Evidence of adverse effects of current air pollution on human health substantially increased in recent years.1 2 Despite improvements in air quality in many regions of the world, primary and secondary pollutants from traffic and other sources of fossil fuel combustion such as particulate matter, diesel soot or ozone remain of particular concern. Photochemical air pollution represented by ozone, oxides of nitrogen produced by vehicles and respirable fine and ultrafine particulates are of interest owing to their toxic properties, and asthmatics are a particularly sensitive subgroup.1 Associations between daily changes in air pollution and various acute respiratory outcomes including subclinical functional changes, symptoms, doctors or emergency room visits, hospitalisations and death,3 have been reported among asthmatics.4 5 Evidence of acute effects is based on several panel studies conducted in children, and to a lesser extent in adults. Greater susceptibility to acute effects of ambient air pollution in asthmatics with more severe asthma was observed in children6 but was not consistent in adults.5 7 A challenge and potential source of inconsistent results is the complex inter-relation of asthma symptoms, its triggers and subjects’ coping strategies through adaptation of treatment. As the typically moderate daily changes in air quality may lead to fluctuations in the expression of the disease, it is of interest to evaluate markers of asthma severity on a clinically relevant scale along the continuum between non-symptomatic not regularly treated asthma and symptomatic asthma despite regular treatment.

The concept of the definition of asthma severity has evolved over the past years. Despite development of guidelines for clinical practice, it remains difficult to define asthma severity in epidemiology. While different definitions of asthma severity are in use,8 9 the concept is certainly clinically relevant. Determinants of asthma severity are not much investigated10 and, to our knowledge, no study investigated the effect of air pollution on asthma severity.

The aim of the present study was to investigate the relations between individually assigned exposure to ambient air pollution and asthma severity assessed in two complementary ways. First, we use a clinically relevant score that integrates both symptoms and treatment. Second, we use a simpler score based on symptoms alone. We hypothesise that severity among adult asthmatics from the Epidemiological study on the Genetics and Environment of Asthma, bronchial hyper-responsiveness and atopy (EGEA) correlates with the average air quality at the residential location.

METHODS

Study population and asthma severity

The EGEA study combines a case-control study and a family study.8 11 12 The protocol has been described elsewhere.13 Asthmatics, living in predefined geographical areas were recruited from chest clinics in five French cities (Paris, Lyon, Marseille, Montpellier and Grenoble) between 1991 and 1995. Within each city, subjects’ residences spread across large geographical areas, ranging from 300 km2 (for the area of Lyon) to 9300 km2 (for the area of Marseille). Cases were classified as living far from the university hospital based on a distance of at least 10–15 km. Asthmatic probands were those who answered positively to the four following standardised questions: Have you ever had attacks of breathlessness at rest with wheezing? Have you ever had asthma attacks? Was this diagnosis confirmed by a physician? Have you had an asthma attack in the last 12 months? Among relatives from the families of asthmatic probands, asthma was defined by a positive answer to at least one of the first two questions. The present analysis concerned 328 adult asthmatic subjects from 255 families (188 asthmatic cases and 140 asthmatic relatives). Information was recorded for respiratory symptoms, environment, smoking habits and treatment by a detailed face-to-face interview administered international standardised questionnaire.14

We used two operational definitions of “asthma severity”. The first approach consists in a four-class asthma severity score defined for the last 12 months that is based on the concept of 2002–2004-GINA (Global Initiative for Asthma) international guidelines.15 First, asthma severity was assessed using a clinical severity score derived from frequency of asthma attacks, symptoms between attacks and hospitalisation for asthma.11 12 16 17 Second, the four-class asthma severity score was developed,8 which combined the above clinical score, with information about the occurrence of any attack in the last 12 months, and treatment reported in the past 12 months (no treatment, treatment without inhaled corticosteroids and treatment with inhaled corticosteroids).

The second definition of severity corresponded to the five-level asthma score introduced by the European Community Respiratory Health Study.9 The score, allowing assessing the activity of asthma, was derived as the sum of affirmative responses to any of the following five asthma-like symptoms: breathless when wheezing, woken up with a feeling of chest tightness, woken by attack of shortness of breath (SOB), attack of SOB at rest, and attack of SOB after exercise. All questions referred to the occurrence of symptoms during the last 12 months.

The independence of asthma severity with the distance from the university hospital was checked, as it could have biased the association of air pollution with asthma severity.

Air pollution

Two approaches were used to assign air pollution exposure to each individual: The first assessment of air pollution relied on the 1991–5 data from the French Agency for Environment and Energy (ADEME) monitoring network. In the case of nitrogen dioxide (NO₂) and sulphur dioxide (SO₂), annual arithmetic means were used. In the case of ozone (O₃) we derived both the 8-hour average (O₃-8 h) across the six summer months (April to September) and the number of days with 8-hour averages above 110 μg.m⁻³ (O₃-days). No data were available for particulate matter (PM) for the period under study. The individual assignment of exposure is described in the online appendix. In brief, annual mean concentrations (O₃, NO₂, SO₂) from the closest monitor were assigned to each residence. For subjects living in proximity to busy roads we used NO₂ and SO₂ data from the closest traffic monitor station. In all other cases, only data from background monitors were used.

In the second approach we derived concentration estimates from a geo-statistical model carried out by Geovariances for the French Institute of Environment (IFEN) and ADEME for the year 1998.18 A validation of the procedure has been performed. Modelled data were compared with classic interpolation algorithms such as inverse distance or univariate kriging. High correlation coefficients (r>0.73) between interpolated and measured values, together with low mean quadratic errors, confirm the validity of the multivariate geostatistical models (Nicolas Jeannée, personal communication). The geostatistical interpolation techniques (kriging-like techniques)19 have been commonly applied for several years to estimate air pollution at unsampled locations.20 21 The estimation process, based on measured concentrations, takes into account the spatial structure of each pollutant; the better the spatial continuity of the pollutant, the better the confidence one can have in the estimate. The interpolation is realised for annual mean concentrations coming from background monitoring stations on a 4 km×4 km grid covering France. Land cover was integrated to the interpolation process as well as specific cofactors correlated with the pollutants: altitude and north-south concentration gradient for O₃, emission registries of nitrogen oxides, road network and population density for NO₂, and industrial emissions of SO_x for SO₂. Residential addresses of all participants were geo-coded to assign estimated annual means for NO₂, SO₂ as well as summer O₃ (O₃-summer) assessed from the monthly means from April to September.

Data analysis

Pearson correlation coefficients were derived across individually assigned air pollution concentrations. Univariate relations between variables were explored with χ2 tests and analyses of variance. Ordinal logistic regressions and, when appropriate, multinomial regressions (see online appendix) were performed. Odds ratios were given for one class difference in the scores and for an increase of one interquartile range (IQR) in air pollution concentrations. Final models took into account age, sex, smoking habits, educational level and occupational exposure. Smoking habits, previously associated to asthma severity,11 were grouped into never, ex-smokers and current smokers. Education was classified in primary, secondary and university level. Occupational exposure was assessed with a job exposure matrix for 18 asthmogens completed by expert review.8 22 Environmental tobacco smoke (current smoking of parents or spouse at home) and indoor air pollution, assessed by considering current use of gas heating or gas cooking and water damage in the last 12 months, were considered but not retained in the final model. In order to test the sensitivity of results, analyses were repeated in subgroups defined by sex, age, smoking habits, occupational exposure and education level as well as by city. Odds ratios were expressed for an increase of 10 days of O₃-days.

In France, O₃ seasons are by and large restricted to the summer period; thus summertime O₃ reflects exposure much better then annual concentrations. While severity was assessed for the last 12 months, one may argue that patients would primarily report the more recent experience. Therefore, if O₃ affects asthma severity in an acute or subacute manner, associations between severity and O₃ may be stronger among subjects answering the questionnaire during the mid summer to fall (autumn) season. We thus stratified the analyses by season of participation (April to September versus October to March). Familial dependence between observations was taken into account using generalised estimated equations (GENMOD procedure) and a random population with one subject per family were performed, but results did not change (data not shown). More details are presented in the online appendix. The statistical software package SAS was used.

RESULTS

Clinical characteristics of the population are reported in table 1 and environmental exposures in table 2. With the first approach (that is, “closest monitor”) we were able to assign annual mean concentrations to most but not all subjects: 88.4% for NO₂, 77.7% for SO₂, 64.0% for O₃-8 h and 88.1% for O₃-days. The distribution of assigned pollutant concentrations is presented in figure 1. Figure S1 in the online appendix presents data by city. The second exposure assignment approach allowed assigning concentrations to 93.9% of subjects.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 The bottom and top of the boxes represent the 25th and 75th percentiles, respectively. The upper and lower whiskers represent the maximum and minimum, respectively. The horizontal line through the middle of the boxes is the median. Median values were 36 μg.m⁻³ for nitrogen dioxide (NO₂) 19 μg.m⁻³ for sulphur dioxide (SO₂), 68 μg.m⁻³ for the 8-hour average of ozone (O₃-8 h) and 30 days for number of days with 8-hour ozone averages above 110 μg.m⁻³ (O₃-days).

Correlations between assigned pollutant concentrations across the two indicators of O₃ based on the first method were strong (r = 0.79, p⩽0.0001). The correlation between NO₂ and SO₂ was lower (r = 0.48, p⩽0.0001). O₃-days and NO₂ as well as O₃-days and SO₂ were not correlated (r = −0.05 and −0.04, respectively). Regarding the two different methods of air pollution assignment, the Pearson coefficients showed significant correlations for NO₂ concentrations (p = 0.48), for SO₂ concentrations (p = 0.33) and for O₃ (p = 0.66 between O₃-days and O₃-summer and p = 0.76 between O₃-8 h and O₃-summer).

All pollutants assigned from the closest monitors (first approach) were positively associated with the integrated four-class asthma severity score, but the associations reached statistical significance only for the two indicators of O₃. Results remained significant for both O₃-8 h and O₃-days also in models adjusting for age, sex, smoking habits, occupational exposure, and educational level (table 3). Taking into account the duration of residence did not change the results. Associations between O₃ and the four-class asthma severity score remained significant among the 94.8% of the subjects living for at least one year at their current address (OR = 2.64 (1.74 to 4.00) for O₃-days and 1.65 (1.27 to 2.14) for O₃-summer). Results were unchanged when adjusting for the number of years in the current residence.

Effect estimates of O₃ were similar in three-pollutant models including O₃, SO₂ and NO₂ (odds ratio (OR) for O₃-days 2.74 (95% confidence intervals (CI) 1.68 to 4.48). Relations with ambient NO₂ were not affected by gas cooking and no association was observed between asthma severity score and cooking or heating with gas. Exposure values assigned from the IFEN model correlated with the four-class asthma severity score in a similar way, and the positive association observed with annual SO₂ became statistically significant.

Results for the simpler five-level asthma score were very similar for all pollutants and both exposure assessment approaches (table S1 in the online appendix). In the three-pollutant model with pollutants from the closest monitor, the five-level asthma score was significantly associated only with O₃ days (OR = 2.02 (1.26 to 3.23) per one IQR of O₃-days). With the IFEN model values, O₃-summer correlated significantly with severity (2.20 (1.54 to 3.15)).

Table 4 presents relations between the four-class asthma severity and O3 (O₃-days) among subgroups. Associations remained robust among strata defined by sex, age, smoking habits, occupational exposure and education level. None of the interaction terms was significant. Stratification on periods of examination showed significant associations of O3 exposure to asthma severity score among subjects participating both from October to March as well as from April to September (OR = 1.24 (1.08 to 1.43) and 1.48 (1.19 to 1.85), respectively). Consistent results regarding stratification by the previous factors were observed for the five-level asthma score (data not shown). For the five-level asthma score, stratification was further studied according to treatment. There was no interaction with treatment (odds ratios were 1.04 (0.73 to 1.49) and 1.11 (0.99 to 1.25) for an increase of 10 days of O₃-days, for untreated and treated, respectively; interaction p value  = 0.35). Effects of O₃ across centres were statistically not heterogeneous (p for interaction 0.07). The size of the geographical areas where subjects lived was unequal, with subjects from Marseille residing across a region some 30 times the size of the Grenoble area as a result of the hospital-based sampling. Accordingly, the range of O₃ concentrations across subjects also varied across areas—for example, among participants in Paris the IQR of the O₃-days was only 5–26 but rather wide in Montpellier (IQR 3–90) (fig S1 in the online appendix). Excluding the centre with the highest value of O₃ (Marseille) or the centre with significant effect (Lyon) did not affect the effect estimates for O₃-days (OR = 1.28 (1.10 to 1.48) and 1.23 (1.10 to 1.38), respectively). Defining “city” as a random effect did not alter the results (OR = 1.24 (1.08 to 1.42)).

DISCUSSION

In the well characterised sample of more than 300 adult asthmatics from the EGEA study, higher residential concentrations of O₃ were associated with more severe asthma. The findings were very similar for the integrated score of clinical symptoms and treatment over the past 12 months and the novel five-level asthma symptom score. Moreover, results were not sensitive to the chosen exposure assessment approach nor to adjustment for age, sex, smoking habits, occupational exposure to asthmogens and educational level. Results were also similar for asthmatics with or without treatment, and for smokers or non-smokers. To the best of our knowledge, this is the first study using asthma severity scores to investigate effects of air pollution. As one score is anchored on symptoms and treatment, it better integrates the course of this disease where deteriorations of the state are usually coupled with adaptations in treatments among well managed patients. The consistent associations observed in this study consolidate the findings. While several panel studies and time series studies suggest the course of asthma to be correlated with ambient oxidant concentrations in adults, the severity score is a promising integrated outcome.

The interpretation of the temporal nature of the observed association between O₃ and asthma severity is challenging and must be put in context of the study design. Depending on the assessment of outcomes and exposures, cross-sectional analyses may investigate not only long-term effects but also subacute or acute effects. Our questionnaires asked about the occurrence of symptoms and treatment in the last 12 months, thus results may be interpreted as a summary of all acute effects experienced during the past year. In this case, results are comparable to similar findings in panel studies showing asthma symptoms among adults to increase with atmospheric pollutants, both in North America and in Europe,23 although some heterogeneity was observed in European cities. Deleterious effects of O₃ in moderate/severe adult asthmatics studied over one year have been reported from the Paris area.5 The asthma severity score used in the present analysis includes frequency of attacks as well as symptoms between attacks and adaptive treatment. A complementary interpretation of the observed cross-sectional associations could be that asthma severity is increased as a consequence of chronic processes because of repeated long-term exposure. The fact that subjects who participated during winter showed similar associations with summertime O₃ concentrations as those participating during the summer period may be an indication of subacute or longer term effects of O₃ on asthma severity, as the O₃ seasons in France are by and large restricted to the summer period. However, a conclusive distinction of acute, subacute or long-term effects of ambient O₃ exposure cannot be made with this cross-sectional approach nor in panel studies but would require large cohort studies. It would be particularly relevant to have repeated measurements at various seasons in large cohorts to disentangle subacute and chronic effects.

The association between acute, subacute and carryover chronic pathologies as a result of O₃ exposure is also not fully elucidated in experimental studies. Nevertheless, plausible biological mechanisms could explain the association of O₃ with asthma severity. The powerful oxidant capacity of inhaled O₃ is well known and could play a part for both acute and chronic effects by maintaining airway inflammation. O₃ has also been shown to favour a Th2 pattern24 and increase eosinophils,25 but neither IgE nor eosinophils in our population were related to asthma severity.26 Endogenous and exogenous determinants of susceptibility to the adverse effects of O₃ are most probably important, and a source of random noise if not controlled in the analyses. Interactions of O₃ with genes involved in the regulation of oxidative stress, such as glutathione S-transferases have already been observed in children,27 and may have an important role in adults as well. Beyond acute effects, the pulmonary inflammatory response to inhaled O₃ and other oxidants, such as cigarette smoke, is a mechanism that could explain sustained effects of these environmental factors. Acute and chronic effects of air pollution may be inter-related, especially among asthmatics, as the repeated acute inflammatory effects may contribute to the chronic processes of airway remodelling and to severity of asthma.28 The importance of the various inflammatory patterns in severe and persistent asthma has recently been underlined10 and further studies are needed to understand which pathways may be implicated, by combining environmental exposure, inflammatory markers and relevant genetic polymorphisms. Further studies should be conducted to improve exposure assessment in order to disentangle acute and chronic effects of air pollution. Information on the activity of the asthmatic disease over short periods, such as the last 15 days or 3 months should be collected as well. Indeed, the last recommendations of GINA, which focus on asthma control, will favour such data collection.29

Strengths and limitations

A strong asset of the EGEA study is the detailed phenotypic characterisation of the sample recruited through chest clinics, allowing studying a wide range of asthma activity and severity. It can never be totally excluded that some subjects, especially among severe asthmatics do have a mixed form of chronic obstructive pulmonary disease (COPD) and asthma. However, in the context of the EGEA study, in which asthmatics were specifically recruited in specialised chest clinics, it is rather unlikely that COPD patients were misdiagnosed as having asthma. Population-based studies usually lack sufficient numbers of severe asthmatics. Evidence is increasing regarding the heterogeneity of asthma and severe asthma.10 Asthma severity combining clinical items and treatment was assessed in close accordance to the 2002–4 GINA guidelines principles,15 but using the type of treatment rather than the daily doses of treatment. Our asthma severity score has previously been shown to be associated with other risk factors in the EGEA study as occupational exposure8 and body mass index in women12 and linked to the 2p23 genetic region.30

It is interesting that we see no difference between results based on the integrated severity score and the simpler symptoms score, reflecting more the activity of the disease. The continuous asthma score based on symptoms has recently been recommended following observations from the European Community Respiratory Health survey as a relevant phenotype to assess associations with environmental determinants.31 While this score is simpler to assess as it requires no treatment information, one may encounter difficulties in assessing the effect of air pollution on severity because of the interaction between exposure, symptom occurrence and changes in treatment. In our sample of asthmatics, effects of O₃ were similar for those with or without treatment. More research is needed to understand the strength and limitations of the two scores and their use in aetiological research.

The association of asthma severity with air pollution could have been biased by recruitment characteristics. All hospitals were homogeneous regarding their recruitment. They were all secondary hospitals with a relatively high proportion of severe asthmatics. If greater distance from the main city was related to greater severity, it could have biased the association of asthma severity with O₃. Indeed, the severity was independent of the distance from the hospital and therefore such bias is unlikely.

The estimation of long-term exposure is a challenge. Estimates of concentrations differed between the two approaches owing to the inherently different methods and data used. Both approaches have their strengths and limitations and it is impossible to evaluate which approach provides estimates closer to the true ambient concentrations. Having two approaches available is a strength of our study and the similarity in the O₃ and NO₂ results across methods support our findings. As mentioned in the Methods section, the IFEN model was available for 1998. Thus, for participants in the early 1990s, the assigned exposure may be less accurate than for participants during the later years. In line with this assumption, ORs were slightly smaller among early (1991–3) participants than late (1994–5) participants, but remained significant in both groups (data not shown). Most importantly though, O₃ effects remained unchanged even in three-pollutant models including adjustment for NO₂ as a marker of traffic-related emissions.

The null findings for NO₂ must be put in context of the exposure assignment methods. NO₂ concentrations are affected by local traffic sources, and concentrations may substantially differ within the first 100 metres of busy roads. The two approaches used in the present study were not designed to capture these local contrasts. Our “simpler approach”—that is, the use of the closest monitor, did take into account the closest traffic monitor of NO₂ for people living close to streets, while the IFEN model, based on background monitors only, and a 4×4 km grid, is less sensitive to proximity to traffic. However, the 4×4 km grid model was refined with network traffic and emission registries of NO_x. Thus, to investigate effects of traffic emissions on asthma severity, other exposure assignments may be needed such as personal measurements taken with passive samplers, traffic or land use data within narrow buffers around the residence, or geospatial models using higher spatial resolution. It is worthwhile conducting projects targeted at traffic-related exposures given the large number of people in Europe living very close to often extremely busy roads.32 Moreover, we only assigned exposure to residential address as no information on workplace or time activity patterns were available, as in most studies conducted in adults.33 In the first approach using the closest monitor, the maximum distance chosen between residential address and monitor was 40 km and this distance varies among studies, with some studies going as far as 80 km,34 but most of our subjects (93%) lived within 10 km of a monitor. While this rather short distance minimises the error in assigned exposure for homogeneously distributed pollutants such as O₃ or SO₂, errors are expected to be larger for spatially heterogeneous pollutants such as NO₂. We expect these errors to be non-systematic, thus biasing associations towards the null.

The contrast in the assigned exposure was rather low in the case of NO₂ and SO₂, with, respectively, a 1.5-fold and 1.8-fold change across the interquartile range (IQR) of concentrations (31–45 and 15–27, respectively; closest monitor approach), thus affecting the statistical power to observe effects. In the case of O₃, power was clearly better with a 2.8-fold difference (IQR 20–56). The IFEN model resulted in similar contrasts for NO₂, whereas those for SO₂ were stronger (2.5-fold increase). This may explain why associations became significant in this model. IQR for O₃ was lower in the IFEN model (1.2-fold increase), nevertheless results remained statistically significant and support the interpretation that oxidant pollutants indicated by ambient O₃ affect asthma severity.

We found some evidence for heterogeneity of effects across centres with far smaller estimates in some cities. In both analyses performed, adding centre as a random effect or after exclusion of the centres, possibly driving the association in relation to the highest O₃ exposure, did not change the main estimate. However, owing to the limited sample size within cities, it is not possible to further elucidate whether effects of O₃ to be centre-specific, driven, for example, by other environmental or population factors present in only a few locations, or whether heterogeneity was a random finding. It is of note though that the size of the geographical areas was unequal across these cities and that the within-city contrasts of pollutants also differed across cities. These factors may all contribute to heterogeneity.

The question is whether the observed associations reflect an effect of O₃ or whether the latter is a proxy of some other pollutants. Particle measures were not available nor did we have personal measurements of O₃ exposure. There is some argument that personal O₃ exposure may be more strongly correlated with ambient particles than with ambient O₃ concentrations.35 This finding, while not universally true, may be because of the well-known heterogeneity between indoor and outdoor concentrations of O₃, with usually very low concentrations indoors where people spend 70–80% of their time.36 Our estimates for O₃ were not sensitive to the inclusion of ambient NO₂ in the three-pollutant model. Based on other studies one may expect substantial correlations between ambient PM and NO₂,37 thus explaining the O₃ findings with particles does not seem to be very plausible in our study. Air conditioning was not assessed, but was not frequently used in France at the time of the survey. To what extent O₃ may be a marker of other constituents in the toxic mixture of ambient air cannot be answered with our study and the heterogeneity across centres may partly be explained by O₃ indicating in part different characteristics of air pollution across these regions.

In conclusion, we observed significant associations between ambient O₃ concentrations and asthma severity in adults. Results were consistent across two approaches of individual assignment of exposure as well as for two complementary markers of asthma severity. More studies are needed to elucidate the time sequence of pollution effects and to evaluate the impact of exposure of air pollution on asthma severity. But the present results add to the evidence of adverse effects of O₃ at levels far below current air quality standards among susceptible people—namely, those with asthma, supporting more stringent regulations of ambient O₃,38 thought to be set to protect public health. The use of asthma severity scores to investigate the contribution of environmental factors to this complex and dynamic disease appears very promising.

EGEA COOPERATIVE GROUP

Coordination: F Kauffmann; F Demenais (genetics); I Pin (clinical aspects).

Respiratory epidemiology: Inserm U 700, Paris M Korobaeff (Egea1), F Neukirch (Egea1); Inserm U 707, Paris: I Annesi-Maesano; Inserm U 780, Villejuif: F Kauffmann, N Le Moual, R Nadif, MP Oryszczyn; Inserm U 823, Grenoble: V Siroux.

Genetics: Inserm U 393, Paris: J Feingold; Inserm U 535, Villejuif: MH Dizier; Inserm U 794, Paris: E Bouzigon, F Demenais; CNG, Evry: I Gut, M Lathrop.

Clinical centres: Grenoble: I Pin, C Pison; Lyon: D Ecochard (Egea1), F Gormand, Y Pacheco; Marseille: D Charpin (Egea1), D Vervloet; Montpellier: J Bousquet; Paris Cochin: A Lockhart (Egea1), R Matran (now in Lille); Paris Necker: E Paty, P Scheinmann; Paris-Trousseau: A Grimfeld, J Just.

Data and quality management: Inserm ex-U155 (Egea1): J Hochez; Inserm U 780, Villejuif: N Le Moual, C Ravault; Inserm U 794: N Chateigner; Grenoble: J Ferran.