Short-term effects of air pollution on daily asthma emergency room admissions

There is wide evidence showing that asthma-related morbidity has been increasing in recent years 1, 2. Among the population of the Madrid region aged 2–44 yrs, 7.3% have had an attack of asthma at some time and 3% suffer from current asthma. Also, an increase has been seen both in prevalence and in healthcare attendance for asthma, in terms of primary care and hospital emergency settings. This increase in morbidity has led to the study of different environmental factors that might be associated with the risk of developing asthma 3.

The most extensively studied environmental factor has been air pollution, since it can increase the risk of developing asthma attacks through several different mechanisms including: 1) a direct irritant effect on sensitive airways; 2) a toxic effect on the respiratory epithelium; 3) generating bronchial hyperreactivity, both allergen-specific and nonspecific; or 4) modifying the immune response, by increasing susceptibility to an immunological trigger 4, 5. Although experimental studies conducted on animals and humans suggest an effect for sulphur dioxide (SO₂), ozone (O₃) and nitrogen dioxide (NO₂), and more recently, for particulates 4, 5, other studies have yielded contradictory results 6–9. Furthermore, a number of studies have investigated the acute effects using time-series methodology, but here again the results reported have proved somewhat inconsistent concerning the pollutants implicated 8, 10, 11. Inconsistencies in the associations between single pollutants and asthma morbidity can be partly due to differences among studies in the levels of pollutants and in prevailing mixtures 6, 10. Thus, it is of interest to examine the effect of each pollutant adjusted for the others, and the effect of pollutant mixtures. Also, pollen levels can be a potential confounding factor in the relationship between air pollution and asthma. However, few studies have included the measurement of pollen as a confounder to obtain a more valid estimate of the effect of air pollutants 10, 12. Finally, other studies have addressed interactions between pollens and air pollutants on the one hand, and allergies on the other 10, 13.

The city of Madrid offers an excellent challenge to study the relationship between atmospheric pollutants and asthma-related hospital emergencies. Madrid constitutes the centre of a dense metropolitan area. The main source of pollutant emission is road traffic. In addition, Madrid's geographical setting has climatological conditions directly linked to emission levels, with frequent anticyclonic situations in summer and winter, which impede the dispersion of pollutants. Moreover, owing to the prevailing botanical and climatic conditions, Madrid has especially high concentrations of different types of pollen 14. The peak Poaceae pollenisation season from May–June accounts for 20% of all asthma attacks that take place throughout the year in Madrid 3.

The objective of this study was to analyse the short-term association between asthma emergency room admissions andair pollutants in Madrid during 1995–1998, controlling for thepotential confounding effect of pollen levels, as well astocompare results obtained by using different statistical methods.

Materials and methods

Results

A total of 4,827 asthma attacks were registered during the period 1995–1998, with a daily mean of 3.3 and range of 0–26 emergencies (table 1⇓). A total of 50% of all attacks involved children aged 0–14 yrs, 25% of whom were aged <5 yrs, and 52.3% were female. The temporal distribution for daily asthma emergency room admissions registered a seasonal pattern, with two epidemic peaks occurring in the second fortnight of May 1996 and May 1998 (fig. 1⇓).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.—

Distribution of asthma-related hospital emergencies and air pollution levels in Madrid during 1995–1998.

PM₁₀ and SO₂ registered a predominantly winter-based pattern, whereas NO₂ was distributed homogeneously and O₃ showed a strong seasonal component that peaked during the summer months (fig. 1⇑). In general, pollution levels remained below the standards proposed by the European Community. PM₁₀, SO₂ and NO₂ were positively correlated, while O₃ was negatively correlated with these pollutants and with SO₂ in particular (table 2⇓). The levels of O. europaea, Plantago sp., and Poaceae had a strong seasonal distribution showing peak levels in May and June. Urticaceae showed lower levels than other pollens, as well as a bimodal distribution, which increased in April and June.

The core parametric model selected included linear trend, sinusoidal terms of up to the sixth order, dummy variables for years, days of the week and public holidays, linear and quadratic terms for temperature and relative humidity, and alinear term for acute respiratory infections. Although this model showed no residual autocorrelation, it did show overdispersion (ϕ²=1.42). Hence, the model was only corrected for the overdispersion. For its part, the core GAM model selected included a cubic smoothing spline with 72 df to control for trend and seasonality, dummy variables for days of the week and public holidays, and cubic smoothing splines with four df for temperature and two df for relative humidity and acute respiratory infections. This model showed no residual autocorrelation, and also reduced overdispersion (ϕ²=1.11).

Table 3⇓ shows the relative risks (RR) of asthma emergency room admissions for a 10 µg·m⁻³ rise in pollutant levels in models for a single air pollutant, with assessment of individual lags up to the fourth order, by using Poisson regression. All the RR proved positive, and were statistically significant (p<0.01) for some effects of PM₁₀, NO₂ and O₃. The highest magnitude was registered for O₃ and PM₁₀, with an increase of asthma emergency admissions of ∼4% for a 10 µg·m⁻³ rise of air pollutant. The lag that describes the strongest association is day 3, with the exception of O₃ that relates best to the previous day. Furthermore, statistically significant associations were observed in the structure of third- and fourth-order lags for NO₂, and all orders except fourth day lag for O₃. Table 4⇓ sets out the results by means of Poisson regression for the single-pollutant model that yielded the best lag, taking pollen levels into account, individually and jointly. Control for pollen failed to lead to great changes in the magnitude of the effect of the air pollutants.

Multipollutant models were fitted for cold-season pollutants (including PM₁₀ and SO₂, jointly with the four types of pollen) and for photochemical pollutants (including NO₂ and O₃, jointly with the four types of pollen) (fig. 2⇓). Associations remained consistent for PM₁₀ (RR=1.066, 95% confidence interval: 1.027–1.107), NO₂ (1.024, 1.005–1.045) and O₃ (1.031, 1.003–1.060), and remained unaltered by the introduction of the other pollutants and the four pollen types. SO₂ was the most affected (0.966, 0.925–1.009), becoming negative after the introduction of PM₁₀. Since PM₁₀ and NO₂ showed a rather mixed pattern, other combinations of multipollutant models were accounted for. Although associations for PM₁₀ remained statistically significant (p<0.05), it decreased ∼2% when adjusting for photochemical pollutants (1.044, 1.001–1.090, and 1.044, 1.015–1.073, when adjusting for NO₂ and O₃, respectively). Results for NO₂ did not change when adjusting for SO₂ (1.031, 1.004–1.059), but it was affected after introducing PM₁₀ (1.001, 0.971–1.032), probably due to their high correlation. Finally, slight differences were observed between the estimates obtained with parametric modelling and those obtained with GAM models (PM₁₀: 1.058, 1.025–1.091; SO₂: 0.979, 0.948–1.010; NO₂: 1.022, 1.005–1.040; and O₃: 1.028, 1.006–1.049 (fig. 2⇓)).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.—

Relative risks of asthma emergency room admissions, with 95% confidence intervals, for an increase of 10 µg·m⁻³ in air pollution level, by means of Poisson regression and generalised additive models (GAM) for multipollutant models. Adjustment includes all types of pollen (Olea europaea, Plantago sp., Poaceae, and Urticaceae). *: photochemical air pollutants; ^#: cold season air pollutants; ^¶: particles measured with a median aerodynamic diameter of <10 µm (PM₁₀) with pollen and sulphur dioxide (SO₂) (lag 3); ⁺: SO₂ with pollen and PM₁₀ (lag 3); ^§: nitrogen dioxide (NO₂) with pollen and ozone (O₃) (lag 3); ^ƒ: O₃ with pollen and NO₂ (lag 3). ▪: Poisson regression; □:GAM.

Discussion

The main results yielded by this study suggest that, in thesingle pollutant models, PM₁₀, NO₂ and O₃ are positively and statistically significantly associated with asthma-related hospital emergencies. These associations hold stable for the cold-season and photochemical pollutant models, where the remaining pollutants of each type are simultaneously introduced. Likewise, no variations are in evidence when pollen isincluded as a confounding variable. The association of highest size occurs with PM₁₀. The most immediate effects are seen for O₃, associated with previous-day pollutant levels. These results are in line with those reported by a number of studies based on time-series methodology 10, 22–32.

Particulates have been one of the pollutants most frequently studied in the assessment of acute health effects, particularly in the case of respiratory diseases. Although particulate sources tend to vary from place to place, the chief source of emission is diesel-engine combustion, inasmuch as diesel engines emit 100 times more particulates than gasoline engines equipped with catalysers 33. Laboratory studies confirm that the emission of diesel particulates can have effects on the immune system, by acting as an adjuvant that boosts allergic inflammation 34. One of the first studies ever to analyse the short-term effects of particulates on asthma-related emergencies was conducted on a <65‐yr-old population in Seattle by Schwartz et al. 22, who reported a positive association with PM₁₀ at an average lag of 0–4 days; their figure of a 3.7% increase in emergencies for every 10‐µg·m⁻³ rise in PM₁₀ is very similar to that observed in the current study. In the same period, Rossi et al. 23 also described a positive relationship between total suspended particulates and asthma-related emergencies in the population aged 15–85 yrs. Subsequently, Castellsagué et al. 24 reported an association at an average lag of 0–3 days between black smoke and emergency room visits in Barcelona among subjects aged >14 yrs. Again, the magnitude was of the same order as that described both by Schwartz et al. 22 and the current study. Recently, studies in Santa Clara 25, London 26 and Belfast 27 have also observed a significant association between PM₁₀ and asthma-related emergencies. In contrast, other authors, working in different geographical settings and in populations of different age ranges, have failed to observe this association 28–32, 35.

Furthermore, experimental studies have demonstrated thatSO₂ can induce bronchospasm in nonasthmatics and asthmatics alike. Although the concentrations required provoking these effects in healthy individuals are very high, among asthmatic subjects changes in pulmonary function are brought about at low dose 36. In time-series studies, SO₂ has been the pollutant for which an association with asthma-related emergencies has least frequently been detected. In Vancouver, Bates et al. 37 observed an association between this pollutant and asthma-related emergency visits among subjects aged >15 yrs. Rossi et al. 23 also observed positive results, as did other recent studies in Birmingham 38, TelAviv 39, London 26 and Belfast 27, where a statistically significant relationship was detected with current- or previous-day pollutant levels. Against this, other studies 22, 29, 35, 40, 41 have failed to demonstrate asthma-related effects. In Spain, under pollution conditions very similar to those prevailing in Madrid, Castellsagué et al. 24 and Sunyer et al. 30 in Barcelona, and Tenias et al. 32 in Valencia, have also failed to observe a statistically significant relationship.

In the case of photochemical pollutants, high concentrations of NO₂ act through oxidative mechanisms to produce epithelial cell damage, reduce resistance to infectious agents, alter alveolar macrophage function and favour the release of mediators of inflammation 42. In epidemiological time-series studies, this pollutant has been described more frequently in European cities. Once again, Castellsagué et al. 24 describe significant associations in Barcelona between previous-day pollutant levels and asthma-related emergencies among persons aged >14 yrs. On the other hand, Sunyer etal. 30, also working in Barcelona within the framework of the APHEA study, failed to detect an association in a subsequent series. Recently, Tenias et al. 32 observed a relationship of considerable magnitude in Valencia. Likewise, studies in London 26 and Belfast 27 reported a significant positive association, while other authors have found no such associations 35, 38.

O₃ leads to changes in lung function, which appears to fall off after repeated exposures on consecutive days, and induces bronchial hyperreactivity, both allergen-specific and nonspecific, thereby giving rise to inflammatory changes in the airways 43. Although in Madrid there were only two O₃ measurement stations available, in contrast to a large number of stations for other air pollutants, these were located in areas of the city that they are not directly influenced by the emission of traffic. Furthermore, it has been well guaranteed that the functions, as well as the quality, of data provided were representative for the entire city 44. O₃ has been the pollutant to be most frequently associated with asthma-related emergencies. Cody et al. 40 in New Jersey were among the first authors to report statistically significant associations. Subsequently, positive associations have also been observed in European cities 30, 32, 35, as well as in North America (Mexico 41, Canada 29 and Nevada 28, 31), though the magnitude of the relationship varies considerably, as does the lag that corresponds to the best association, even though this could depend on the relative prevalence of other interacting pollutants in those different places. However, most of the reviewed studies where a positive association with O₃ was found reported the strongest association in early lags, lag 1 the most frequently reported 30, 32, 35, 41. In this sense, the O₃ has a great oxidant capacity, causing inflammation of the airways and bronchial reactivity. It has also been shown that repeated exposures to O₃ cause tolerance. This could explain the relationship with immediate exposures to this pollutant. The mechanisms of action of NO₂ are less well known, although its oxidant capacity is smaller than that of O₃ 9. In contrast, in other European 24, 26 and North American studies 22, 25, 37 of similar setting, this relationship has not been observed.

In epidemiological time-series studies aimed at assessing the short-term effects of air pollution on health indicators, the suitability of basing results on single or multipollutant models is debatable, since separation of the individual effects may prove extremely difficult, owing to collinearity between the various air pollutants. In the current study, PM₁₀ estimates show a slight increase in response to the introduction of SO₂, the other pollutant predominant in winter. SO₂, in contrast, is vulnerable to the introduction of PM₁₀. The reported levels of SO₂ in Madrid were low, showing a decreasing trend. This could be mainly due to the change of coal to gas heaters. Although SO₂ is a strong bronchoconstrictor, it is mainly absorbed in the upper tract when air is uptaken by the nose. In experimental studies, although effects have been observed at low levels, these have been of slight intensity and of quick improvement 4. Then, the association found in single-pollutant models with SO₂ can be due to concealing the effect of other pollutants, and disappear when also adjusting for PM₁₀ due to its high correlation. NO₂ and O₃ remained unaltered on inclusion of the other photochemical pollutant. This behaviour pattern was also in evidence in the study by Tenias et al. 32, where SO₂ proved to be the most vulnerable pollutant, while the relationship with the photochemical pollutants remained unchanged when these were simultaneously introduced.

An aspect of special interest is the possibility of pollen confounding or modifying the association between air pollution and asthma 45. In Madrid, the temporal distribution ofasthma emergencies plots epidemic increases during the second fortnight in May and first fortnight in June, coinciding with the abrupt release of Poaceae and Plantago sp. pollen 3. Recently, Anderson et al. 10 in London, and Fauroux etal. 35 in Paris have also included different types of pollen in the analysis and, like the current study, have observed that the relationships between the different air pollutants and asthma remain independent of, and also not modified by, the effect of such pollen types. Experimental studies have shown that exposure to pollutants, especially particulates and O₃ (but also NO₂ and SO₂), may enhance airway response to environmental allergens in susceptible individuals 46–49. In time-series studies there is still little information on this point. Due to skewed pollen distributions this variable (pollen) was fitted using different transformations, such as the square root, and alternatively as a dummy variable comparing days without pollen versus days with pollen levels 10. However, results for air pollutants did not change when adjusting pollens transformed with square root, while using dummy variables there were slightly changes, <3% for cold-season pollutants and 1% for photochemical pollutants.

With the use of GAM models, bias can occur both due to too lax convergence criteria and concurvity 16, 21. However, the results obtained in this study on the basis of parametric modelling do not differ substantially, in terms of magnitude of association and of choice of best lag, when compared to themore flexible modelling using GAM. This consistence coincides with the GAM‐based reanalysis, which was performed in the APHEA study 50 to assess heterogeneity in the regional results, and which in the case of West European cities yielded similar estimates to those obtained using the parametric methodology of the initial analysis. In this regard, the authors can only agree with the recommendations made by NMMAPS 16 and APHEA 21 researchers to the effect that cautious use must be made of GAM models, with care being taken to avoid automatic restriction to the default convergence parameters included in statistical programmes, in order to prevent possible biases in the estimates. The authors have also considered the daily number of acute respiratory infection cases because it is widely demonstrated that this is an important cause of asthma, or even of asthma attacks. Rhinovirus is the main virus that causes asthma exacerbations 51. However, in most of the time-series studies to assess the short-term effects of air pollution on asthma researchers used influenza because it is easier to obtain from registers. However, the authors found a best goodness of fit, in terms of reduction of core model likelihood, when adjusting for thedaily number of acute respiratory infections instead of influenza. When available it should be desirable to use information on acute respiratory infections because it is more directly related with asthma.

This study suffers from the usual limitations of ecological designs, particularly insofar as lack of accuracy in the measurement of exposures is concerned. Despite the small magnitude of the associations, these may be relevant to the public heath point of view. In terms of attributable risk, the association observed with PM₁₀ would imply that if the levels of this pollutant were reduced an average of 5 µg·m⁻³, it could prevent between 0.5–3.3% of the asthma emergencies, that is, between six and 39 annual asthma attacks. However, these results must be interpreted with caution, given that variables not controlled for in the model or complex relationships between pollutants may confound the results. For instance, it has recently been observed that when other pollutants that are not usually measured, such as benzene, are incorporated into multipollutant models, the relationship with the classic pollutants decreases to a great extent 27, 52. Caution must therefore be exercised when it comes to drawing conclusions of a causal nature.

In conclusion, and taking into consideration the limitations of this study, the evidence produced to assess the association between air pollution and asthma-related hospital emergencies in this study, suggests that the usual air pollution levels in Madrid, specifically those of particles measured with a median aerodynamic diameter of <10 µm, ozone and nitrogen dioxide, are associated with an increase in asthma emergency room admissions, and that this effect is independent of ambient pollen concentrations. The authors also advise the use of generalised additive models, which provides the best fit, for the reduction of both autocorrelation and overdispersion.