The primary objective of our study was to investigate associations between ambient concentrations of NO2
and the risk of stroke using a spatially refined measure of air pollution that was available from a land use regression model. This study did not evaluate effects of fine particulate matter for which both short term and long term exposures have been associated with the risk of stroke [2
]. Concentrations of NO2
, to a far greater extent than fine particulate matter, vary considerably within intra-urban areas, and are regarded as a surrogate measure of traffic related air pollution [38
]. Our previous analyses of the associations between day to day changes in air pollution and stroke revealed associations that were dominated by NO2
during the summer months [20
In this case-control study, we found no association between an IQR increase in LUR-derived measures of NO2
and stroke. LUR models are based on land use and traffic patterns within urban centers and can provide high resolution estimates of within-city ambient NO2
concentrations. In comparison to exposures estimated from interpolation models of air pollution where monitor locations are sparse, data from an LUR generally have less spatial error [39
]. In this respect, our present work uses a much more refined measure of exposure than we applied in our previous ecologic analysis of stroke risk and long-term exposure to ambient NO2
in Edmonton. In that ecological study we used an interpolation model of air pollution based on data from two fixed-sited monitoring stations [21
], and no association with stroke was found for either NO2
or fine particulate matter (PM2.5
) after adjusting for other contextual confounding variables. So, even with the improvements of individual-level data and higher spatial resolution, we were still unable to find a longer-term effect of pollution (medium- or long-term) on stroke events in Edmonton. In contrast, we found positive associations between day to day increases in NO2
and ischemic stroke risk in Edmonton during the summer months [7
]. The difference may be due to a greater pathogenetic influence of air pollution exposure on processes involved in triggering stroke over the short-term compared to the slow progression of atherosclerosis or venous thrombosis over a longer time frame [18
]. Combined, the findings from across our studies suggests that only day to day elevations in ambient air pollution, but not medium term exposures, increase the risk of stroke. While findings from the case-control study should be interpreted with some caution given the lack of data on individual level risk factors, we found very little difference in NO2
levels among current and never smokers in the CCHS.
Elsewhere, cohort studies published to date have generated an inconclusive body of evidence on the effects of long-term exposures on stroke. While an association with an IQR increase in NO2
was reported in a cohort analysis in Denmark (fully adjusted hazard ratio = 1.05; 95% CI: 0.99-1.11) [3
], no association was found with a 10 μg/m3
increase in particulate matter <2.5 μm in aerodynamic diameter (PM2.5
) among a 10-year cohort study in Canada [12
]. However, data from a cohort of women in the US found that stroke events were strongly associated with a 10 μg/m3
increase in PM2.5
(hazard ratio: 1.28, 95% CI: 1.02-1.61) [11
] and a positive association was also observed with a 10 μg/m3
increase in PM10
in the California Teachers cohort (1.06, 95% CI: 1.00,1.13) [19
]. Although we note that our exposure data were not historical and we were not modeling long-term effects on stroke risk, we did not find a similar effect of NO2
on stroke among women in our case-control population.
In contrast to cohort studies, case-control studies lack the ability to examine time-at-risk effects; however, they may offer improvements in cost and statistical efficiencies. Our study findings align closely those from a recent case-control study of acute strokes in Scania, Sweden where no association was found between stroke and 10 μg/m3
increases in NOx
(OR = 0.93; 95% CI: 0.82-1.95) [1
]. That study and ours differed in regards to the patient population. While their case series pooled ICD-10 codes I61x, I63x, and I64x [41
], our overall dataset of stroke patients also included those with discharge diagnoses I60x, I62x, and G54x. Our total stroke dataset represent a more heterogeneous set of clinical outcomes; however, these data permitted us to further consider the possibility that air pollution exerts different effects on clinically different stroke types.
In our analyses, we assumed medium-term residency at the postal code given in the hospital database, based on the observation that 18.1% of Edmonton residents reported moving residence within the previous 12 months, in the 2006 Census [42
]. Given the association between age and stroke and the higher tendency to move among younger Canadians compared to older residents [43
], even medium-term exposure misclassification could have been more common among controls than cases.
It appears that even with high spatial resolution exposure data, our results do not differ substantially from previous studies; there is no evidence of a medium-term effect of NO2
on risk for any subtype of stroke, or overall stroke. In the only study of short-term effects of NOx
on TIA, Henrotin et al
found no association between 10 μg/m3
increases in NOx
and TIA (OR = 0.86; 95% CI: 0.74, 1.02) [6
]. Similar to our present findings, we found no association between 5-year average concentration of NO2
and TIA in our previous ecologic analysis of stroke in Edmonton [21
]. Hemorrhagic stroke risk is unaffected by long-term ambient NO2
]; on this point our results are also consistent with those from other studies.
Our study is, to our knowledge, the first case-control study of air pollution effects on stroke to use hospital controls. While a population-based control group would be more representative of the source population, such controls are not readily available in Canada. Regardless, our use of hospital controls was methodologically sound. Because there are no plausible biological mechanisms to suggest that long term exposure to ambient pollution increases the risk of experiencing lacerations, odds ratios were not affected by this type of Berksonian bias [45
]. Traffic density is associated with traffic injury, but patients suffering from those events are coded as ICD-10 V01 – X59 [46
]; thus, our control group, patients presenting with lacerations, strains, sprains would not have been involved in events associated with exposure.
In case-control studies, selection bias may also arise when a confounder influences risk estimates and the distribution of the confounder between case and control groups arises due to control group selection [47
]. Age was strongly associated with case status, weakly and variably associated with exposure, and was controlled for in our models. We conducted 2 sensitivity analyses which suggest that these models sufficiently controlled potential confounding due to the difference between age distributions of case and control groups. Hospital-based control sampling ensured that the controls would have had access to the study EDs had they had stroke and allowed us to eliminate patients who were stroke and laceration patients during the study period. We feel any residual confounding may have less impact on our results than would any self-selection bias that is inherent in population-based control sampling with low response rates [48
]. Strokes are more common among women than among men and admissions for accidental wounds, strains, and sprains may be more common among men [49
]. We controlled for the confounding due to sex in our analysis, thereby reducing potential selection bias inherent in our control group.
Increased risk for ischemic stroke among smokers has been documented extensively and reviewed recently [51
]; however, with group-level data we did not find that NO2
was strongly different across smoking groups. Oudin et al.
observed stronger effects of NOx
on acute strokes among non-smokers than among smokers [52
]. It is possible, then, that smoking could have been an effect modifier of our results. Data on other known risk factors for stroke such as diabetes, hypertension, and previous stroke [53
] could have also allowed us to define sub-populations with increased susceptibility to the effects of ambient air pollution, but these data were not available among hospital administration databases. Data on anti-coagulant use would help us to determine those who have been controlling their risk of stroke, if they already have known risk factors for cardiovascular disease or stroke. Also, without historical medical data, we cannot rule out the additional possibility that individuals with existing risk factors for stroke, including chronic diseases, chose to move closer to high-traffic areas of Edmonton to shorten travel times to specialized health care. With regionalization of stroke and cardiac care units into higher level tertiary facilities [54
], generally, located close to universities, there may be unintended harmful effects on those populations with potentially greater vulnerability to air pollution effects. Our study, however, included patients from 11 hospitals of various levels of specialization (primary, secondary, and tertiary care) throughout the city, so, the potential for intensive health care needs among a higher stroke-risk group to cause selection bias is minimal.
When socioeconomic variables that captured small area effects of education and household income were entered into the models they produced only a small increase in the odds ratios for most outcomes. However, there is an inherent limitation due to the reliance on CT-level data for SES indicators. In metropolitan centers, CT-level data may poorly represent individual-level deprivation [55
]. However, while the association between health and SES is stronger at the individual level, the direction of the association is the same at the ecological level. The probability of misclassification could be argued to be greater among those in areas of higher traffic density, as the population density within the CT would be greater than in areas at the city boundary.