The Association between Air Pollution and Subclinical Atherosclerosis

Rivera et al. (2013) investigated the association between air pollution and subclinical atherosclerosis by using carotid intima media thickness (IMT), ankle–brachial index (ABI), and several indicators of air pollution. As a main outcome, air pollution was positively associated with an ABI of > 1.3, and also with changes in IMT. In contrast, they observed no significant association between air pollution and an ABI of < 0.9. 
 
I have some concerns about their study (Rivera et al. 2013). First, the study included a small number of subjects with ABIs 1.3 (56 and 116, respectively). The authors used multinomial logistic regression analysis; for the full-adjustment model, > 16 air pollution variables were used. There is a limitation in the number of independent variables appropriate for multiple logistic regression analysis (Novikov et al. 2010; Peduzzi et al. 1996), and enough events should be included to maintain statistical power for multivariate analysis. According to the criteria that at least 10 events per variable are required to keep stable estimates (Peduzzi et al. 1996), Rivera et al. (2013) needed ≥ 170 events with an ABI 1.3 for their analysis. 
 
Second, Rivera et al. (2013) used systolic and diastolic blood pressure to adjust for the relationship between air pollution and indicators of atherosclerosis. But multicollinearity among independent variables should have been considered in the analysis (York 2012). 
 
Finally, Rivera et al. (2013) could not clarify the lack of association between ABI < 0.9 and indicators of air pollution. An ABI of 0.9–1.0 is also associated with cardiovascular risk (Ono et al. 2003). Therefore, the association between air pollution and subclinical atherosclerosis should be evaluated by selecting a higher cut-off value of ABI, such as 1.0. This procedure will increase the number of events for multivariate analysis. 
 
Other researchers have reported a significant association between air pollution and IMT (Bauer et al. 2010; Diez Roux et al. 2008). Lenters et al. (2010) also examined the association between air pollutants and indicators of vascular damage but observed no association between air pollution and IMT. Lenters et al. (2010) used nitrogen dioxide (NO2), black smoke, particulate matter ≤ 2.5 μm in aerodynamic diameter (PM2.5), and sulfur dioxide (SO2) as indicators of air pollution, and they used pulse wave velocity and augmentation index in addition to IMT as indicators of vascular damage. Traffic intensity and proximity of residence to roads were also used as indicators of air pollution. Lenters et al. found significant associations only between NO2 and pulse wave velocity and augmentation index and between SO2 and pulse wave velocity. Because contradictory results for this association have been reported, further longitudinal studies are needed to assess this association.

The Association between Air Pollution and Subclinical Atherosclerosis http://dx.doi.org/10.1289/ehp.1307403 Rivera et al. (2013) investigated the association between air pollution and subclinical athero sclerosis by using carotid intima media thickness (IMT), ankle-brachial index (ABI), and several indicators of air pollution. As a main outcome, air pollution was positively associated with an ABI of > 1.3, and also with changes in IMT. In contrast, they observed no significant association between air pollution and an ABI of < 0.9.
I have some concerns about their study (Rivera et al. 2013). First, the study included a small number of subjects with ABIs < 0.9 and > 1.3 (56 and 116, respectively). The authors used multi nomial logistic regression analysis; for the full-adjustment model, > 16 air pollution variables were used. There is a limitation in the number of independent variables appropriate for multiple logistic regression analysis (Novikov et al. 2010;Peduzzi et al. 1996), and enough events should be included to maintain statistical power for multi variate analysis. According to the criteria that at least 10 events per variable are required to keep stable estimates (Peduzzi et al. 1996), Rivera et al. (2013) needed ≥ 170 events with an ABI < 0.9 or > 1.3 for their analysis.
Second, Rivera et al. (2013) used systolic and diastolic blood pressure to adjust for the relationship between air pollution and indicators of atherosclerosis. But multi collinearity among independent variables should have been considered in the analysis (York 2012).
Finally, Rivera et al. (2013) could not clarify the lack of association between ABI < 0.9 and indicators of air pollution. An ABI of 0.9-1.0 is also associated with cardiovascular risk (Ono et al. 2003). Therefore, the associa tion between air pollution and subclinical athero sclerosis should be evaluated by selecting a higher cut-off value of ABI, such as 1.0. This procedure will increase the number of events for multi variate analysis.
Other researchers have reported a significant association between air pollution and IMT (Bauer et al. 2010;Diez Roux et al. 2008). Lenters et al. (2010) also examined the association between air pollutants and indicators of vascular damage but observed no association between air pollution and IMT. Lenters et al. (2010) used nitrogen dioxide (NO 2 ), black smoke, particulate matter ≤ 2.5 µm in aero dynamic diameter (PM 2.5 ), and sulfur dioxide (SO 2 ) as indicators of air pollution, and they used pulse wave velocity and augmentation index in addition to IMT as indicators of vascular damage. Traffic intensity and proximity of residence to roads were also used as indicators of air pollution. Lenters et al. found signifi cant associations only between NO 2 and pulse wave velocity and augmentation index and between SO 2 and pulse wave velocity. Because contradictory results for this association have been reported, further longitudinal studies are needed to assess this association.
The author declares he has no actual or potential competing financial interests. We appreciate Kawada's interest in our study (Rivera et al. 2013). His letter focuses on some limitations of our study, which were already discussed in our article.
Kawada notes that the small number of partici pants with low ankle-brachial index (ABI) provided low power to detect an associa tion; this is a limitation that we acknowledged in our article (Rivera et al. 2013). Kawada also points out that the number of covariates in the model exceeds the common rule of thumb of having at least 10 events per variable. We recognize that, with a small number of cases, one is inevitably faced with the trade-off between including all rele vant confounders and keeping the number of covariates to a minimum. However, in our article we provided results from a minimally adjusted model including only five confounders and an interaction term. In this model, the number of events per variable was > 10, and the results were not significantly different from those of the fully adjusted model. Kawada suggests selecting a "higher cut-off value of ABI, such as 1.0" for low ABI given that "an ABI of 0.9-1.0 is also associated with cardio vascular risk (Ono et al. 2003)." The findings of Ono et al. were for patients on hemo dialysis due to end-stage renal disease and thus cannot be extrapolated to healthy population samples such as the one considered in our study (i.e., with no history or current signs of cardiovascular disease). We selected a cut-off value of 0.9 for low ABI because of the strong evidence of increased risks of incident cardiovascular disease, morbidity, and mortality in individuals with ABI < 0.9 (Allison et al. 2008;Ankle Brachial Index Collaboration 2008;Lee et al. 2004;McDermott et al. 2005). A cut-off value of 0.9 is also more specific (Lee et al. 2004) and much more common in the literature, which allows compari son with other studies.
Kawada's second argument involves the consideration of multi collinearity in the fully adjusted model (model 2), in which systolic and diastolic blood pressure were included. Systolic and diastolic blood pressure were only moderately correlated (correlation coefficient, 0.62), and according to the variance inflation factor (VIF), there were no multicollinearity problems (VIF was 2.45 for systolic blood pressure and 1.98 for diastolic blood pressure).
Kawada points at "contradictory results" in the association between air pollution and carotid intima media thickness (IMT), mainly based on the null and weak associations found by Lenters et al. (2010) between several markers of air pollution [nitrogen dioxide, sulfur dioxide, PM 2.5 (particulate matter ≤ 2.5 µm in aero dynamic diameter), black smoke, and traffic intensity] and three indicators of vascular damage (IMT, pulse wave velocity, and augmentation index). We consider, however, that these do not constitute results contradictory to the positive Environmental Health Perspectives • volume 122 | number 1 | January 2014

A 9
Correspondence association between air pollution and subclinical markers of athero sclerosis found in the six studies thoroughly discussed in our article (Bauer et al. 2010;Diez Roux et al. 2008;Hoffmann et al. 2007Hoffmann et al. , 2009Künzli et al. 2005Künzli et al. , 2010 as well as by Wilker et al. (2013). The study by Lenters et al. (2010) involved a cohort of young adults, on average 28 years of age. Exposure to air pollution was estimated at the current address only. As the authors acknowledged, the young age of partici pants and the exposure misclassification, which resulted from exposure estimated at the current address only, are likely explanations for their mixed results.
Finally, we agree with Kawada's closing remark on the need for longitudinal studies, as we concluded in our article.