The economic burden of asthma is considerable, and therefore, it is valuable to identify exposures that can be modified on a population-level basis to reduce the health care costs to treat it. In 1994, the economic costs of asthma for the US were estimated to be in excess of $10 billion dollars . The corresponding estimate for Canada, in 1990, was estimated to fall in the range between $504 and $648 million . A review of published studies found that hospital costs typically account for 20–25% of the overall direct costs of asthma . Given that there are approximately 150,000 ED visits for asthma in Canada annually, a modest reduction in these numbers alone would provide considerable costs saving. We found positive associations between outdoor levels of air pollution and asthma ED visits, between April and September, in each age-group examined. Associations were generally stronger for NO2 and CO, however, they were also evident for O3 and particulate matter. These findings provide compelling evidence that reductions in emissions from the sources that give rise to these pollutant levels may decrease the associated direct health care costs of asthma in this region. Elsewhere, interventions to reduce outdoor air pollution levels have proven to be successful as they were accompanied with a concomitant decrease in the number of hospital visits and admissions for asthma, particularly in young children [30, 31].
This study was undertaken, in part, because there have been few Canadian studies that have evaluated how associations between ambient levels of air pollution and hospitals visits for asthma vary by age. In addition, the composition mix of pollution in Edmonton differs from other Canadian cities due to the close proximity of coal and petrochemical industries. The findings of our study are similar to those reported for a case-crossover study in Toronto where positive associations with CO and NO2 and hospital admissions for asthma in both males and females aged 6 to 12 were observed, however no effect was found for O3 . Further support for the relevance of vehicular traffic on asthma comes from the work by Oyana who reported an increased prevalence of asthma among children and adults who lived in close proximity to ambient sources of pollution at a US-Canada border crossing [32, 33]. In contrast, associations with ozone have been noted in the Canadian cities of Saint John, New Brunswick , but not with NO2 . Positive associations with ozone have been also noted in a study of ED visits in the province of Ontario and Toronto [35, 36]; for the province wide study, effects were more pronounced among children under the age of one, however, asthma remains an unclear diagnosis in children under the age of two.
Positive associations were observed with both particulate and gaseous phase pollution and as previously mentioned, they were most evident with NO2 and CO, both typically regarded as markers of vehicular traffic. In the province of Alberta, transportation accounts for a much smaller percentage of overall nitrogen oxides (NOX) emissions (26%), than it does in Canada as a whole (50%) . Therefore, it is possible that industrial sources of NO2 in the Edmonton area contribute to the increased risk of asthma visits. The nature of the hospital or pollution data does not allow us to evaluate the respective contributions of industrial versus transportation sources of pollution to the increased risk of asthma visits.
To better understand the interrelationship between CO, NO2 which were highly correlated with each other (r = 0.70), two-pollutant models were fit. These analyses revealed stronger associations with ED asthma visits among children and the elderly for NO2, relative to CO, in all age ranges except among those aged 15 to 45. There exist several biological mechanisms whereby NO2 can affect respiratory health. It has been shown to make people more susceptible to respiratory viral infections that exacerbate asthma , and enhance allergic responses after subsequent challenge . NO2 has also been shown to increase bronchitis symptoms among asthmatics , and reduced lung function among children who spend more time outside . More recently, research from the California Children's Health Study found that prolonged exposure to traffic pollution, including NO2, increases the incidence of childhood asthma [42, 43]. Taken together, there is growing support for the relevance of traffic related pollution in the exacerbation and development of asthma.
The ambient pollutant that has been most frequently associated with asthma hospitalizations has been ozone. In our study, an association between ozone and asthma ED visits was observed in patients of all ages; this association was strongest among those 5 – 14 years of age, while not statistically significant in other age ranges. Ozone (O3) is formed from photochemical reactions between NOX and volatile organic compounds in the presence of sunlight. Ozone levels are highest on warm sunny and calm days, with exposures peaking in mid-afternoon. Controlled laboratory studies have shown that O3 can invoke acute lower inflammatory responses in both healthy and asthmatic subjects, however, asthmatics appear to experience more severe responses . Given that the oxidant capacity of NO2 is smaller than that for O3 , our finding of a stronger association with NO2 and asthma visits is somewhat surprising. However, similar patterns have been observed in several recent asthma studies that have evaluated both pollutants [14, 46]. Recently, an eight city panel study of 990 children found that NO2 and CO were strongly related to asthma exacerbations while no such association was noted for O3 . Differences in meteorology, the complex mixture of pollution between regions, and the possibility of a threshold effect for ozone  may contribute to equivocal findings reported in different regions.
Emergency department studies of asthma have also evaluated the role of ambient particulate matter. Particulate air pollution is a mixture of solid particles and liquid droplets that can differ considerably in origin, size, and composition. Particulate matter includes aerosols, smoke, fumes, dust, ash and pollen. Fine particulate matter, which comprises those particles with an average aerodynamic diameter of 2.5 microns or less, has been studied more of late because it can better penetrate the respiratory system than particles of larger size. Positive associations between particulate matter and hospital visits for asthma have been reported in many international studies [20, 46, 49–56], but not all [9, 19, 34, 47, 57, 58]. Future short-term health effect studies of ambient pollution need to better isolate the biologically important constituents, and physical properties of particles that invoke responses in persons with asthma.
The validity of our findings relies on the accuracy of diagnosing asthma within the ED, and this accuracy is known to vary by the patient's age. As mentioned before, we excluded asthma ED visits among children less than two years of age as it is often confused with bronchiolitis . In older patients, while clinicians in theory are able to distinguish between asthma and chronic obstructive pulmonary disease (COPD), some diagnostic misclassification does occur [59, 60]. To evaluate the extent that such misclassification affects our presented risk estimates, we evaluated the association between air pollution and COPD visits in the elderly. We found that outdoor levels of air pollution were unrelated to ED visits for COPD in our patient population. This indicates that two different disease entities are being captured through the diagnostic patterns in place in the Edmonton area hospitals. However, the misdiagnosis of COPD as asthma would serve to underestimate the strength of our associations.
We found that the associations were strongest between outdoor air pollution and ED visits for asthma among children between 2 – 4 years of age, and among the elderly (= 75). Children are widely regarded to be a susceptible population for air pollution health effects for several reasons. They have higher minute ventilation and higher levels of physical activity, spend more time outside than adults, and their peripheral airways more susceptible to inflammatory narrowing . In addition, they retain a disproportionately higher amount of air pollution per unit body weight than adults . Factors that may increase the susceptibility of the elderly to air pollution include: higher airways deposition rate of particulate matter, deficits of dietary factors such as antioxidants, and compromised immune systems due to comorbidities and increased medication use . While further work is needed to evaluate how air pollution differentially affects the exacerbation of acute asthma by age, diverse findings from previously conducted Canadian studies highlights the continued need to look at both gaseous and particulate phase component of the air pollution mix.
The risk estimates presented here have been adjusted for meteorological effects of temperature, and relative humidity. They have also been adjusted for daily ED counts for influenza in order to control for viral respiratory seasonal epidemics. The case-crossover study design is also effective in controlling for the influence of individual-level risk factors that are unlikely to vary over short time intervals. For asthma, such factors are numerous and include: age, sex, cigarette smoking, household pets, and genetic predisposition to asthma. While cigarette smoke has been identified as an important risk factor for asthma, it is unlikely that it would confound our results as these exposures, as suggested by recent analyses of Canadian national survey data, are not related to with outdoor air pollution levels from fixed sited monitoring station . Similarly, indoor sources of NO2 from cooking and heating are unlikely to be correlated with outdoor sources over the short time interval of the study. Perhaps most importantly, the time-stratified case-crossover approach has also been demonstrated as a suitable method to control for time trends in both air pollution exposures and outcomes .
The case-crossover approach relies on the assumption that the event of interest, here ED visits for asthma, define the case intervals while no such visit can occur during the matched control intervals. This assumption can be violated under the scenario of recurrent events. For example, individuals may present themselves to the ED for asthma multiple times, and therefore, the control periods associated with some individuals could be misclassified. With the time-stratified design, this would occur if an individual had an ED visit for asthma on the same day of the week more than once in a given month. Unlike many other hospital-based case-crossover studies of recurrent outcomes, patient identification data were available for most visits; this allowed us to evaluate the extent of this possible bias. In our dataset for which patient identification data were available, approximately 33% of these patients visited the ED more than once over the study period. However, there were very few instances (n = 411) where an individual who visited the ED for asthma had a subsequent visit for asthma on the same day of the week, within the same month in a given year. Neither the exclusion of these matched sets, nor the re-coding of the control intervals to properly reflect the fact that these were case intervals changed the risk estimates in any appreciable way.
Our risk estimates are reliant on the use of air pollution levels derived from fixed-site monitoring stations. Measurement error from fixed site monitoring stations can occur from the devices themselves, or from an inability to account for heterogeneous pollution levels that exist spatially within the region. The magnitude of these measurement errors vary between pollutants. For example, pollutants such as NO2 exhibit tremendous spatial variability and have been shown to be correlated to traffic measures . In contrast, meteorological conditions strongly influence the efficiency of photochemical processes that lead to ozone formation [65, 66], and for this reason ground-level ozone air pollution is largely characterized on a regional-scale basis, rather than on an intra urban scale. While individual-level exposure estimates are generally recognized to be superior for evaluating risk of environmental exposures, as pointed out by Schwartz the use of a daily mean exposure for an entire city is relevant . Namely, the mean of personal exposures among residents in that city is likely more highly correlated with central monitoring station than individual exposures. Recent work into the measurement error associated with the use of fixed site monitoring stations suggest that most of the difference between personal and fixed site monitoring measurements of exposure are Berkson error, and therefore, do not bias the risk estimates. The work by Zeger suggests that the remaining measurement error follows the classical error model, and therefore, the overall net effect would be risk estimates that are biased towards the null . As a result, the measurement error associated with the use of fixed site monitoring stations is not the source of the positive associations found in our study population.
Along the same lines, aeroallergen levels likely varied within the Edmonton census area. One sampling device was used to infer daily aeroallergen levels. We expect that the mixing of spores in air, and transport by wind provides a more uniform mixture of aeroallergens throughout the study region. Because of this, and the fact that people are not stationary, we feel that the use of one sampling device is a valid means to represent daily aeroallergen levels in the study region. Further support for this comes from a sampling study that found high correlations in pollen counts between two sampling sites located 5.6 km apart .
Exacerbations of asthma are often caused by viral illnesses, exposure to irritants or allergens. In this study, we partitioned ED visits into two seasons, April to September and October to March. Daily monitoring of aeroallergen indicate that their relevance pertains strictly to the period between April and September. In contrast, as evidenced by seasonal patterns in the number of daily visits for influenza, a viral etiology predominates the winter period. Like others, we modeled the daily number of visits for influenza to control for seasonal viral respiratory epidemics [20, 70, 71]. While the frequency of daily influenza visits were correlated with the number of asthma visits during the winter, the addition of this term produced no appreciable change in the air pollution risk estimates. Similarly, aeroallergen levels did not confound the air pollution risk estimates between April and September.
Air pollution levels were generally higher in the period between October to March, than April to September where associations with asthma ED visits were evident. Differences in the air pollution risk estimates between the two seasons themselves may possibly be explained in part by differential exposure misclassification. As Edmonton-area residents spend a greater proportion of their time outside during the spring and summer seasons, fixed-site monitoring data likely more accurately reflects the average exposure to ambient pollution in the summer. Therefore, if the association between air pollution and asthma is real, there would be greater attenuation in risk estimates for winter time exposures.
Finally, it is important to note that this study undertook a large number of comparisons as we explored associations between multiple ambient measures of air pollution pollutants using several types of metrics, across different age groups and seasons. Due to the large number of statistical tests performed, the chances of detecting a spurious finding are increased. We did not change our p-values to take into account the multiple testing performed in this study as such an approach has criticized for introducing more problems than they are intended to solve . Despite the large number of tests performed, it is also important to recognize that our findings of stronger association in the young and elderly are consistent with the hypotheses we had a priori .[30, 31].