Short-term effects of air pollution on hospital admissions of respiratory diseases in Europe: a quantitative summary of APHEA study results

Short-term effects of air pollution on hospital admissions of respiratory diseases in Europe: a quantitative summary of APHEA study results – Air Pollution and Health: a European Approach

Claudia Spix

THERE IS INCREASING INTEREST in the use of hospital admission data in studies of short-term effects of air pollution on health. This reflects the improved availability of admission data from routine systems and the possibility that, as a health outcome, admissions may offer some advantages over mortality data. Advantages include the likelihood that the diagnosis will be more accurate, the possibility that admission may be a more sensitive indicator of pollution effects, and the availability (for respiratory disease) of information on children and young adults. On the other hand, this source of data may be influenced by access to the health-care system and behavioral patterns.

The respiratory disease group mainly comprises infections of the lung or obstructive airways disease, either in acute form (asthma) or chronic form (chronic obstructive pulmonary disease [COPD]). As age increases, there is more overlap between COPD and asthma clinically, and it has been demonstrated that European doctors differ substantially in their diagnoses of these two conditions.[1] In all groups, infection is one of the major reasons for an exacerbation of airways disease. From this it follows that in a European study it might be better to study time series of all respiratory diagnoses than individual diagnoses; also, higher numbers give more statistical power. A possible disadvantage is the biasing of effect sizes to the null, if a relevant fraction of the admissions were insensitive to air pollution.

From what is known about toxicity of the air pollutants studied in Europe, it is possible that ambient concentrations might affect the respiratory system.[2] Whereas healthy people are unlikely to experience anything more than minor effects on the airways that are not associated with symptoms, subjects with preexisting disease could experience a worsening of symptoms that might precipitate an admission to hospital-or even death.

To date, most studies of respiratory hospital admission have been done in North America, where associations have been reported for particles and ozone, and to a lesser degree for sulfur dioxide ([SO.sub.2])-4-10 Fewer studies have been reported from Europe, and with the exception of the study by Walters et al.,[11] most researchers have focused on specific respiratory causes.

The Air Pollution and Health: a European Approach (APHEA) study is a Europe-wide collaborative effort to investigate the short-term effects of air pollution and to standardize the methods for analyzing epidemiological time series of counts. The health endpoints were total mortality, selected cause specific mortality, and respiratory emergency hospital admissions; details are described elsewhere.[12] A unique feature is the use of standardized methods of data selection and analysis for each center (see Katsouyanni et al.[13] for details). In this article, we present a quantitative summary of the respiratory hospital admissions results from five APHEA cities (i.e., London, Amsterdam, Rotterdam, Paris, and Milano). Results of admission time series of specific respiratory subgroups (i.e., COPD, asthma) are presented elsewhere.[14,16]

Material and Method

Material. Selected respiratory causes were defined by international Classification of Diseases (ICD9)-460-519. Except for asthma, these causes apply mainly to adults. They were analyzed separately for adults (i.e., 15-64 y of age) and the elderly (i.e., 65 y and above).

We obtained daily admission data from routine registers in all cities. The registration covered all hospitals in London, the Netherlands, and Milano, as well as hospitals selected for admitting short-stay patients in Paris. Registration was almost complete in the Netherlands, 92% in Milano, and 90% in Paris, and registration rose from 73% to 95% during the study period in London. The diagnosis by which the cases used here were selected was defined as the diagnosis at or after discharge (i.e., when all examination results were evaluated).

An overview of the data available for meta-analysis is provided in Table 1. When possible, we used daily counts of emergency admissions because this is likely to be a more sensitive indicator than general admissions. We were unable to differentiate between emergency and nonemergency admissions in Paris and Milano.


The pollutants studied were those generally available in European cities: sulfur dioxide (SO.sub.2]); nitrogen dioxide ([NO.sub.2]); ozone ([O.sub.3]); and indicators of particulate matter, Black Smoke (BS), and total suspended particulates (TSPs). The levels of [SO.sub.2] and [NO.sub.2] were obtained as a daily mean and a 1-h maximum, particles were obtained as a daily mean, and03was obtained as a daily 8-h maximum (9 A.M.-5 Pm.) and a 1-h maximum. (Table 1).

Each center performed the Poisson time series regressions of their data individually. This approach was necessary because the amount of data studied was too large to be studied at one place; it was also desirable because access patterns, weather patterns, and special events (e.g., strikes, holidays, epidemics) differ between places; therefore, we accounted for each city separately. The confounders included in each city were trend; seasonality; calendar effects (e.g., day of week, holidays); unusual events (strikes, reorganizations) as applicable; and meteorology (i.e., temperature and humidity). Where necessary, we included an autoregressive error term. For pollutants, each center determined a best-fitting 1-d result that allowed a delay of up to 3 d (5 d for 03), as well as a best-fitting cumulative result that corresponded to the mean of the same day and from up to 3 d earlier (5 d for [O.sub.3]). Each center established its own definition of warm season and cold season, depending on local climatic conditions, but in the main, April-September was considered warm. We used local medians for defining a pollutant as high or low for models testing effect modification of one pollutant by the level of another. Poisson regression coefficients can be expressed as relative risks per units of change.

Details about the principles of the time series analysis for this type of data are available elsewhere,[17] and information about the practical rules set by researchers to ensure maximum comparability–but allowing for the necessary flexibility within APHEA–is presented by Katsouyanni et al.[13] The individual center’s results are published elsewhere.[15,18-24]

Method. The protocol required each center to fit the dose-response curve transformation that suited its data best. These were mostly nontransformed or log-transformed values–the latter describing a flattening of the dose-response curve with higher pollution levels. Given that these log-transformed curves tend to fit better when higher levels of air pollution are studied, we, in an effort to facilitate meta-analysis, refitted those models by using untransformed pollution values and by deleting all days from the series on which pollution levels exceeded 200 mg/[m.sup.3]. The relative risks given herein, therefore, apply best to relatively low levels of air pollution and should not be extrapolated, especially for the winter-type pollutants. Information about the conditions under which transformed curves are better fitted is available elsewhere.[15,18-24]

To provide a quantitative summary of results across the centers, we applied methods of meta-analysis by obtaining a pooled regression coefficient as a weighted mean of local regression coefficients–the weights being inversely proportional to the local variances. We performed calculations only for endpoint-pollutant combinations available from three or more countries, except for particulate matter, for which this restriction would have likely prohibited meta-analysis completely. Consequently, the particulate matter results were less stable.

We determined the weights, assuming a fixed-effects model, when a chi-square test failed to detect heterogeneity at the sensitive level of [Alpha] = 20% (see Appendix). When we had to reject the assumption of homogeneity, a random-effects model seemed more appropriate. In this model, the between-cities variance is added to each estimated local variance, thus giving more similar weights, but also a larger variance; this approach was an appropriate way of expressing that we were less sure of the pooled result in the case of heterogeneous local results. The local weights are expressed in the figures as “bubbles” of corresponding size around the parameter estimate. The between-cities variance can be estimated in several ways, we used an iterative ML-approach.[25]

In those instances in which heterogeneity was present and coefficients from at least five cities were available, we sought explanations for this in the form of weighted linear regressions of local coefficients on nontime-dependent properties of the cities in question. Candidates that might describe differences in sensitivity between populations or sources of bias were as follows: (a) indicators of the general population health status (e.g., age-standardized mortality rate, life expectancy, proportion of elderly, mortality rates from respiratory causes, smoking prevalence); (b) climate indicators (e.g., temperature, humidity–by season) during the study period and latitude; (c) an indicator of the outcome data quality differentiating between general and emergency admissions; (d) indicators of pollution data quality (e.g., number of stations, inhabitants represented per station, correlation between stations); and (e) indicators of the air pollution situation (e.g., number of inhabitants, pollution level, correlations between pollutants).


Daily counts of adult respiratory admissions were not associated consistently with daily mean [SO.sub.2]. A random-effects model was necessary, and the pooled coefficient was close to 0. The heterogeneity between the cities was explained either by the number of stations measuring [SO.sub.2] or mean winter temperature or by the mean life expectancy. Amsterdam and Rotterdam had (a) only one measuring station, (b) the lowest mean winter temperature (2.5 [degrees] C) of all cities, and (c) the highest life expectancy (77 y), and no adverse effect of [SO.sub.2] could be detected. It should be noted, however, that differences in life expectancy were very small in the cities examined here (between 75 and 77 y), and the association could be explained by chance. Misclassification of exposure via use of just one station may have biased the measurable effect to the null. The [SO.sub.2] measurements in the other three cities were based on four stations in each of these cities, and effects (i.e., small, nonsignificant) were seen. In the elderly age group, results were homogeneous, and only in Paris were they mostly positive and significant. However, the joint parameter for the daily mean was significant, and we expected an overall increase of 2% (95% confidence interval [CI] = 1, 5) in elderly admissions with a concomitant increase in [SO.sub.2] of 50 [micro]g/[m.sup.3] (Table 2).

Table 2.–Summary Effects of Pollutants on Daily Respiratory Admissions as Relative Risk (RR) per 50[micro]g/[m.sup.3] Increase in Pollutant

Pollutant Cities Age group (y)

[SO.sub.2] daily mean L, A, R, P, M 15-64


BS daily mean L, A, R, P 15-64


TSP daily mean A, R, M 15-64


[NO.sub.2] daily mean L, A, R, P 15-64


[NO.sub.2] daily maximum 15-64


[O.sub.3] 8-h average L, A, R, P 15-64


[O.sub.3] 1-h maximum 15-64


Pollutant RR 95% CI

[SO.sub.2] daily mean 1.009 0.992, 1.025

1.020(*) 1.005, 1.046

BS daily mean 1.028(*) 1.006, 1.051

1.020 0.996, 1.046

TSP daily mean 1.010 0.989, 1.031

1.016 0.994, 1.039

[NO.sub.2] daily mean 1.010 0.985, 1.036

1.019 0.982, 1.060

[NO.sub.2] daily maximum 1.004 0.996, 1.011

1.005 0.977, 1.033

[O.sub.3] 8-h average 1.031 1.013, 1.049

1.038(*) 1.018, 1.058

[O.sub.3] 1-h maximum 1.019(*) 1.005, 1.033

1.031 1.015, 1.047

Notes: [SO.sub.2] = sulfur dioxide, BS black smoke, TSP total suspended particulates, RR = relative risk, CI = confidence interval, A = Amsterdam, L = London, M = Milano, P = Paris, and R = Rotterdam.

(*) Significant at 5% level.

Although results within the Netherlands seemed unstable, most Black Smoke regression results for adult admissions tended to be positive. The joint effect was small, but it was positive and significant, and we expected a 3% increase (95% CI = 1, 5) in admissions with a concomitant increase in BS of 50 [micro]g/[m.sup.3]. The 1-d effect was larger than the accumulated effect. No effect of TSP was visible. For the elderly–and for both TSP and BS–the effects were close to 0, without heterogeneity. The TSP effects were slightly larger, but still were not significant (Table 2, Fig. 1).


An inconsistent picture was displayed by the [NO.sub.2] regression results in both adults and elderly admissions. The Netherlands’ results had large random variation, and random-effects models were needed. Only the pooled result for accumulated daily mean [NO.sub.2] was borderline significant, and we estimated an almost 2% increase (95% CI = 0, 3) with an [NO.sub.2] increase of 50 [micro]g/[m.sup.3]. Associations with other [NO.sub.2] indicators and adult respiratory admissions were smaller. For the elderly, only Rotterdam reported a consistently positive, significant association. Overall, there appeared to be no evidence of an [NO.sub.2] effect in either age group (Table 2).

The [O.sub.3] results in the adult group showed good agreement between cities. In London, a significantly positive association was seen with each type of [O.sub.3] indicator, and most results from the other cities also had a positive tendency. London and Paris results were very similar, joint results were positive and significant and were even more so among the elderly. The strongest association was with the daily 8-h average. The most common lag was with the same or previous day; therefore, we may term the effect “almost immediate.” An approximate 3% increase (95% CI = 1, 5) in adult admissions, and an approximate 4% (95% CI = 2, 6) in elderly admissions, were estimated with a concomitant increase of 50 [micro]g/[m.sup.3] daily 8-h average (Table 2, Fig. 2).


Results, by season. The by-season models were less stable than the all-year results and varied strongly among cities. Consequently, a random-effects model was almost always required. No seasonal differences were significant, except for one, but some trends became apparent.

The small effect of 1-d[SO.sub.2] in the elderly (2% all year) resulted from an effect in the warm season, although this difference was not significant (Table 3). The differences in BS effects between seasons were quite large, but not significant, and they were quite inconsistent between 1-d and cumulative effects. This points at instability–and at little else (Table 3). Though based on even fewer cities (n = 3), with respect to the adults, TSP indicates rather consistently–and, for cumulative effects even significantly–a stronger effect in the warm season (1-d TSP 3% [CI = 0, 61 per 50 [micro]g/[m.sup.3]). This was not the case in the elderly (Table 3). Nitrogen dioxide had no effect on either age group during either season (Table 3). The effects of [O.sub.3] on adults were similar in both seasons, but in the elderly they were slightly larger in the warm season (Table 3, Fig. 3).


Table 3.–Summary Effects of Air Pollutants on Respiratory Admissions, by Season

Pollutant Cities Season(*)

[SO.sub.2] daily

mean L, A, R, P, M Warm


BS daily

mean L, A, R, P Warm


TSP daily

mean A, R, M Warm


[NO.sub.2] daily

mean L, A, R, P Warm


[NO.sub.2] daily

maximum Warm


[O.sub.3] 8-h

average L, A, R, P Warm


[O.sub.3] 1-h

maximum Warm


15-64 y olds

Pollutant RR([dagger]) 95% CI

[SO.sub.2] daily

mean 1.01 0.98, 1.04

1.01 0.97, 1.07

BS daily

mean 0.99 0.90, 1.09

1.04([double dagger]) 1.02, 1.07

TSP daily

mean 1.03([sections]) 1.00, 1.06

0.97 0.93, 1.02

[NO.sub.2] daily

mean 1.00 0.96, 1.04

1.01 0.98, 1.04

[NO.sub.2] daily

maximum 1.00 0.99, 1.02

1.00 0.98, 1.01

[O.sub.3] 8-h

average 1.02 0.99, 1.05

1.03 0.98, 1.08

[O.sub.3] 1-h

maximum 1.01 0.99, 1.05

1.02 0.99, 1.05

65 + y olds

Pollutant RR([dagger]) 95% CI

[SO.sub.2] daily

mean 1.06([double dagger]) 1.01, 1.11

1.02 0.99, 1.04

BS daily

mean 1.07([double dagger]) 1.00, 1.15

1.00 0.95, 1.04

TSP daily

mean 1.01 0.98, 1.04

1.02([sections]) 1.00, 1.05

[NO.sub.2] daily

mean 1.02 0.99, 1.06

1.00 0.98, 1.03

[NO.sub.2] daily

maximum 1.00 0.98, 1.02

1.00 0.98, 1.03

[O.sub.3] 8-h

average 1.04([double dagger]) 1.02, 1.07

1.02 0.99, 1.05

[O.sub.3] 1-h

maximum 1.04([double dagger]) 1.02, 1.05

1.03([sections]) 1.00, 1.06

Notes: RR = relative risk, CI confidence interval, [SO.sub.2] sulfur dioxide, BS = black smoke, TSP total suspended particulates, [O.sub.3] = ozone, A = Amsterdam, L = London, M = Milano, P = Paris, and R = Rotterdam.

(*) Cold season mainly October-March; warm season mainly April-September.

([dagger]) per 50-[micro]g/[m.sup.3] increase in pollutant.

(*) Significant at 5% level.

([sections]) Significant at 10% level.

Results, by level of another pollutant. There was an indication that an [SO.sub.2] effect in adults might be observable only at higher levels of particles. Perhaps [SO.sub.2] acts as a proxy for particles instead of having an effect of its own, but in our study the difference was small, inconsistent, and was absent in the elderly. On the other hand, particle effect differences, by [SO.sub.2] levels, were either not detectable or inconsistent, and we conclude that the BS effect found was independent of [SO.sub.2]. There was, however, a significant difference in the effect of BS, by level of [NO.sub.2] on the same day (i.e., it was much stronger or only detectable when [NO.sub.2] levels were high). This effect was stronger in the adults, whereas the BS effect was stronger overall. We expect up to 7% more adult respiratory admissions per 50 [micro]g/[m.sup.3] BS on high [NO.sub.2] days (95% CI = 0, 15). Whether BS effects depend on general local [NO.sub.2] levels, or whether the correlation between BS and [NO.sub.2] depends on a local level, could not be tested because of an insufficient number of data points (Fig. 4). Nitrogen dioxide, however, showed no consistent difference in effect size or showed no effect at all, by different same-day levels of BS or [O.sub.3].



Problems with materials. One problem with data comparability was that Paris and Milano data were confined to all admissions, whether emergency or not. To explore the possible effect of this, we examined the available London data to examine the relationship between the two categories of admission. Some data on diagnosis categories were also available from Milano. The relationship varied with both diagnosis and age, and emergency admissions generally formed the vast majority of respiratory admissions among the elderly, but only approximately 50-70% of the admissions in adults were emergency admissions. From the point of view of the meta-analysis, admission patterns driven by nonemergency admissions were expected to bias the results downward. Results that include Paris and Milano coefficient estimates–especially for respiratory cases of adults–may, therefore, have been rather conservative. It it is possible that some of the differences seen in effect between elderly and adults resulted from this phenomenon. We compared local results, however, and we did not find direct evidence of biased results.

The analysis of particle effects was hampered because different measurement methods were used. The methods used most frequently were the Black Smoke (smoke stain) method and TSP (gravimetric or [Beta]-attenuation). Size-fractionated particulate mass (PMx, particulate mass of particles below x [micro]m in size) was rarely measured in Europe at the time of this study; within this data set, such measurements were available only from Paris (no meta-analysis possible). The average relationship between the different methods is not necessarily the same everywhere and would be affected by the local size distribution, the “blackness” of the particles, and perhaps by season. In the context of this study, results can be compared only qualitatively.

Methodology problems. It should be noted that certain frequently encountered problems of meta-analysis do not apply.[26] There was no selection bias. The participating cities were not selected by the results of the short-term analysis (results were unknown when the studies started), and no cities or results were excluded later. There were no important differences in health endpoints, exposure data, and analysis procedures used; we took great care to set rules for inclusion and to ensure comparability. Therefore, this meta-analysis was not an afterthought, but was planned from the onset of the study.

The interpretation of the by-season or by-level-of-another pollutant results is not completely straightforward. For example, if a threshold exists, and if the values of a pollutant lie mostly below this threshold in one season, then the differences in effect would reflect this threshold, rather than an effect modification by weather conditions. There was no pollutant in this analysis, however, that had no effect in its respective lower season and a strong effect in the other. If one looks at effects by level of another pollutant, differences may also be caused by thresholds if the two pollutants in question are strongly correlated. In this data set the correlation between [NO.sub.2] and BS was positive–but to a different degree in different cities. However, if this were the main explanation for the effect difference observed, a similar pattern would have arisen for [SO.sub.2], which was similarly correlated with particles (data not shown). Inasmuch as this was not the case, there must be another (at least partial) explanation. The dependence of the BS effect on [NO.sub.2] levels might point at a synergy between the adverse effects of various components of automobile exhausts, which produce both particles and [NO.sub.2], but are less responsible for [SO.sub.2] levels in the air. It might point at different effects of particles, by source of emission or composition of the particles. Given that [NO.sub.2] itself shows no effect, high [NO.sub.2] days could also be associated with another component of air pollution that causes this effect modification.

Overview of Results

Ozone. Associations of respiratory admissions with [O.sub.3] were large, significant, homogeneous, and immediate. They were stronger in the elderly than in adults and were stronger with the 8-h daytime average than with the daily maximum. The elderly group was also more affected than the adults in the warm season.

Suspended particles. There was a tendency toward an association of respiratory admissions with Black Smoke, but the very limited number of cities prevented final conclusions. Total suspended particles may show some effect on adults in the warm season. The BS effect appeared quite independent of the concurrent [SO.sub.2] level; it was, however, very dependent on the [NO.sub.2] level, and significantly larger effects were seen when [NO.sub.2] on the same day was above the local median.

Sulfur dioxide. No consistent evidence of an influence on respiratory admissions was found. The heterogeneity between cities was best explained by number of stations providing data (i.e., effects were larger when three or more stations provided data). Perhaps the elderly form a more sensitive subgroup.

Nitrogen dioxide. Although there were some positive associations with respiratory admissions, an [NO.sub.2] effect could not be confirmed for either age group.

Comparison with respiratory mortality in APHEA cities. Cause-specific mortality was available for nine APHEA cities (i.e., London, Paris, Lyon, Barcelona, Milano, Lodz, Poznan, Cracow, and Wroclaw). We saw evidence of an association between respiratory mortality and [SO.sub.2], BS, and [O.sub.3] in Western European cities, with relative risks of 1.05 for [SO.sub.2] daily mean (95% CI = 1.01, 1.04), 1.04 for BS (95% CI = 1.02, 1.07), and 1.05 (95% CI = 1.02, 1.08) for [O.sub.3] daily 8-h averages per 50[micro]g/[m.sup.3] increase in pollution. This association did not hold in the four Polish cities (no [O.sub.3] data available), whose pollution mixtures, with relatively high levels of particles and [SO.sub.2], might be different from those of Western European cities.[27]

Other studies. In most published studies, except those published or co-authored by Schwartz and Dockery, investigators used methods that are not quite comparable with those described here. Comparisons should be made cautiously. In many of the U.S. studies, sulfates were used as a pollutant. It is known that [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]-forms very fine particles; in fact, Schwartz et al.[17] converted them to particulate matter with an aerodynamic diameter less than or equal to 10 [micro]m ([PM.sub.10]) (with locally different factors) for comparison purposes.

Bates and Sizto[3] analyzed the effects of [O.sub.3], [NO.sub.2], [SO.sub.2] (all 1-h maxima), coefficient of haze (COH), and sulfates (daily means) on respiratory admissions during January/February and July/August in Southern Ontario. The method of analysis was very different from the one we used. They found no associations in winter (except with temperature). In summer, respiratory admissions were correlated with [SO.sub.2] (2-d lag), [O.sub.3] (1-2-d lag), [NO.sub.2] (2-d lag), and [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (1-d lag), but they were not correlated with COH. Bates and Sizto observed the largest correlations between [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] and all respiratory causes, followed by [O.sub.3] (1 -d lag).

Thurston et al.[6] investigated summer data (July/ August) gleaned during a 3-y period in Toronto (Canada) w4th respect to particle strong acidity (PSA) (H+), sulfates (daytime 1-h maximum), [O.sub.3] (daily 1-h maximum), TSP, [PM.sub.10], an [PM.sub.2.5] (daily means). Their method of analysis was different from the one we used. Ozone (same day) had the strongest influence on all admissions and particles had a moderate influence, whereas [SO.sub.2] and [NO.sub.2] had none. No comparable relative risks can be given, but qualitatively this is consistent with our findings.

During a 3-y period, Walters et al.[11] examined respiratory admissions in Birmingham (United Kingdom) relative to [SO.sub.2] and BS. Their method of analysis was different from the one we used. Both pollutants had an effect, but BS tended to show up more often in their study than in ours.

Dockery and Pope[9] reviewed and meta-analyzed studies (some of which were hospital-admission studies) for short-term effects of particulate air pollution with respect to daily mean [PM.sub.10]; they converted other particle measurements to this standard, particularly TSP = [PM.sub.10]/0.55 and BS [approximately equals] [PM.sub.10]. They calculated an approximate 4% increase in all respiratory admissions per 50 [micro]g/[m.sup.3] [PM.sub.10] (i.e., three studies). This increase is consistent with the approximate 3% we found for BS in our study.

Burnett et al.[10] investigated data collected during 6 y in Ontario (Canada) with respect to [O.sub.3] and sulfates. Emergency admissions could have been selected from the database, but apparently the authors used all respiratory admissions. The method of analysis was very different from the one we used, especially because there was no meteorology correction. Their parameters can be expressed as approximate risk ratios. The study by Burnett et al.,[10] another by Thurston et al.,[5] and two by Schwartz[7,8] are summarized with the APHEA results in Tables 4-6.

Table 4-International Comparison of Effect of [SO.sub.2] Daily Mean on Respiratory Admissions of Cases 65 y or Older

Study City Lag RR(*) 95% CI

APHEA London 2 1.04 0.99, 1.08

Amsterdam 2 1.02 0.98, 1.06

Rotterdam 0-2(*) 1.02 0.98, 1.07

Paris 0 1.03 1.00, 1.06

Milano 0 1.00 0.97, 1.03

APHEA pooled result — 1.02 1.00,1.05

United States New Haven 2 1.03 1.02, 1.05

(Schwartz 1995[7])

Tacoma 0 1.06 1.01, 1.12

(Schwartz 1995[8])

(*) In Rotterdam, results were pooled from three different study periods.


Table 6.–Effects of Ozone Daily Maximum on Respiratory Admissions During the Warm Season/Summer Only: An International Comparison

Study City Age group (y)

APHEA London 65+




APHEA pooled result 65+

United States New Haven (Schwartz 1995[7]) 65+

Tacoma (Schwartz 1995[7])

Spokane (Schwartz 1995[8])

Buffalo (Thurston 1992[5]) All

New York (Thurston 1992[5]

Ontario (Burnett 1992[10])

Study City Lag

APHEA London 1

Amsterdam 1

Rotterdam 0-2+

Paris 0

APHEA pooled result —

United States New Haven (Schwartz 1995[7]) 2

Tacoma (Schwartz 1995[7]) 2

Spokane (Schwartz 1995[8]) 2

Buffalo (Thurston 1992[5]) —

New York (Thurston 1992[5] —

Ontario (Burnett 1992[10]) —

Study City RR(*)

APHEA London 1.04

Amsterdam 1.05

Rotterdam 1.05

Paris 1.02

APHEA pooled result 1.04

United States New Haven (Schwartz 1995[7]) 1.03

Tacoma (Schwartz 1995[7]) 1.10

Spokane (Schwartz 1995[8]) 1.24

Buffalo (Thurston 1992[5]) 1.06

New York (Thurston 1992[5] 1.03

Ontario (Burnett 1992[10]) 1.02

Study City 95% CI

APHEA London 1.01, 1.06

Amsterdam 0.99, 1.12

Rotterdam 0.97, 1.13

Paris 0.99, 1.06

APHEA pooled result 1.02, 1.05

United States New Haven (Schwartz 1995[7]) 0.99, 1.07

Tacoma (Schwartz 1995[7]) 1.03, 1.15

Spokane (Schwartz 1995[8]) 1.00, 1.54

Buffalo (Thurston 1992[5]) 0.99, 1.12

New York (Thurston 1992[5] 1.02, 1.04

Ontario (Burnett 1992[10]) 1.01, 1.03

In two cities in the United States, [SO.sub.2] effects were similar to those in the APHEA cities. Schwart[27] interpreted them as being caused by the correlation between [SO.sub.2] and particles; however, among the APHEA cities, no association between the locally different [SO.sub.2]-BS or TSP correlations and the locally different [SO.sub.2] effects was found (Table 4).

The effects of [PM.sub.10] on elderly respiratory admissions in the U.S. studies were larger than those found for BS and TSP in the APHEA cities. Perhaps this difference occurred because [PM.sub.10] is a more appropriate measure of the fraction of particles relevant for health effects. Also, the correlations between particulate mass and other pollutants tend to be different in the United States (Table 5).

With respect to [O.sub.3], only summer or warm season results ([O.sub.3] measurements are often discontinued during the cold season) are quoted in the literature, as are effects for daily maximum or daily mean; we found the 8-h daytime average to be the best predictor. Except for the very large effect found in Spokane (but with a 95% CI = 0, 54), the European and U.S. results for warm season and daily maximum appear quite similar in magnitude (i.e., all CIs overlap largely with the 2-5% effect size CI found for 50 pg/ml daily 1-h maximum).

A review of the literature concerning, specifically, COPD admissions and asthma admissions is provided elsewhere.[14-16]

Comments and interpretation. In considering whether the associations observed are causal, one must examine the possibility of confounding by factors that could be associated with both pollution and health effects. Differences in diagnostic habits, treatment regimes, and health-care systems–as well as lifestyles–are unlikely confounders inasmuch as they may vary strongly between cities and countries, but not according to daily local pollution levels. More plausible confounders are weather and climate. In this study, the case for causality was strengthened by our finding of such consistent [O.sub.3] effects across the European cities, as well as in U.S. studies, in which different climates and weather patterns were noted. Similarly strong and consistent effects were found in APHEA in the subgroup of COPD admissions, for which additional data were available from Barcelona, which has a Mediterranean climate quite unlike that of London or the Netherlands. 14 Differences in actual effect size might be the result of differences in the pollution mix, differences in the spectrum of diseases admitted to hospital, or differences in the underlying susceptibility of people with those diseases to admission, based on national differences in primary care systems that affect the way exacerbations are handled. it must be noted, however, that from a statistical point of view, the strong [O.sub.3] effects were homogenous between cities.

The respiratory group mainly comprises infections of the lung or obstructive airways disease in either acute form (asthma) or chronic form (COPD); diagnoses are difficult to differentiate, especially in the elderly. A common exacerbating factor for all respiratory conditions, which may in turn be exacerbated or promoted by air pollution, is acute infection. Specifically, we may see an impairment of airway defenses against infections, an increase in airways hyperresponsiveness, toxic inflammation of the lung, modification of the response of asthmatics to inhaled allergens, airways obstruction, and impairment of gas exchange and ventilation/perfusion balance. Plausible mechanisms exist for respiratory disease to be affected by all four pollutants.[2]

The coherence of results across various cities and studies, especially for [O.sub.3], together with what is known about possible mechanisms, strengthens the argument that the associations found in this study were causal.

The APHEA project was supported by the European Commission, DGXII, Environment 1991-1994 Programme (Contract number EV5V CT92-0202; scientist responsible, Dr. C. Nolan).

The APHEA collaborative group contains the following members: K. Katsouyanni, G. Touloumi, E. Samoli (Athens, Greece–Coordinating Center); H. E. Wichmann, C. Spix (OberschleiBheim, Germany); H. R. Anderson, A. Ponce de Leon, R. Atkinson, J. Bower, D. Strachan, M. Bland (London, United Kingdom); M. A. Vigotti, G. Rossi, L. Bisanti, F. Repetto, A. Zanobetti (Pisa, Italy); W. Dab, P. Quenel, S. Medina, A. LeTertre, B. Thelot, B. Festy, Y. LeMoullec, C. Monteil (Paris, France); J. P. Schouten, J. M. Vonk, A. C. M. deGraaf (Groningen, The Netherlands); D. Zmirou, P. Ritter, I Barumandzadeh, F. Balducci, G. Laham (Lyon, France); B. Wojtyniak, T. Piekarski (Warsaw, Poland); L. Bacharova, F. Fandakova (Bratislava, Slovakia); J. Sunyer, J. Castellsague, M. Saez, A. Tobias (Barcelona, Spain); and A. Ponka (Helsinki, Finland).

Submitted for publication October 18, 1996; revised; accepted for publication March 31, 1997.

Requests for reprints should be sent to Dr. C. Spix, Uni Mainz, IMSD, D55101 Mainz, Germany.


[1.] Burney PG. The effect of death certification practice on recorded national asthma mortality rates. Revue D’Epidemiol et De Sante 22. Publique 1989; 37:385-89.

[2.] American Thoracic Society. Health effects of outdoor air pollution. Am J Respir Crit Care Med 1996; 153:3-50.

[3.] Bates CV, Sizto R. Air pollution and hospital admissions in Southern Ontario: the acid summer haze effect. Environ Res 1987; 43:317-31.

[4.] Pope CA III. Respiratory hospital admissions associated with [PM.sub.10] pollution in Utah, Salt Lake, and Cache valleys. Arch Environ Health 1991; 46:90-97.

[5.] Thurston GD, Ito K, Kinney PL, et al. A multi-year study of air pollution and respiratory hospital admissions in three NY State metropolitan areas: results for 1988 and 1989 summers. J Expo Anal Environ Epidemiol 1992; 2:429-50.

[6.] Thurston GD, Ito K, Lippmann M, et al. Respiratory hospital admissions and summertime haze air pollution in Toronto, Ontario: consideration of the role of acid aerosols. Environ Res 1994; 65:272-90.

[7.] Schwartz J. Short-term fluctuations in air pollution and hospital admissions of the elderly for respiratory disease. Thorax 1995; 50:531-38.

[8.] Schwartz J. Air pollution and hospital admissions for respiratory disease. Epidemiology 1996; 7:20-28.

[9.] Dockery DW, Pope CA III. Acute respiratory effects of particle air pollution. Ann Rev Public Health 1994; 15:107-32.

[10.] Burnett R, Dales RE, Eaizenne ME, et al. Effects of low ambient levels of ozone and sulfates on the frequency of respiratory admissions to Ontario hospitals. Environ Res 1994; 65:172-94.

[11.] Walters S, Griffiths RK, Ayres J. Temporal association between hospital admissions for asthma in Birmingham and ambient levels of sulfur dioxide and smoke. Thorax 1994; 49:133-40.

[12.] Katsouyanni K, Zmirou D, Spix C, et al. Short-term effects of Air Pollution on Health: A European approach using epidemiologic time series data. The APHEA project: background, objectives, design. Europ Respir J 1995; 8:1030-38.

[13.] Katsouyanni K, Schwartz J, Spix C, et al. Short-term effects of Air Pollution on Health: A European approach using epidemiologic time series data. The APHEA protocol. J Epidemiol Commun Health 1996; 50(suppl 1):12-18.

[14.] Anderson HR, Spix C, Medina S, et al. Air pollution and daily hospital admissions for chronic obstructive pulmonary disease (COPD) in 6 European cities: results from the APHEA project. Europ Respir J 1997; 10:1064-71.

[15.] Sunyer J, Castellsague J, Saez M, et al. Air pollution and mortality in Barcelona. J Epidemiol Commun Health 1996; 50(suppl 1): 76-80.

[16.] Sunyer J, Spix C, Quenel PH, et al. Urban air pollution in and emergency admissions for asthma in four European cities. The APHEA project. Thorax 1997; 52:760-65.

[17.] Schwartz J, Spix C, Touloumi G, et al. Methodological issues in the studies of air pollution and daily counts of deaths or hospital admissions. J Epidemiol Commun Health 1996; 50(suppl 1):3-11.

[18.] Schouten JP, Vonk JM, Graaf A de. Short-term effects of air pollution on emergency hospital admissions for respiratory disease: results of the APHEA project in two major cities in the Netherlands during 197789. J Epidemiol Commun Health 1996; 50(suppl 1):22-29.

[19.] Zmirou D, Schwartz J, Saez M, et al. Time-series analysis of air pollution and cause-specific mortality: a quantitative summary in Europe (APHEA study). (Submitted for publication.)

[20.] Wojtyniak B, Piekarski T. Short-term effects of air pollution on mortality in Polish urban population-what is different? J Epidemiol Commun Health 1996; 50(suppl 1):36-41.

[21.] Touloumi G, Samoli E, Katsouyanni K. Daily mortality and “winter-type” air pollution in Athens, Greece: a time-series analysis within the APHEA project. J Epidemiol Commun Health 1996; 50(suppl 1):47-51.

[22.] Dab W, Medina S, Quenel P, et al. Short-term respiratory health effects of ambient air pollution: results of the APHEA project in Paris. J Epidemiol Commun Health 1996; 50(suppl 1):42-46.

[23.] Spix C, Wichmann HE, Daily mortality and air pollutants: findings from Koln, Germany. J Epidemiol Commun Health 1996; 50(suppl 1):52-58.

[24.] Ponce de Leon A, Anderson HR, Bland JM, et al. Effects of air pollution on daily hospital admissions for respiratory disease in London between 1987-88 and 1991-92. J Epidemiol Commun Health 1996; 50(suppl 1):63-70.

[25.] Berkey CS, Laird NM. Nonlinear growth curve analysis: estimating the population parameters. Ann Human Biol 1986; 13:111-28.

[26.] DerSimonian R, Laird N. Meta analysis in clinical trials. Controlled Clin Trial 1986; 7:177-88.

[27.] Zmirou D, Brumandzadeh T, Balducci F, et al. Short-term effects of air pollution on mortality in the city of Lyon, France, 1985-90. J Epidemiol Commun Health 1996; 50(suppl 1):30-35.


[chi square] Test of Homogeneity

Given local estimates [[Beta].sub.i] (parameter or parameter vector) and [[Sigma].sub.i] (variance or covariance matrix of [[Beta].sub.i]), i = 1, . . . N, we obtain a fixed effects model estimate of the joint parameter or parameter vector


with weights or weight matrices


The test statistic


is [chi square]-distributed with (N – 1) p degrees of freedom, where N is the number of studies and p the dimension of [Beta].

CLAUDIA SPIX GSF Forschungszentrum fur Umwelt und Gesundheit Institut fur Epidemiologie Neuherberg, Germany H. ROSS ANDERSON Department of Public Health Sciences St. George’s Hospital Medical School London, United Kingdom JOEL SCHWARTZ Harvard School of Public Health Boston, Massachusetts MARIA ANGELA VIGOTTI Institute of Clinical Physiology National Research Council Pisa, Italy ALAIN LeTERTRE Observatoire Regional de la Sante Paris, France JUDITH M. VONK State University of Groningen Faculty of Medicine Department of Epidemiology and Statistics Groningen, The Netherlands GIOTA TOULOUMI University Athens Medical School Department of Hygiene and Epidemiology Athens, Greece FRANCK BALDUCCI Universite Joseph Fourier Medical School Lyon, France TOMASZ PIEKARSKI National Institute of Hygiene Department of Medical Statistics Warsaw, Poland LJUBA BACHAROVA National Center for Health Promotion Bratislava, Slovakia AURELIO TOBIAS Institut Municipal d’Investigacio Medica Barcelona, Spain ANTTI PONKA Helsinki City Center of the Environment Helsinki, Finland KLEA KATSOUYANNI University Athens Medical School Department of Hygiene and Epidemiology Athens, Greece

COPYRIGHT 1998 Heldref Publications

COPYRIGHT 2000 Gale Group