Statistics from Altmetric.com
It is widely believed that current levels of outdoor air pollution may provoke or exacerbate asthma. This is not surprising because a characteristic aspect of asthma is a tendency towards hyperreactivity to inhaled environmental gases, particles, and allergens. Experimental studies in animals and humans have shown that commonly measured pollutants including ozone (O3), sulphur dioxide (SO2), respirable particles, and nitrogen dioxide (NO2) all have the potential to aggravate asthma by either direct irritation or by enhancing the effects of allergens to which the individual is sensitive.1-4 An effect of air pollution on symptoms or use of medications has been reported mainly from areas with high levels of oxidant pollution5-7 and some studies have only observed small changes in lung function, similar to those experienced by non-asthmatic subjects.2 It is plausible that such effects, if occurring in an asthmatic with already compromised lung function, could lead to an increase in use of primary care or hospital services. Exposure to aeroallergens (pollens, fungal spores) is related to weather conditions and is a potential confounder in analyses of the effects of air pollution on daily hospital admissions for asthma. Experimental evidence suggests that the effects of aeroallergens may be increased by exposure to air pollution8-10 but the importance of this biological interaction at the population level is unknown.
Analyses of daily time series of emergency hospital attendance or admissions for asthma offer a convenient method of testing the hypothesis that air pollution provokes asthma in the population at large. Taken overall, existing studies lack consistency as to the presence of effects or, where effects have been observed, the type of pollutant.2 Few studies have addressed the issue of confounding or effect modification by pollens or examined associations in children, a subject of particular popular concern.
In this paper we attempt to address all of these questions in the context of London from 1987 to 1992. We use Poisson regression techniques to examine the association between air pollution (O3, NO2, SO2, and black smoke) and daily emergency admissions for asthma in children, adults (15–64) and the elderly (65+). We examine the possible confounding effect of airborne pollens and test the hypothesis raised by chamber studies8-10 that there is an interaction between pollens and pollution effects. This work was part of a multi city European project (Air Pollution and Health: a European Approach (APHEA)) which used a standardised approach to the data assembly and statistical analysis.11-13 A meta-analysis of selected asthma admissions results from APHEA has been published14 and elsewhere15 we have reported the results for all respiratory admissions, but not specifically for asthma. The published results for all respiratory diagnoses15 do not address possible confounding or effect modification by aeroallergens.
The sources of health and air pollution data and the way in which the database was constructed for analysis have been described in detail elsewhere.15 This followed the procedures adopted by APHEA11 All air pollution variables were obtained from routine monitors sited to measure urban background concentrations. Ozone concentrations were measured by the ultraviolet absorption technique at a single background monitor in inner London and expressed as eight hour (09.00–17.00 hours) and maximum one hour averages. Nitrogen dioxide concentrations were measured by the chemiluminescence method at two urban background sites in inner London and expressed as one hour maximum and daily averages. Daily average concentrations of particles were measured using the black smoke (BS) method at four monitors in central, north, northeast, and south London; this method measures only black particles of less than 4.5 μm diameter.16 Daily average concentrations of SO2 were obtained from the same four sites using the acidimetric bubbler system. If up to two stations did not provide data we used a regression method to obtain an estimate of the average values across the four stations on each day.17 Mean 24 hour temperature and humidity were calculated from data obtained from Holborn in central London.
The airborne pollen concentrations were monitored on the flat unobstructed roof of a seven storey building in North London using a Burkard volumetric spore trap. This draws air through a critical orifice at 10 l/min onto a tape on a rotating drum (speed 2 mm/h) coated with a smooth adhesive mixture of Vaseline and paraffin wax to provide a time related sample of airborne material. The tape is stained with basic fuschin and examined under ×400 magnification. Pollen grains and spores are identified with as much taxonomic detail as possible and a 24 hour average for each taxon is compiled from 12 two hour samples. This technique of pollen and spore monitoring is the standard adopted by the British Aerobiology Federation for use by the National Pollen Network. Daily counts for birch, grass, and oak are used in this analysis.
For the period April 1987 to February 1992 inclusive, counts of daily emergency admissions for asthma (ICD9 493 coded at discharge) to all National Health Service hospitals in London were obtained from the Hospital Episode System. This covers the great majority of emergency admissions for acute medical conditions such as asthma. It does not include attendance at hospital emergency departments which do not result in admission.
The statistical analysis followed the approach developed by APHEA which has been described elsewhere.12 Details of the analysis of the London admissions data for this period were described in our previous paper on all respiratory admissions15) but are briefly summarised here. The general approach was to use Poisson regression to model the daily counts of asthma admissions, taking account of overdispersion and autocorrelation. We first controlled for time trends, seasonal cycles, days of the week, public holidays, and influenza epidemics. The residuals from this process were then examined in relation to temperature and humidity to identify the lag (if any) with the strongest effect and the transformation (if any) which gave the best fit. The last stage in the development of the basic model was to correct for the remaining autocorrelation of the residuals. The final model was checked and revised as necessary using autocorrelation functions, Durbin-Watson statistics, plots of fitted values and residuals, and periodograms. A log linear model with autoregressive Poisson errors was then fitted to the data. No transformation of the pollutant variable was used.
Our a priori hypothesis concerned associations with pollution on the same or previous two days. The most significant single lag was selected whether the effect was positive or negative. We also examined the effect of cumulative lags up to three days. The effects of each air pollutant were examined for the whole year and for the “warm” (April to September) and “cool” (October to March) seasons separately. We analysed the effects of pollens in the warm season by either including all warm season days or only those days for which the pollen count was greater than zero. In examining the interaction between pollens and air pollutants we postulated a concurrent rather than sequential effect so that, in these models, the best respective lag for both pollen and pollutant was used. Where a pollutant was found to have a significant effect the effect of putting a second pollutant into the model was examined.
The results are presented as a percentage increase associated with a 10 unit increase in pollutant level. This was calculated by the formula 100 × (1—exp (β coefficient × 10)—1). A 1% relative increase corresponds to a relative risk of 1.01. A significance level of 5% (two sided) is used.
The descriptive statistics of the admissions and environmental data are shown in table 1. The associations between the various air pollutants and asthma admissions are shown in table 2 by age and season. For the all-ages group over the whole year, significant associations were observed with NO2, SO2 and black smoke, with the cumulative lags tending to show stronger and more significant effects. Some of these associations have p values of <0.001 and provide good evidence for a relationship even after the Bonferroni correction for 24 tests (eight pollutants and three seasons).
Ozone was negatively associated with admissions in the 0–14 age group (whole year and cool season) and positively with admissions in the 15–64 age group (whole year and warm season). The seasonal differences in the effects of ozone in these two age groups were statistically significant. NO2 was significantly associated with admissions in the 0–14 age group (whole year and warm season) and in the 65+ age group (cool season). SO2 was associated with admissions in the 0–14 age group (whole year and warm season) and in the 65+ age group (cool season), while black smoke was associated with admissions in the 65+ age group (whole year and cool season).
The various pollutants covary to some extent and this covariation is not constant across the seasons. For example, the correlation between ozone and NO2 is positive in the warm season and negative in the cool season. For every significant association, two-pollutant models were examined to obtain insight into which single pollutant might influence asthma admissions independently of the effects of others. The results of this analysis are shown in table 3. The results did not follow a pattern that is easy to summarise. We draw attention to those analyses in which the effect of a particular pollutant remained significant after each of the other three pollutants was included in the model.
In the all-year analysis the effects of O3 (15–64 age group) and NO2 (all-ages) were robust to inclusion of each of the other three pollutants. In the warm season analysis similarly robust findings were obtained for SO2 (0–14 age group) and ozone (15–64 age group and all-ages). In the cool season the only robust result was for the negative effect of ozone in the all-ages analysis.
The effect of birch, grass, and oak pollens on admissions is shown in table 4. No consistent pattern is apparent. There is a significant positive effect of birch pollen in the 0–14 and 15–64 age groups, a significant negative effect of grass pollen in the 0–14 age group, and a significant negative effect of oak pollen in the 0–14 and 15–64 age groups. Because the production and dispersal of pollens is related to weather conditions, it is possible that any effects of pollens on asthma admissions could confound the effects of pollutants in the relevant period of the year. There was little correlation between birch pollen and any pollutant but grass pollen was correlated positively with ozone (r = 0.122) and SO2 (r= 0.115) and negatively with black smoke (r = –0.117). Oak pollen was positively related to ozone (r = 0.138) and NO2 (r = 0.182).
To examine the possibility of confounding by pollens the effects of those pollutants found to have warm season effects were examined after inclusion of each of the pollen categories in the model. As table 5clearly shows, the pollutant effects are not influenced by introducing pollen into the model. The analysis was repeated for only those days on which the pollen count was above zero with the same result. Table 6tests the hypothesis that there is a pollution/pollen interaction by comparing the effects of pollutants on days when pollens were present with those when they were not. The only interaction in a positive direction—that is, as hypothesised—was between SO2 and grass pollen in the 0–14 age group (p <0.01). In two comparisons (ozone and birch pollen and ozone and oak pollen in the all-age group) the interaction was significantly negative (p <0.001 in each case).
This is one of the most comprehensive investigations of daily asthma admissions and air pollution yet reported, and one of the few to take airborne pollen into account and to examine interactions between pollution and pollen levels. We found associations between all of the four air pollutants and daily asthma admissions, but a lack of consistency across age groups and seasons. Some of the associations were not strongly significant and others were weakened or became non-significant after including another pollutant in the model. For all ages combined the most robust associations were with SO2and NO2. When analysed by age the most robust associations were for SO2 in the warm season in children, and for ozone in younger adults, also in the warm season. With the exception of ozone, there was a tendency for the cumulative lags to have a stronger effect.
Although the statistical method employed was a widely accepted one,13 its appropriateness and the manner in which potential confounders, especially weather and cyclical factors, are controlled have both been criticised.18 Studies which have addressed these issues in the context of daily mortality have been reassuring19 but may be less relevant to admission data which may require more complex models than mortality data. This applies especially to children because of their greater susceptibility to respiratory epidemics and the effects of school holidays. In addition, it must also be recognised that large numbers of statistical tests were performed using a variety of lags, and it is therefore also possible that some of the significant results were due to chance. For this reason more weight should be placed on the larger and more strongly significant findings, especially those which are consistent across seasons and age groups and robust to the inclusion of other pollutants in the model.
Our assessment of exposure was based on a relatively small number of monitors and may not have been a good indicator of personal exposure to outdoor pollution. Where several monitors were available, there was a significant and often strong correlation between them. Since this study the number of automatic monitors in London has expanded greatly. These new monitors, which cover both central and suburban London, tend to show good tracking between background levels of ozone, NO2and PM10 particles, though less so for SO2.
Twenty one other studies of air pollution and daily asthma admissions were identified in the literature. Of these, 15 appeared to have a satisfactory methodology and these, together with the present study, are listed in table 7 to obtain an overview of age groups and pollutants studied and the direction and statistical significance of the associations reported. Calculation of summary estimates using meta-analysis was outside the scope of this paper and would, in any case, have been difficult because the studies differed in many respects, including methods, statistical power, the pollutants investigated, and the age ranges reported. Our assessment of consistency will rely mainly on whether statistically significant findings were reported. In the all-age group three of 14 studies did not find significant associations with any of the pollutants assessed and the proportions with significant findings for ozone, SO2, NO2 and particles were 7/14, 6/12, 2/9, and 7/15, respectively. For the adult age group (15–64) three out of 13 did not find significant associations and the proportions with significant associations with ozone, SO2, NO2and particles were 6/13, 1/12, 3/13, and 3/13, respectively. One of the ozone associations was negative. Similarly, in children, 4/13 did not find significant associations and the proportions with significant associations with ozone, SO2, NO2 and particles were 6/11, 5/12, 3/13, and 2/11, respectively. Two of the associations with ozone were negative (both in the winter season). Taken overall, it is apparent that the evidence is not coherent as to whether there is an effect of pollution or the responsible pollutant. Ozone, SO2 and particles were significant in no more than half of all studies and only about a quarter found significant effects for NO2.
Some of this inconsistency may result from false negative results arising from lack of statistical power or from false positive results due to chance, multiple significance testing, post hoc hypothesis testing, or publication bias. It could also result from differences between cities in the pollutant levels and prevailing mixtures. Differences between the age-specific results may also reflect differences in the clinical and pathological nature of asthma and its provoking factors at different stages of life. Many of the pollutants considered were not independent of one another and may have been indicating some other causal pollutant which was not measured. For example, summer pollution comprises a complex mixture and varying relationships of gaseous pollutants such as ozone and SO2and secondary particles such as sulphates, nitrates and associated acids. Significant associations between hospital admissions for asthma and individual pollutants should not therefore be interpreted narrowly as necessarily being an effect of that pollutant. Conversely, the lack of strong associations with black smoke does not exclude an effect of particles such as sulphates not measured by this technique.
Debate about the causal nature of associations between daily mortality and particle pollution has been dominated by issues of methodology, potential confounding, and biological plausibility. A major argument in favour of causality has been the consistency of results obtained from a wide variety of cities throughout the world. In the case of asthma admissions lack of consistency weakens the argument for causality. Perhaps the effects of outdoor air pollution on asthma admissions are more dependent on local environments and population vulnerability than is the case for particles and mortality. If this is the case, results from one area should not be assumed to apply to another.
Hospital admissions for asthma may also be influenced by ambient aeroallergens. Extreme examples are point source epidemics such as those associated with the unloading of soy beans36 and with rye grass pollen in association with thunderstorms.37The rise in admissions in the spring has been attributed to pollens.38 A number of studies have not found associations between pollens and daily admissions or emergency room attendances for asthma.39-41 One explanation could be that pollens are generally too large to penetrate to the small airways. On the other hand, fungal spores which are smaller (and hence more respirable) have been associated with asthma admissions.41 42 In the present study we are confident that the observed associations between admissions and air pollutants are most unlikely to have been confounded by pollens. It also seems unlikely that there is an interaction between pollens and air pollutants. Further work with more recent data will investigate whether the same is true for airborne fungal spores.
Chamber studies have found evidence of interactions between aeroallergens and gaseous ambient pollutants in their effects on bronchial hyperresponsiveness.8-10 43 We found little evidence to suggest that this mechanism occurs in the ambient situation, apart from the interaction between SO2 and grass pollen in children’s asthma admissions which requires confirmation in further studies. The chamber study of Devalia et al 43 found evidence of an interaction between pollens and a mixture of NO2 and SO2 but we did not address the question of interactions with mixtures in this data set.
We conclude that there is evidence that all of the pollutants may have an effect on asthma admissions but that there is a lack of consistency across the age groups in the specific pollutant responsible. Associations with pollutants were not confounded by airborne pollens and there was little evidence for an interaction with pollens. These results, when taken together with 15 published studies of air pollution and daily asthma admissions, indicate that the evidence linking air pollution to asthma admissions lacks consistency.
This study was funded by the National Asthma Campaign (Grant 178) and the European Commission, DGXII, Environment 1991–94 Programme (EV5V CT92–0202).
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.