Оverview and critique of the air pollution and health: a European approach (APHEA) project

Report no. 99/54: The relationship between air pollution and adverse health outcomes is a subject of current concern in the area of environmental health. The Air Pollution and Health: a European Approach (APHEA) project is a co-ordinated study of the short-term effects of air pollution on mortality and hospital admissions in sixteen cities in eleven European countries. The Institute for Environment and Health have been requested by CONCAWE to carry out an independent review and critique of the APHEA project.

Twenty-three published papers were identified and reviewed. Methodology A standardised protocol was developed and adopted by all centres in the APHEA project for the collection of air pollutant data, health outcome data and information on potential confounders, such as temperature, relative humidity and unusual events. Poisson regression, allowing for auto correlation and over dispersion was used by each centre to analyse their own data and the procedure for the building statistical models was specified in detail. Each city fitted the model of ‘best fit’ to their data, so that the city-specific models differed in the number of variables they included and the transformations and lag times used for the time-dependant variables. The meta-analyses combined the relative risks obtained from the best fitting models for the individual cities to give overall summary estimates.

A fixed effects model was first fitted and a test of heterogeneity was carried out. If heterogeneity was significant, a random effects model was then fitted, with explanatory variables. Results:

• Small increases in the relative risks were consistently found in the individual;
• Cities and in the combined analyses for several air pollutants and health;
• Outcomes;
• The relative risks for total mortality ranged from 0.98 to 1.13, most being between 1.01 and 1.03;
• The strongest effect for total mortality was found for sulphur dioxide (SO₂), with particulates showing a slightly weaker association;
• Consistently elevated relative risks were found for both nitrogen dioxide (NO₂) and ozone (O₃) and the estimates from the combined analyses were statistically significant;
• The results for mortality attributable to respiratory disease and cardiovascular disease were less consistent with a wider variation in risk estimates between cities and few being significantly elevated;
• Similar patterns were found for digestive system diseases, chosen as a control group, although risk estimates tended to be lower;
• Increased risk of hospital admissions for respiratory disease was found for SO₂ in the elderly (aged 65 and over), and for black smoke (BS) in the 15-64 age group;
• Results for NO₂ and respiratory disease admissions were inconsistent;
• A significantly positive association was found for respiratory admissions in London with O₃, and the combined estimate was significantly raised and was larger in the elderly;
• A combined analysis from four cities for emergency hospital admissions for asthma found associations with NO₂ in both children under the age of 15 years and adults aged from 15-64 years, and for SO₂ in children that were significantly raised;
• Emergency admissions for chronic obstructive pulmonary disease were significantly associated with all pollutants using the combined estimates, with the strongest and most consistent effect being found for O₃;
• The effects of temperature and season varied between cities and by pollutant and health outcome. There was a tendency for effects to be stronger in the summer;
• Correlations between pollutants within cities and the extent to which synergy between pollutants existed also varied considerably.

Critique The systematic and co-ordinated approach is one of the strengths of the APHEA project, and the results have contributed to the previous lack of meaningful information in Europe. The results support those found in many other studies of the adverse effects of air pollution. There was a tendency, however, for there to be a stronger association with SO₂, than studies, for example from the USA, and risk estimates for particles were generally lower than those obtained in the American studies. However, the project was limited by the quality of some of the data, in particular the exposure information. There were variations in the numbers of monitors per geographical area and per population, the location of the monitors, the measurement and sampling techniques, and the correlations between values from the monitors. Ecological studies such as this are subject to a high degree of misclassification due to the use of ambient air exposure data as a surrogate for personal exposure. There were also problems of completeness of diagnosis for some of the hospital admission data, and it was not possible to separate emergency from planned admissions in some centres. The published papers do not report the results in a standardized format. Information important for the interpretation and comparison of the results between the cities was not consistently presented.

For example, it would have been useful to know, for all the cities, the values of the correlations between pollutants measured in order to assess potential collinearity, the risk estimates for all the covariates included in the models, changes in the values of the coefficients of the pollutants as variables were added to or removed from models, and an indication of the fit of the models. The correct adjustment for potentially strong co-founders, such as those relating to climatic conditions, is particularly important when elevation of the relative risk estimates is so small. Variables, such as population size, age structure, migration patterns and proportion of smokers, which might have undergone change over the study periods and influenced either health or pollution or both, were described for some cities but were not considered as variables for inclusion in the models. Other time-varying factors, such as pollen, which might influence respiratory diseases such as asthma, were not included in the analyses.

The lack of homogeneity in some of the data and the use of risk estimates derived from the ‘best fit’ models, with varying choice of variables, lag times and transformations, necessitates caution when evaluating the summary estimates obtained from the meta-analyses. Recommendations Although the results from the APHEA project concur with many other studies in showing an association between a range of airborne pollutants and adverse health outcomes, they have also helped to emphasise the wide variation in both the magnitude and direction of the estimates and the predominant pollutants. Clarification of the relationship between levels of air pollution measured by outdoor fixed-site monitoring, levels indoor and personal exposures is required. The role of each pollutant, both as an individual risk factor and in its contribution to a synergistic effect, the importance of different sources and mixtures of pollution, the interrelationships of pollution and climate and the influence of micro environments are all areas which require further research. The development of a European air pollution monitoring network, which uses the same method of sampling and analysis and is more evenly distributed by area and population is also essential.

In view of the lack of consistency of reporting of the results from the APHEA project, a paper summarising these in a standard format, as suggested earlier, for all cities, pollutants and health outcomes, would be useful. The result for diseases of the digestive system should also be given for all cities to assist comparisons. Although there are limitations in the APHEA data, reanalyses of the combined data sets using the raw data from all the relevant centres, would be helpful. Lag times, transformations and weights could be fitted, and explanatory variables and interaction terms characterising the individual centres could also be included. Ecological study designs are limited in their inherent assumption that individual risk can be estimated using group data. Spurious associations may be identified through an unobserved variable or by incomplete adjustment for confounding. Collinearity between pollutants may make it difficult to identify exactly which pollutant is important and there may be incomplete knowledge about the time-exposure response relationship. The regression coefficients obtained from the models used cannot be used to quantify the effect of pollution. A Cohort study approach is needed to assess causality and also estimate the potential public health benefit of are duction in pollution.