Air pollution and daily admissions for chronic obstructive pulmonary disease in 6 European cities: results from the APHEA project

Air pollution and daily admissions for chronic obstructive pulmonary disease in 6 European cities: results from the APHEA project. H.R. Anderson, C. Spix, S. Medina, J.P. Schouten, J. Castellsague, G. Rossi, D. Zmirou, G. Touloumi, B. Wojtyniak, A. Ponka, L. Bacharova, J. Schwartz, K. Katsouyanni. ERS Journals Ltd 1997. ABSTRACT: We investigated the short-term effects of air pollution on hospital admissions for chronic obstructive pulmonary disease (COPD) in Europe. As part of a European project (Air Pollution and Health, a European Approach (APHEA)), we analysed data from the cities of Amsterdam, Barcelona, London, Milan, Paris and Rotterdam, using a standardized approach to data eligibility and statistical analysis. Relative risks for daily COPD admissions were obtained using Poisson regression, controlling for: seasonal and other cycles; influenza epidemics; day of the week; temperature; humidity and autocorrelation. Summary effects for each pollutant were estimated as the mean of each city's regression coefficients weighted by the inverse of the variance, allowing for additional between-cities variance, as necessary. For all ages, the relative risks (95% confidence limits (95% CL)) for a 50 μg·m-3 increase in daily mean level of pollutant (lagged 1–3 days) were (95% CL): sulphur dioxide 1.02 (0.98, 1.06); black smoke 1.04 (1.01, 1.06); total suspended particulates 1.02 (1.00, 1.05), nitrogen dioxide 1.02 (1.00, 1.05) and ozone (8 h) 1.04 (1.02, 1.07). The results confirm that air pollution is associated with daily admissions for chronic obstructive pulmonary disease in European cities with widely varying climates. The results for particles and ozone are broadly consistent with those from North America, though the coefficients for particles are substantially smaller. Overall, the evidence points to a causal relationship but the mechanisms of action, exposure response relationships and pollutant interactions remain unclear. Eur Respir J 1997; 10: 1064–1071. *Dept of Public Health Sciences, St. George's Hospital Medical School, London, UK. **GSF National Research Centre for Environment and Health, Institute for Epidemiology, Neuherberg, Germany. ***Observatoire Régional de la Santé, Paris, France. +Dept of Epidemiology and Statistics, University of Groningen, The Netherlands. ++Institute Municipal D'Investigacio Medica, Barcelona, Spain. +++Institute of Clinical Physiology, National Research Council, Pisa, Italy. ‡Faculté de Médicine, Université de Grenoble, France. ‡‡Dept of Hygiene and Epidemiology, University of Athens Medical School, Greece. ‡‡‡National Institute of Hygiene, Warsaw, Poland. #Helsinki City Centre of the Environment, Finland. ##National Centre for Health Promotion, Bratlslava, Slovakia. ###Harvard School of Public Health, Boston, USA.

There is considerable evidence that severe air pollution episodes may be associated with an increase in morbidity and mortality [1,2]. Recent studies have found that daily morbidity and mortality may also be associated with levels of air pollution which are well below those observed in episodes and are within current air quality standards [3]. A vulnerable group is likely to be older people with pre-existing cardiorespiratory disease, including chronic obstructive pulmonary disease (COPD) [1, 3,4]. This condition is characterized by chronic and usually progressive impairment of airflow due to obstruction, damage and disorganization of the airways, as well as to loss of alveolar tissue. Advanced stages of the disease are associated with poor respiratory reserve, and affected individuals are likely to be especially vulnerable to additional stress on the respiratory system, such as might be caused by the toxic effects of inhaled pollutants. Evidence from panels of patients with COPD suggests that they experience small reductions in lung function in association with increased pollution levels in the ambient range [5,6]. Hospital admissions for COPD might, therefore, be a sensitive indicator of the adverse effects of outdoor air pollution. Studies from Birmingham (AL, USA) [7], Detroit (MI, USA) [8], Minneapolis-St Paul (MN, USA) [9], Ontario (Canada) [10] and Spokane (WA, USA) [11] have reported associations between daily admissions for COPD and particulate and ozone pollution.
In the Air Pollution and Health, a European Approach (APHEA) collaboration, a standardized prospective approach was used to examine the short-term effects of air pollution on mortality and morbidity in a wide range of European cities [12]. In six cities (Amsterdam, Barcelona, London, Milan, Paris and Rotterdam) data on admissions for COPD were analysed. In this paper, we present the results of an analysis, in which the individual city results have been combined using meta-analytical techniques to provide summary estimates of the relative risks of daily admissions for COPD associated with ambient levels of sulphur dioxide (SO 2 ), nitrogen dioxide (NO 2 ) ozone (O 3 ) and particles (black smoke (BS) or total suspended particulates (TSP)). Papers concerned with all emergency respiratory admissions (International Classification of Diseases 9th Revision (ICD9) 460-519) and asthma (ICD9 493) will be published separately.

Methods
Details of each city's methods have been reported previously [13][14][15][16][17]. From routine sources, daily counts of emergency hospital admissions for ICD9 490 (unspecified bronchitis), 491 (chronic bronchitis), 492 (emphysema) and 496 (chronic airways obstruction) were obtained. For the purpose of this analysis, these four codes comprise COPD. In Barcelona, data collection was part of a special project. The admissions covered all hospitals in each city which admit medical emergencies, except for Barcelona, where six participating hospitals, which cover 90% of emergencies, provided data on emergency COPD admissions. In Milan and Paris, emergency admissions could not be separated from total admissions, but based on an analysis of London data, which found that 95% of COPD admissions in that city were "immediate", it is likely that the large majority of COPD admissions in Paris and Milan were also "immediate" i.e. unplanned emergency admissions. The proportion of all medical admissions with diagnostic coding was over 90% in all cities except London, where it rose from 73 to 95% during the period of study. These systems record the diagnosis of the condition responsible for admission at the time of discharge.
The effects of pollutants that were already available from routine monitoring systems measuring background concentrations were studied. The criteria for inclusion of monitors and for dealing with missing values were decided in advance by the APHEA group [12]. SO 2 and NO 2 were analysed as 24 h and maximum 1 h means for each day. Indicators of particles used in this analysis were TSP and BS analysed as 24 h values. Ozone was analysed as an 8 h mean (09:00 to 17:00 h) and as the daily maximum 1 h mean. Temperature and humidity were analysed as mean 24 h values.
The APHEA group agreed the analytical approach in advance to ensure the maximum degree of comparability. This was Poisson time series regression controlling for trend, seasonal and other cycles down to 2 months (6 weeks in the case of London) [15], day of the week, holidays, influenza epidemics, temperature, humidity and autocorrelation [18,19]. Within the constraints of the agreed approach, each centre analysed their data individually rather than on a pooled basis. This decision was made because factors, such as access patterns, pollution mixtures, climate and seasonal influences differed between cities, and might need to be taken into account on a city by city basis. Similarly, each centre determined for each pollutant the best 1 day lag (up to 3 days) and cumulative lag (the mean over several previous days) for each pollutant. For ozone, up to 5 days were allowed. For centres with higher levels of pollution (days above 200 µg·m -3 ), the exposure response was sometimes logarithmic, flattening out at higher levels. To simplify the meta-analysis, each centre fitted a linear relationship between the pollutant and COPD admissions for days below 200 µg·m -3 . In addition to all year models, coefficients were estimated for the cool (October to March) and warm (April to September) seasons separately.
The summary effect of each air pollution indicator on COPD admissions was estimated by calculating the weighted mean of each city's regression coefficients, the weights being inversely proportional to the local variances. The weights were calculated assuming a fixed effects model when a Chi-squared test failed to detect heterogeneity at the sensitive level of alpha=20%. When the assumption of homogeneity had to be rejected, a random effects model was adopted; this gives weights which are more similar between cities but a larger variance, reflecting greater uncertainty about the summary estimate when local results are heterogeneous [20].
Where heterogeneity was observed, we tried to explain this using weighted linear regressions of local coefficients on non-time-dependent properties of the cities, including indicators of general population health status, climate, quality of outcome data, quality of pollutant data, and pollutant features, such as heterogeneity in overall levels of pollution within and between cities. Table 1 summarizes the health and environmental data used in the analysis. The cities varied considerably in size and environmental characteristics. The median summer temperature ranged from 14°C in Amsterdam and Rotterdam to 22°C in Milan, and the median winter temperature from 5°C in Amsterdam and Rotterdam to 13°C in Barcelona. The contrast between summer and winter temperatures (expressed as percentage difference) was greatest for Amsterdam (47%), Rotterdam (47%) and Milan (48%), intermediate for London (30%) and Paris (29%), and lowest for Barcelona (16%). For pollution levels, the highest and lowest cities, respectively, were: SO 2 , Milan and Amsterdam; NO 2 , London and Paris; BS, Barcelona and Amsterdam; and O 3 , Amsterdam and Paris.

Results
The daily number of COPD admissions varied from 1 (Amsterdam and Rotterdam) to 20 (London). The proportion of COPD admissions for patients aged ≥65 yrs ranged from 48% in Paris to 70% in Barcelona. Figure 1 a-e shows, for the 24 h concentration of each pollutant (8 h for ozone), the relative risks for each city and the summary estimate for an increase of 50 µg·m -3 in each pollutant. Table 2 shows the relative risk estimates for single day and cumulative lags, and also includes maximum 1 h values for SO 2 , NO 2 and ozone.
The effect of SO 2 varied considerably across cities but the summary estimate was significant for the 1 h measure and borderline significant for the daily mean.
The lags were inconsistent, being either same day or day 2. This heterogeneity was largely from Amsterdam and Rotterdam; there were indications that this was associated with the use of fewer monitoring stations (one compared with three or more in the other cities) and relatively low temperatures.
The effect of particles was more consistent, and the summary estimates for BS and TSP were both statistically significant or borderline significant. The lags varied from same day to day 2. For increases of 50 µg·m -3 in BS and TSP all-age COPD admissions were increased by 3.5% (95% CL 1-6) and 2.2% (95% CL 1-5), respectively.
Both 24 h and 1 h NO 2 were significantly associated with COPD admissions and the cities tended to be consistent, apart from Amsterdam which was a negative outlier causing heterogeneity in the 1 h NO 2 effects. This heterogeneity could not be explained using the variables described in the Methods section, as there were few differences between Amsterdam and Rotterdam in these respects. The lags varied from 0 to 2 days. An increase of 50 µg·m -3 in 24 h NO 2 was associated with a 1.9% increase in admissions (95% CL 0-5).
The most consistent and significant findings were for ozone, and there was no significant heterogeneity between the cities. The lags varied from 0 to 2 days. A 50 µg·m -3 increase in 8 h ozone was associated with a 4.3% (95% CL 2-7) increase in admissions. With one exception, all the cities had a zero or 1 day lag. The exception was Rotterdam (also the smallest city), which had a lag of 2 days.
In general, the use of cumulative lags did not give stronger effects than single day lags. Table 3 shows the effects of pollutants in the warm and cool seasons separately. In the warm season, significant or borderline significant effects were observed for SO 2 , NO 2 and ozone, but not for BS or TSP. In the cool season, marginally significant effects (10% level) were obtained for BS and ozone. The difference in pollution effect between warm and cool seasons was significant only for 8 h ozone, with a much stronger effect in the warm season.
Too few cities provided analyses of the ≥65 age group to justify meta-analysis of this age group. However, because most other reports of the effects of air pollution on COPD admissions have been confined to this age group, the results will be mentioned here. London, Milan and Paris analysed the effects of 24 h SO 2 on this age group; all the relative risks were positive and of similar size: 1.043 (NS), 1.069 (p<0.05) and 1.048 (p<0.1), respectively. All three individual effects were higher than the all-ages summary estimate of 1.022.  1977-1989 1986-1992 1987-1991 1980-1989 1987-1992 1977-1989   The summary estimate of 1.053 for SO 2 for the ≥65 age group was significant (p<0.05). London and Paris showed similar effects of BS on COPD in the ≥65 age group (1.039 and 1.032, respectively). These effects were not significant and the summary estimate of 1.034, while not significant, was almost identical to the all-ages summary estimate of 1.035.

Discussion
In a prospective standardized study of six European cities, the effect of various air pollutants on daily admissions for COPD were analysed using a Poisson regres-sion technique. Using meta-analytical statistical techniques to combine the individual city effects, it was found that relative risks were significantly increased for a number of pollutants. The most consistent effects were for ozone in the warm season, but significant effects were also observed for SO 2 , NO 2 and measures of particles (TSP and BS). The concentrations of pollutants were generally well within World Health Organization (WHO) Guidelines for health protection in Europe [21].
This study differs from previous meta-analyses [3,11] in that the meta-analysis was a prospective part of the APHEA project and analysed a wider range of pollutants. The parametric Poisson regression approach chosen has some potential deficiencies. Complex seasonal patterns might not be appropriately modelled by harmonic waves, while other estimates potentially depend on the assumed shape of the exposure response curve [19,22,23]. However, investigations into the sensitivity of this approach using different models and comparisons with more sophisticated nonparametric techniques suggest that the approach used in the present study is quite robust [8,9,24]. Control for meteorological variables is always a critical issue in temporal air pollution studies. A recent study, which compared the synoptic approach, favoured by some biometeorologists, with controlling for weather variables using the same methods as the APHEA collaboration found that similar results were obtained [25]. The interpretation of the present data should take into account that each centre selected the lag which gave the greatest effect, rather than a priori. This policy was agreed because at the outset of the study there was insufficient epidemiological or biological information upon which to base an a priori hypothesis as to lag, and there was the strong possibility that different environments and health care systems might be associated with different lags.
The final meta-analysis of the results has to be judged differently from retrospective meta-analyses published previously, as it was planned from the start, and care  was taken to ensure standardized procedures where possible. There is no bias in selection of cities for study or subsequent analysis. The estimate of pollutant exposure, being based on one or several city background monitors was necessarily imprecise because ambient concentrations probably vary throughout the city due to the varying nature of emission sources, topography, air mixing, dispersal and removal processes. Furthermore, indoor levels, which comprise the main exposure, do not necessarily reflect outdoor levels. For example, ozone levels are lower indoors but small particles may penetrate indoors quite easily. Nitrogen dioxide may be higher indoors due to indoor combustion sources. If outdoor levels correlate with indoor levels, then the present estimates will be biased but still represent an association with outdoor air pollution. If, on the other hand, the misclassification is random, the tendency will be for the present effects to be biased towards the null, leading to an underestimate of effects [26].
The diagnosis of COPD is known to be subject to variation both within and between countries [27,28]. Furthermore, several of the cities in this study included a small proportion of nonemergency admissions. It is possible, therefore, that the clinical spectrum of COPD admissions differed from city to city. This might lead to some variation in the size of the estimate, depending on how misclassification affected the average sensitivity of the group coded as COPD, but we are unable to estimate whether and to what extent this occurs.
Are these effects likely to be causal? Although the associations observed are unlikely to be due to chance, the small size of the relative risks raises the question of whether the results could be explained by unknown confounding factors or inadequate control of known confounders. Being an observational ecological study, such a possibility cannot be disproved; however, the consistency of some of the findings together with reports of significant effects on COPD admissions in North America [7][8][9][10][11] suggests that this is less likely.
Few previous studies of this type have systematically analysed pollution effects by season. We observed that the size and significance of effects tended to differ between cities, though the winter/summer differences were statistically significant for one pollutant (ozone). Seasonal differences might have several explanations. One is that there is a threshold effect, exceeded mainly in one season. A second is that the effect depends on complex interactions with the rest of the pollution mix, which also varies seasonally. A third is that some of the associations observed for a particular pollutant are due to confounding by factors, which themselves vary by season. At present, we have insufficient information to explain this variation.
The other important issue is that of plausibility. Patients with advanced COPD tend towards a state of respiratory failure, in which blood levels of oxygen and carbon dioxide become abnormal and which, in turn, leads to problems in other systems, such as the circulatory system. They are particularly susceptible to acute chest infections. It is plausible that such patients might be made worse by the toxic inflammatory effect of small increases in atmospheric pollution. Experimental chamber studies indicate that both healthy subjects and those with asthma or COPD exhibit considerable individual variability in susceptibility to SO 2 , ozone and NO 2 [29][30][31][32]. However, there is little evidence from chamber studies that ambient levels of ozone [33,34] or NO 2 [35][36][37][38][39], have clinically significant effects on COPD patients. It is conceivable that the disparity between ambient and chamber studies may arise because chamber studies do not involve very severe patients or because they are inadequate for simulating the pattern and duration of ambient exposure or the complex mix of pollutants found in the ambient situation. Panel studies do, however, suggest that patients with COPD may experience small short-term effects at ambient levels of particles, NO 2 and SO 2 [5,6]. It should be noted that, in the ambient situation, the actual exposure of some individuals may be substantially higher or lower than indicated by a background monitor. Previous North American studies of air pollution and COPD admissions have been concerned mainly with the effects of particles or ozone (table 4). The present findings confirm the presence and scale of effect reported for ozone, and suggest that this is a widespread and fairly consistent phenomenon on both sides of the Atlantic.
The results for particles are less consistent than those for ozone. While the APHEA study found that there are significant effects of particles on admissions for COPD, the coefficients were considerably smaller than in the North American studies.
Most of the North American studies have been confined to the ≥65 yrs age group, whereas 50-70% of European COPD admissions were in this age group. This could account for some of the transatlantic differences in particle effects but is unlikely to be the main explanation, because in the few European cities where analyses were available for the ≥65 yrs group the coefficients for particles were similar to those for all ages, and the coefficients for SO 2 , though higher in the ≥65 yrs group, were not significant. Furthermore, the effects of ozone were similar in both continents. The smaller size of particle effects in Europe may be explained by differences in the chemical nature and size distribution of the particle mixture, as well as by differences in the composition of the whole pollution mixture. There is increasing interest in the role of the fine (<2.5 µm) and ultrafine (<0.1 µm) fractions, of which a substantial part is composed of chemical particles, such as sulphates, nitrates and acid aerosols [40][41][42]. These are inadequately indicated by BS or TSP.
We have established that significant associations between air pollution and daily admissions for chronic obstructive pulmonary disease can be detected, and that among the candidate pollutants, the associations with ozone are the strongest and most consistent. Further research is required to examine interactions between pollutants, exposure response relationships and health impact. The comparatively smaller effects of particles in Europe compared with North America should be investigated further by using comparable measurements of particles and taking into account differences in the overall pollution mixture. For the present, our results indicate that current levels of air pollution are likely to be harmful to people with chronic obstructive pulmonary disease and that policies to further reduce air pollution should be continued.