Living in close proximity to traffic was associated with increased prevalence of asthma and asthma symptoms last 12 months. No statistically significant effects were seen from traffic exposure at workplace address, daily time spent in traffic, or commuting time to work, after adjustment for potential confounders. A combined exposure estimate did not give higher association with asthma.
Discussion of exposure assessment
This is to our knowledge the first epidemiological study on asthma to use GIS not only to estimate traffic at residential address but also at workplace address and with information about commuting time to work or other outdoor time in traffic. However, while this more complete exposure information could be expected to strengthen any association with asthma, this was not found in this study.
A potential reason that no significant adverse effect was seen on workplace address could be if misclassification of exposure, due to invalid geocoding, was larger for workplace address. Since geocoding for the workplace address was made for the exact address, the geocoding technique in itself is not likely to be the reason for no association. However, if the study subjects are not stationary at their work location, or the company address might refer to larger commercial areas or buildings there might be little association between the personal exposure and the outdoor-indoor levels for that location. Exposure estimates at the residential addresses might on the other hand have inaccuracies due to imprecise geocoding since individuals are positioned at the centre of their real estates. In urban areas there might therefore be substantial misplacement for individuals living in large family-housing, or for large estates with vast land areas in the rural areas. It is well known that geocoding error generally gives conservative estimates , as does exposure misclassification in general if not related to disease.
It should be noted that effects of traffic on asthma symptoms were indicated at workplace addresses, but the effect estimates were lower than at residential address, and not statistically significant.
Since the associations between traffic-related air pollution and asthma generally shows distance-dependent relationship with strongest effects on asthma from living within 50 m of roads, and with sharp decline of many air pollutants within 30-150 m, a modelled resolution on NOx of 250 × 250 m might be too low to detect any effects from traffic. This must be weighted against the fact that a higher spatial resolution may not be meaningful considering the likely location uncertainty of workplace address.
An effect from daily time spent in traffic on asthma symptoms was indicated in unadjusted estimates, but not after adjustment for confounders. Exposure studies and simulations studies have shown that personal NOx dose per se is only marginally influenced by commuting time , but if NOx is seen as a proxy for NO and ultrafine particles, or other pipe-exhausts, the contribution from time in traffic outdoor at street-level i.e in congested traffic, may be many times higher and very influential of total exposure. In this study we regarded NOx as a proxy for traffic pollution and treated use of gas stove as a potential confounder rather than exposure. When calculating the contribution of "time in traffic" to total exposure, we let the "dosecontribution" vary between 30 μg/m3 and a more extreme scenario of 300 μg/m3, but this did not give a stronger association with asthma, although some of the asthma cases were moved from the lowest to a higher exposure category.
The major source of exposure misclassification may be the cross-sectional study character, especially for asthma prevalence, which showed an increased association with traffic when analysis was restricted to subjects which had been living at least 5 years at current address. Although asthma may start in adult age, most asthma begin in childhood , hence, a cross-sectional study in adults may poorly reflect retrospective exposure. This however should less affect the results for asthma symptoms last 12 months, a condition which is better related to current exposure, but may have different etiology and be affected differently by air pollution .
Since air pollution is well known to trigger symptoms [1, 23], (even if it is less certain if it contributes to the development of asthma), asthmatics may be more likely to move away from than towards traffic. Therefore a migrational bias is most likely to decrease the effects on asthma prevalence and asthma symptoms. It is also likely that the large proportion (44%) who regularly used asthma medication further would diminish the association between traffic and asthma symptoms, especially since people living closing to roads had a higher prevalence of asthma medication. In conclusion, cross-sectional studies need to be confirmed by prospective studies, not only to establish the casual link, but also to measure the true burden of disease from traffic.
Since this study was conducted in an area with low levels of air pollution in a European perspective, high exposure to traffic was rare and the study was slightly underpowered to estimate effects from residential traffic at traffic levels which has previously shown to be related to effects. This also hindered any further analysis of effect modifications by other risk factors than traffic. Pooling of exposure groups would not help since only the highest exposure groups showed a relation to traffic, thus pooling would severely dilute the effects.
Discussion of potential confounding and selection bias
A strength of the study was the large number of potential confounder information which was collected, such as BMI , occupational exposure , and presence of indoor dampness and mould , which are known risk factors for adult asthma and often associated with socio-economic status of the neighbourhood. Socio-economic status (SEI), with the classification system used in this study, has in Sweden shown an association with asthma incidence in recent years . Confounder adjustment slightly increased the effect estimates for residential address, suggesting that competing risk factors sometimes dilute the effects from traffic, something we have previously suggested . A weakness was that we did not have more detailed data on triggers for asthmatic symptoms, since we previously have observed a association between traffic and asthma triggered by pollen and furred animals, but not with asthma triggered by other factors . Degree of confounding (measured or unmeasured) is not likely to be directly generalizable between studies since the association between covariates such as socio-economic status and air pollution (NOx) has been shown to be reversed depending on area in Scania . Confounding is better controlled for with respect to asthma symptoms than to asthma prevalence in this study, since we had information about current but not past exposure to risk factors.
The effect estimates for residential traffic were stronger in the case-control study than in the first survey, indicating potential selection bias. In previous public health surveys in the region it has been shown that the response rate is dependent on geographical strata . It is thus not unlikely that selection bias can have occurred, however the objective exposure assessments used in this study is a true advantage. Ideally, since this study was sampled on geographical strata, an analysis conditional on geographical stratum might have increased the validity. This was however not possible since exposure ranges were not comparable between the different stratas/communities (figure 2). This also excluded the possibility to use a dummy variable for urban/rural areas to adjust for potential residual urban-rural confounding. It should be noted that accounting for total traffic exposure could further have strengthened any residual urban-rural confounding by comparing people who are both working and living in rural environments, with people who are both working and living in urban environments.
To our knowledge, all previous studies on adult asthma prevalence have only estimated traffic exposure at residential address. A previous cross-sectional study in southern Sweden found asthma triggered by allergic factors to be associated with high traffic intensity within 100 m of residence, and with modelled NO
> 19 μg/m3 [17, 28]. A cross-sectional study in northern Sweden found that asthmatic symptoms increased significantly with modelled NO2-concentrations and self-reported heavy vehicles outside the kitchen window . A Swedish case-control study found measured home outdoor NO2 (min-max: 0-29 μg/m3) to be associated with asthma incidence among atopics . The Swedish cities in the Nordic Rhine study found modelled NO2 to be associated with incident asthma (OR = 1.5, 95% CI 1.0-2.4, per 10 μg/m3) (min-max: 3.3-46 μg/m3) .
A few European cohort studies have supported that traffic pollution increases asthma incidence in adults: The ECRHS study found an association between modelled NO2 and increased asthma incidence (OR 1.4; 95% CI 1.0-2.0, per 10 μg/m3) , The SAPALDIA study found that asthma incidence was associated with modelled change in TPM10, hazard ratio (1.3, 95%CI: 1.1 - 1.6 per 1 μg/m3 change) 
The results from other Swedish studies support that asthma symptoms are affected at relatively low levels of air pollution. Cohort studies in adults, although still few, also supports that the association between traffic exposure and asthma prevalence observed in this cross-sectional study may reflect a true increase in asthma incidence when living close to traffic.
However, if the most recent studies support the association between air pollution and asthma, the relation with asthma incidence is not fully settled and there are also a few recent negative studies in adults [32, 33], and some cohorts in children .
There are two studies in children which have investigated the effects of traffic at both home and school, on asthma. McConnell et al found an increased hazard ratio when combining traffic-related pollutants at school-and residential address, on new-onset asthma, compared to the independent effects . The other study by Kim et al make a reservation that the study was not designed for independent assessment of exposure at school- and residential address, and the sample size was insufficient to properly do so, but they report that they found a slight attenuation of effects on current asthma from residential traffic pollution when adding both residential and school exposure in the same model .
In our study, effects at workplace address in the highest exposure categories were statistically insignificant partly because lack of power to confirm small effect estimates. Further studies in areas where high levels of air pollution is rare, should consider to strongly oversample exposed subjects in relevant exposure ranges and population groups.
However, the lack of power can not explain that the association did not get stronger for total exposure. Alhough our lack of statistically significant association with traffic at workplace address and time spent in traffic may be due to misclassification of exposure, it may also indicate that residence is still the most influential exposure determinant of traffic exposure among adults.