri et al. 2004). The deregulation of those hormones linked with appetite regulation could potentially lead to obesity and higher risk of cardiovascular diseases.

For air pollution, several pathways are hypothesized to explain its impact on CV diseases. Firstly, air pollutants could generate an increase in lung oxidative stress and inflammation. Some pollutants could also migrate through the pulmonary epithelium into blood. Those pathways could lead to a systemic and vascular inflammation, increasing the risk of hypertension and thrombogenesis. Another hypothesized mechanism is the activation of the pulmonary reflex by particulate matter leading to the activation of the sympathetic system. A chronic activation of this system could lead to hypertension, plaque instability and cardiac arrhythmias (Brook et al. 2004; Burgan et al. 2010; Mills et al. 2007).

As motor vehicles are the predominant source of both air pollution and noise in many cities (Allen and Adar 2011), the reported associations between road traffic exposure and CV diseases could be influenced by a confounding or interaction between those two pollutants. Yet, few studies have aimed to untangle the possible effects of road traffic noise and air pollution. This is fundamental to better steer public health interventions and policies aimed at reducing CV effects of road traffic. For example, if road traffic induces CV effects through noise levels, then regulations to reduce car air pollutant emissions may not tackle the problem and zoning by-laws could be more appropriate. In this article, we reviewed epidemiological evidences that looked at the confounding effect of one traffic-related exposure (noise or air pollution) on the association between the exposure to traffic-related noise or air pollution and CV outcomes.

Figure 1 presents a flowchart describing the stages of selection for the studies used in this review. Our keyword search yielded 221 starting articles ("Appendix"), which were reduced to 16 articles by applying the various inclusion and exclusions criteria. A large number of irrelevant articles had to be sifted by hand because of the double meaning of the terms sound and noise. We added four articles identified by snowballing and expert consultation. Of these 20 articles, a final 11 studies were discarded because it was impossible to assess the impact of the confounding variables of interest (Fig. 1). We did not find any document in the gray literature that could benefit this review.

Nine studies evaluating the relationship between CV outcomes, noise and traffic-related air pollution met our criterion (Beelen et al. 2009; Clark et al. 2012; de Kluizenaar et al. 2007; Dratva et al. 2012; Gan et al. 2012; Selander et al. 2009; Sorensen et al. 2011, 2012a, b). Of those articles, one assessed CV disease mortalities in general (Beelen et al. 2009), four assessed ischemic heart diseases (Beelen et al. 2009; Gan et al. 2012; Selander et al. 2009; Sorensen et al. 2012a), two assessed cerebrovascular diseases (Beelen et al. 2009; Sorensen et al. 2011), four looked at blood pressure or hypertension (Clark et al. 2012; Dratva et al. 2012; Sorensen et al. 2012b; de Kluizenaar et al. 2007), and one article was identified for both heart failure and cardiac dysrhythmia (Beelen et al. 2009).

Tables 1 and 2 present the associations between cardiovascular outcomes and either road traffic air pollution or noise. The tables also summarize the population characteristics of each study, the health outcomes monitored, as well as the noise and air pollutants indicators used. When available, the background exposure levels to pollution and noise of the subjects were reported.

Six studies used day–evening–night noise levels (L _den), as indicators of exposure to noise levels from road traffic (Beelen et al. 2009; de Kluizenaar et al. 2007; Gan et al. 2012; Selander et al. 2009; Sorensen et al. 2011, 2012a, b). Others studies used equivalent noise levels over 24 h (L _eq,24h) (Selander et al. 2009), night noise levels (L _night) (Dratva et al. 2012) or daily noise levels (L _day) (Clark et al. 2012; Dratva et al. 2012). All studies reported used dispersion noise models, but only four reported validation information on the model used (Beelen et al. 2009; Sorensen et al. 2011, 2012a, b). In those studies, the average difference between the estimated noise levels and measured noise levels was 3.1 dB or less. Regarding exposure to air pollutants, seven studies used nitrogen oxide (NO_x or NO₂) as an indicator (Clark et al. 2012; Dratva et al. 2012; Gan et al. 2012; Selander et al. 2009; Sorensen et al. 2011, 2012a, b). Two articles reported results based on particulate matter (PM₁₀ or PM_2.5) (de Kluizenaar et al. 2007; Gan et al. 2012). Two studies used measurements of black smoke or black carbon to account for road traffic pollutants (Beelen et al. 2009; Gan et al. 2012). Of those studies, two applied land use regression models (Beelen et al. 2009; Gan et al. 2012) to assess the air pollution exposure, while the other studies employed dispersion models (Clark et al. 2012; de Kluizenaar et al. 2007; Dratva et al. 2012; Selander et al. 2009; Sorensen et al. 2011, 2012a, b). As noted in Tables 1 and 2, seven out of the nine studies reviewed reported validation information on their air pollution model (Beelen et al. 2009; Clark et al. 2012; Dratva et al. 2012; Gan et al. 2012; Sorensen et al. 2011, 2012a, b). The Clark et al. study reported an average difference between the estimated and the measured pollutant levels of 2.4 ppb. The remaining studies presented R ² that ranged from 0.52 to 0.75. All articles presented average yearly outdoor exposure for both noise and air pollutant levels as a proxy of individual exposure. The correlations between traffic noise and air pollutants in the studies reviewed range from 0.16 to 0.72. It is surprising that studies where both noise and air pollutants were modeled, relying on some of the same variables (e.g., distance to traffic source), could generate low and disparate correlations.

This review aimed to assess the confounding effects of one traffic-related exposure (either noise or air pollutants) on the association between its counterpart and cardiovascular outcomes. In general, the results of the nine studies reviewed here showed that when associations between noise and CV diseases were adjusted for air pollutants, modifications of the point estimates for cardiovascular diseases were <10 %, after controlling for the air pollutants, with the exception of the studies by de Kluizenaar et al. (2007), Beelen et al. (2009) and Dratva et al. (2012) where the CIE was higher than our cutoff point for confounding effects. The Beelen et al. (2009) study reported a marked decrease of the strength of the association after adjustment for air pollution and traffic intensity. Yet, the simultaneous adjustment for traffic intensity and black smoke makes the evaluation of confounding by black smoke difficult in this study. Nonetheless, no association between road traffic and CV outcomes (before and after adjustment) was found in both the Dratva et al. (2012) and the Beelen et al. (2009) studies, rendering the CIE meaningless. By its definition, a confounder must modify the association between the exposure and the outcome. To confound, such associations must be present at least before or after adjustment. de Kluizenaar et al. (2007) reported CIE of <2 % in both the Groningen sample and the prevent cohort. However, CIEs nearing 10 % were observed in the 45–55 years subgroup, which could indicate the presence of a small confounding between the two exposures in this particular subgroup. Similar findings were found for associations between air pollutant levels and CV diseases, although the number of studies was limited (N = 3): controlling for noise levels either changed the point estimates for CV diseases by <10 % or the study did not present an association between the exposure and the outcome (before and after adjustment). Only the Sorensen et al. (2012b) study presented an indication of confounding by traffic noise in the association between NO_x levels and blood pressure. Nonetheless, overall these findings suggest an independent effect of noise and air pollution on CV diseases, particularly ischemic disease for which there were a greater number of studies. The review also points to the absence of comparability between studies. Most studies were difficult to compare because different noise or air pollution indicators were used, the pollution levels were assessed using different techniques and very few studies assessed comparable health outcomes.

In this review, we also tried to verify if the impacts of noise and air pollutants on CV were subject to the same interactive effects. Though both studies (Gan et al. 2012; Selander et al. 2009) that assessed interaction effects did not find any effect, one cannot conclude that there was an absence of interaction effect between noise and air pollutant levels with so few studies. This is particularly true in the light of the point raised by Selander et al. (2009) that the interaction analysis might have lacked power. It should also be noted that both studies used different noise indicators, the air pollutant or the cardiovascular outcome, to identify possible interaction effects. Those results do suggest, though, that if a multiplicative interaction exists, it is likely to be small.

Due to limitations of the literature, we cannot conclusively ascertain the independence of the effects of the two risks on any CV health outcome. Nonetheless, the results reported tend to indicate that the impacts of traffic noise and air pollution on cardiovascular outcome are distinct, or at least that they are not completely dependent on one another. Furthermore, the correlation between noise and air pollutant levels does not seem to influence the confounding effects. A wide range of correlations between noise and air pollutants were reported in the studies reviewed and this could be partially explained by differences in the urban structure at each location (building height, distance of buildings to sidewalks, street width, traffic intensity and distance to major road). This would suggest that confounding between traffic-related noise and air pollution is a study-dependant issue. However, high correlations between noise and pollutant levels were not associated with greater confounding effects in the studies reviewed. Additionally, the quality of the exposure assessment of the confounding variables and the quality of the studies (number of biases) do not seem to influence the confounding effects. However, the reported CIE seemed higher in studies that used a cross-sectional design. This may be because these studies mainly used linear regressions and presented regression coefficients. In fact, the approach that we used to assess confounding effects was developed for risk ratios and may not be applicable for linear regression. Nonetheless, this approach has been suggested by some authors in linear regression text books (Vittinghoff et al. 2012).

This review is subjected to a few limitations. First, only a small number of studies were available to be reviewed, which reduced the strength of our findings. Secondly, with any systematic review, the possibility of publication bias is present. We tried to minimize this bias by searching the gray literature and reviewing some non-English publications, but this bias cannot be excluded. We also tried to minimize the selective reporting bias, which could be a major one in this review. Indeed, the principal objective of most of the studies reviewed was not to assess the confounding effect between noise and air pollutants from road traffic. As shown in Fig. 1, more than half of the studies presenting the basic characteristics to be included did not report the impact of the co-variables of interest or the regression coefficients. It is also possible that some authors assessing the effect of either noise or air pollutant levels on CV diseases found that the corresponding co-variable was not significant and did not report it. This omission would, however, strengthen our results, as it would seem to indicate the independence of the effects of both exposures. Finally, the strategy used to assess the confounding effect in studies could also be criticized. The 10 % variation in the estimate is an arbitrary cutoff point that does not necessarily rule out confounding or inform on the statistical variability. Nonetheless, this cutoff point is widely used in the literature and was identified as the least biased in a simulation presented in Maldonado and Greenland (1993), in the absence of prior knowledge of confounders. Another potential limitation in our assessment of confounding is that the 10 % cutoff implies that effects are linear. In the case of nonlinear effect estimates, the absence of CIE does not preclude confounding (Janes et al. 2010). This might be the case if noise had a threshold effect, and future studies should address this.

To assess clearly the presence of confounding, future studies should use coherent noise and air pollution indicators. Those indicators should be chosen according to the effects examined. It would be better to use maximum noise levels (L _max) or equivalent noise over 1 h (L _eq,1h) to assess acute effects, and L _eq,24h to assess chronic effects. The use of L _den or L _DN as the noise metrics in the studies included in this review could be contested. Those noise metrics created to assess annoyance increase the weight noise levels occurring in evening or at night and are therefore not representative of the actual sound exposure. The source of the exposure is also an important factor to take into account while choosing an indicator. For a surrogate of all pollutants emitted by on-road traffic, one should use individual pollutants such as black carbon, NO_X and ultrafine particulate that are more source specific. More studies on the relationship between noise outdoor exposure levels and personal exposure levels should be conducted. This is important given that this relationship could differ between noise and air pollution. Furthermore, future studies should present adjusted results for both noise and air pollution modeled separately, so that the impact of the pollutant that was controlled for can be assessed. As shown in Tables 1 and 2, only few studies provided validation information for the exposure models used in their study. The absence of such information prevents the reader from judging the quality of the exposure assessment and thereby precluding the reader to judge from assessing the quality of the adjustments made. Finally, more studies are needed to find out if the confounding effect is specific to subcategories of CV outcomes. Ideally, those studies would need an epidemiologic design enabling them to assess the chronic effects of both traffic-related pollutants. We also recommend that those studies focus their research on ischemic heart diseases, hypertension or the fluctuation in blood pressure, for which mechanisms should yield more conclusive results.