The public health relevance of air pollution abatement

Causality

The assessment of the impact of air pollution inherently assumes that air pollution causes morbidity and mortality. As discussed in the WHO working group report 21, “causality” is not a clear-cut definition but a process, evaluating the entire scientific evidence. It includes a comprehensive identification of all relevant studies and a systematic assessment of their quality. Conclusions and interpretations must be based on the entire body of valid research. So far, no formal process has been implemented on the general issue of air pollution, although processes such as the development of WHO quality guidelines for Europe have implicitly or partially followed this path 23. For single pollutants, such as particulate matter or ozone, the USA Environmental Protection Agency's approach of regularly reviewing the literature on criteria pollutants can be considered such a systematic review process. The consideration of formal criteria may be a useful guide to assess the degree of evidence 24. Whereas some scientists claimed evidence for causality years ago 25, research has since been intensified in many areas of the world. The attempt to jeopardise science with politics or to perpetuate misconceptions about epidemiology 26 has failed, and it is of note that scientists made no progress in dismantling the claim for evidence of a causal association between current levels of air pollution and morbidity and mortality in the public 3, 27–29. Over recent years, the evidence for causality, particularly regarding short-term effects, has become stronger 30.

Although most air pollution experts may claim “causality”, the level of evidence may vary for different health outcomes. Thus, impact assessors may need to choose a restricted set of outcomes to derive attributable cases. Depending on where to draw this line, different projects may include different outcomes. In general, much less research has been undertaken to assess the long-term effects of air pollution. Thus, whereas most impact studies may include hospital admissions, asthma attacks or acute respiratory symptoms, the long-term effect on death rates or years of life lost and chronic bronchitis remain more controversial. Studies, such as the USA assessment by Ostro and Chestnut 31, the trinational study 15, or the Italian air pollution project (ITARIA) 32, included chronic bronchitis among adults, based on the Californian Seventh Day Adventist study 33. However, it is not yet clear whether these air pollution attributable cases may be of the same average severity as chronic bronchitis cases among smokers. Although this uncertainty may be of limited importance in the derivation of attributable cases, in a cost assessment, severity might be of relevance.

Cost of effects

In general, it is true that the process of cost attribution to the number of cases adds a large level of uncertainty 9, 10, 34. This is only partly due to the lack of data in the field of health economics but is much driven by the wide margin of cost estimates for mortality. Depending on methods and concepts, the cost of one death may vary by a factor of >20. For example, whereas in USA studies a death may be valued up to $10 million 20, 35, the trinational study used a value of E0.9 million (E1.0 is equivalent to approximately USA $0.9; table 2⇓) 9, the UK Department of Transport a value of E1.2 million 11, and the first Swiss assessment a value of less than E400,000 34. This introduces a margin of quantitative uncertainty which goes beyond any other aspect of uncertainty. Accordingly, controversies are particularly extensive for economic assessments. Given that economic assessments are strongly driven by national preferences and healthcare systems, cost estimates are usually not transferable. The European trinational study attributed some E600–800 per capita and year to air pollution-attributed health problems with 75% due to mortality estimates 9. The cost values (per case) are indicated in table 2⇓.

The risk function

As shown in figure 1⇓, the slope of the risk function is influential. In air pollution research, this function is usually based on epidemiological studies. The parameter is often called “exposure/response” or “dose/response” although epidemiological studies may measure neither “exposure” nor “dose” in their true sense directly. As in the case of meta-analyses of clinical trials, pooled estimates from several studies may be considered an adequate choice. However, some assessors may prefer to use, if available, local effect functions if the uncertainties in these studies are considered to be smaller than those of transferring results from other places. For most outcomes and regions, estimates may not be available. Accordingly, transferring risk functions from a study population to the target population of the impact assessment may be required. This step carries inherent uncertainties if exposure levels and/or quality of exposure grossly differ in the two populations. Although theoretically of concern, it is of note that the currently available risk functions for outcomes such as hospital admissions or symptoms are rather similar across studies from different regions, i.e. heterogeneity in risk functions may not be the most influential factor.

Mortality: how should the impact of air pollution on death be interpreted?

From a methodological perspective, mortality is a rather unusual and difficult outcome. Due to the high public attention on this outcome and the strong influence of death on cost assessments (table 2⇓), it remains an important and controversial part of the impact assessment. Two concepts of “death” must be distinguished. On one side, death is an event (count data); on the other side it is the delimiter of survival time or life expectancy. Time-series studies are extremely efficient and powerful means with which to investigate the association of the daily number of death counts and the level of air pollution on that day or a few days before. Highly precise estimates of the association between today's level of air pollution and mortality rates in the next few days are available 3. In fact, even on a global scale, these estimates are rather consistent 4. However, these time-series studies have a restricted short-term effect perspective which must be of concern for health-impact assessors. For example, let us assume cases where air pollution may trigger respiratory exacerbation 36 or a myocardial infarction 37. It may well be that some of these cases will survive the very first days but will die within a few weeks, due to this primary, air pollution-related event. Time-series mortality studies will not attribute such “late-onset death” because the vast majority of these studies investigate only the associations between air pollution and death which occur within a few days. In fact, Schwartz 38 has shown that the typical time series do not fully capture the short-term effects of air pollution. Furthermore, the time-series approach does not assess life time lost nor does it account for shortening in life expectancy due to long-term cumulative exposure. For example, air pollution may lead to recurrent infections which, in the long run, may result in chronic respiratory symptoms 6, 33. These conditions are associated with shorter life expectancy 39. Although the actual “event” of dying may not be related to the level of air pollution at death or the few days before death, such premature death should clearly be part of the air pollution health-impact assessment, otherwise the attributable cases will be, by default, underestimated. Cohort studies, however, directly assess the impact of cumulative air pollution experience on person-time or time-to-death. Although some very short-term, high-exposure victims of air pollution may not be captured in the cohort approach, this design clearly gives a more complete analysis of the problem. Therefore, based on conceptual considerations, the impact assessment may rely on concentration/response functions from cohort studies 16. The fact that these risk functions are 5–10‐times larger, than those observed in time-series studies, may suggest that the “within-days” effects of air pollution on death are only a small fraction of the entire dynamic between air quality and life expectancy. The disadvantage of this choice, however, is the sparse availability of cohort risk functions. To date, there are only three USA studies available 40–42, two of which have been extensively reanalysed 28. The preliminary confirmative European study has not yet been published 43. The heterogeneity across these few studies is larger than across time-series estimates, and impact assessments outside the USA may need to assume that the findings apply similarly to other regions. The uncertainties in the attributable death estimates are not quantifiable.

There are two ways to quantify the impact on “mortality”. The above-mentioned cohort studies published the RR for death, thus, it is straightforward to estimate the annual number of attributable deaths. As an alternative, the annual years of life lost may be quantified. These estimates are preferable in evaluating the public health relevance because they are sensitive to the age structure of the victims. Assuming cardiorespiratory causes of death, air pollution-attributed death may be, on average, rather old. So far, published studies did not provide estimates of the years of life lost nor the age structure of the premature death, thus, indirect estimates and assumptions must be relied upon, using life tables of the population. The shortening in population mean life expectancy has been estimated to be in the range of ∼6 to >24 months for an increase in the annual mean PM₁₀ of ∼10 µg·m⁻³ 10, 44, 45.

The impact of a mixture

The implications of air pollution being a mixture rather than one toxic substance are different for epidemiological studies and impact assessments. Epidemiological investigations are very versatile in using many surrogate measures of the air pollution mixture, e.g. nitrogen dioxide, particulate matter of various sizes, ozone, sulphur dioxide, carbon monoxide, etc. These pollutants are often correlated in time and/or space as they share common sources. The best surrogate measure for epidemiological research would be one that: 1) directly is, contains, or is highly associated with the biologically “causal agent”; 2) is geographically homogeneously distributed within a city, thus, a few monitoring stations can efficiently describe the “true” concentration; and 3) the indoor/outdoor penetration is high, thus, outdoor concentrations also fully reflect the indoor exposure to outdoor pollutants, i.e. “personal exposure” may be measured by central monitors.

Pollutants are rather different with regard to the second and third characteristics, thus, it remains a challenge of epidemiological studies to rank the biological importance across the single pollutants clearly. Accordingly, although the evidence is very strong that air pollution causes mortality and morbidity, it remains more difficult to allocate these effects to one or the other pollutant. In fact, it may be a scientifically weak or at least overly simplistic model to expect that the health problem of air pollution may be explained by one or a few single pollutants.

For health-impact assessment, however, the correlation of pollutants requires the selection of one or a few surrogate pollutants to derive the attributable cases. The sum of the “impact” of each pollutant will grossly overestimate the contribution of air pollution. PM₁₀ is currently the most often reported surrogate measure in epidemiological studies and, therefore, in the health-impact assessment. Ozone, which is a serious additional pollution and health problem in the summer, may be used as a further surrogate to estimate the impact of the oxidant pollution separately.

As shown in the model (fig. 1⇓), population-weighed exposure data are required. In many areas of the world, this might be a formidable challenge due to the lack of monitoring networks. Systematic air pollution monitoring ought to be implemented on a regular basis in all populated areas, otherwise health-impact assessments may be rather speculative.

Impact above what level of pollution?

The choice of the air pollution reference value (E₀ in fig. 1⇓) is of major influence for the number of cases attributed to air pollution. This choice may depend on the purpose of the study. In many countries, policy makers currently face the problem that air quality criteria regulations are intended to “protect health”, including the health of the most vulnerable people; to date, research has failed to obtain any evidence for a no-effect threshold. Thus, similar to carcinogens, the natural “threshold” might be zero exposure. Therefore, nonzero target values of clean air acts, inherently assume that some health impact of air pollution may be accepted. Impact assessors must choose a level below which they explicitly want to ignore the impact on air pollution. To demonstrate the influence of this choice, figure 2⇓ shows a model of the air pollution attributable proportion of chronic bronchitis (incidence) in adults as a function of the choice of “exposure reference level”. The model uses the data and additive attributable case functions used in the trinational study 15. Considering zero exposure, i.e. the full impact of air pollution, >20% of cases would be attributed to air pollution. However, the attributable proportion above the Swiss clean air act value (20 µg·m⁻³ PM₁₀ annual mean) would be well below 5%. The trinational WHO study chose 7.5 µg·m⁻³ (mean of exposure category 5–10 µg·m⁻³) to reflect some “natural background” level of air pollution. Given that impact assessments have rather different purposes and may need to address specific policy questions, investigators have used very different values so far 9, 31, 32, 46, 47. Therefore, studies cannot be compared directly, even in cases where the methodological framework may otherwise be the same.

The background frequency of diseases matters

The influence of the background incidence or prevalence of the health outcome in the impact assessment target population may be substantial (P in fig. 1⇓). Given that the calculation of attributable cases applies, in a first step, the RR function to the background frequency level P suggests that the number of attributable cases is clearly influenced by this value. Unfortunately, many health outcomes are not monitored on a population level and are often poorly defined. For example, although there is good evidence for an increase in asthma attacks among adults during days with increased air pollution, asthma attacks are not monitored and definitions of attacks vary across epidemiological studies. To demonstrate the influence of P on the final result, the prevalence of chronic cough and phlegm symptoms are used. Based on the SAPALDIA study 6, a 1.65% (95% confidence interval: 1.28–2.16) increase may be assumed in chronic bronchitis symptoms, among those with exposure to ETS 48. Furthermore, an increase in the annual mean PM₁₀ of 10 µg·m⁻³ was associated with a 1.27% (1.08–1.50) higher RR for these symptoms 6. Based on data from Zemp et al. 6, a crude RR of 2.4 can be derived for chronic bronchitis symptoms in smokers compared with nonsmokers. With the trinational study attributable risk model and the Swiss baseline data (table 3⇓), ∼20,000 prevalent cases may be attributed to air pollution in France, Austria, and Switzerland overall, at ∼7% prevalence of symptoms (fig. 3⇓). However, if the number of smokers is increased up to 100%, the chronic bronchitis symptom prevalence would be expected to increase to >17% 6. In such a population, the impact model of figure 1⇓ would attribute >50,000 cases of chronic bronchitis to the same level of air pollution (fig. 3⇓). Although the background frequency of diseases and health-relevant lifestyles may not be very different across Europe, the problem may become serious for some health outcomes i.e. if findings from mostly Western air pollution studies are applied to impact assessment studies in developing countries that have rather different patterns of morbidity and health hazards, apart from ambient air pollution. For example, in many Asian countries, a large fraction of cases of bronchitis among children may be due to indoor burning of solid fuel for cooking and heating, driving the prevalence of bronchitis 49. Application of Western air pollution/bronchitis risk functions, derived in regions where indoor combustion is almost nonexistent 50, to the Asian background frequency of acute bronchitis may lead to a large overstimate of the cases attributed to outdoor air pollution. Extrapolations from Western countries to developing regions are particularly prone to such errors. To reduce the risk of false overestimation of the health impact of outdoor air pollution, baseline frequency data from populations with low rates may be used, e.g. those reported in the nonsmoking Seventh Day Adventist study 33. Furthermore, case attribution calculations can be developed which transfer an attributable number of cases instead of the RR estimates from the source studies to the target population 51.

Will the health benefits of abatement happen?

It is typically the case in environmental politics that stakeholders may question the need for and benefit of clean air policies. Thus, authorities may be asked to predict the long-term costs and benefits of regulations 20. The question, thus, arises whether attributable cases as derived in the impact assessment truly reflect the number of cases that can be prevented under a clean air scenario. It must be acknowledged that such predictions inherently require further assumptions with additional levels of uncertainty and speculations. In fact, such studies may not add more than an “order-of-magnitude” idea about the distant future, and it remains inherently questionable whether it will ever be possible to rigorously and scientifically prove the long-term benefits of clean air strategies. At the least, long-term monitoring of both air pollution and health outcomes would be needed, in large populations. Given the long-term economic and cultural changes in societies, this will, however, not be enough to prove or disprove the long-term evidence. For instance, air quality may be improved but myocardial infarction may nevertheless increase, if smoking rates strongly increase. Such ecological comparisons are inherently flawed, thus, repeated full-scale epidemiological studies would be required to truly control confounding.

It must be emphasised that the health benefits may depend on the abatement strategy. Given that single pollutants may be surrogates for mixtures, highly pollutant-specific abatements may not yield the full range of health benefits. For more comprehensive strategies the question of the “culprit pollutant” is less relevant. For example, traffic-speed reductions or replacement of current car fleets with low- or no-emission cars would reduce a variety of pollutants and, thus, optimise health benefits whatever the relative importance of single pollutants may be.

For short-term effects of air pollution, there is some preliminary empirical evidence that the health benefit of air quality improvements may happen within short periods of time. Although population-based experimental studies are usually not feasible in air pollution research, at least three “natural experiments” may be cited which are in line with rather prompt health benefits of improved air quality. The first is the “Utah Valley experience” where a 1‐yr closure of a steel mill resulted in strong declines in air pollution, morbidity, and mortality. The reopening of the mill was accompanied by pollution and health rates climbing back to preclosure values 62. As reported by Heinrich et al. 63 the remarkable improvements in air quality in former East Germany were paralleled by a rather instant decline in respiratory symptom rates among children, which would be expected from the concentration risk function from a European epidemiological study 50. The third “experiment” was undertaken during the Olympic games in Atlanta, GA, USA. Although it may be argued that such unusual events may limit the adequate consideration of confounding, the study shows that strong restrictions in inner-city car use was followed by reductions in air pollutants and a decrease in respiratory emergency room visits among children 64. Although none of these studies can be conclusive, they are at least remarkably coherent with expectations under a scenario of causality. The studies suggest that short-term effects may in fact be fully prevented if air pollutants are lower. It may be worthwhile to foresee or plan such “natural experiments” in advance to further assess the evidence of “accountability” of policies. Potential target regions for such “experiments” may be areas where large developing projects are just about to be implemented, such as the closure or opening of new and large traffic arteries or industrial facilities.

However, for long-term effects and benefits, the assessment may be much more difficult. In fact, due to competing risks and large-scale changes in exposure profiles, over long periods of time, air pollution attributable cases of long-term effects (e.g. mortality and chronic bronchitis among adults in the trinational study), may not be fully “preventable cases”, or only many years after achieving better air quality 10. This limitation of case attribution is not restricted to air pollution but also applies to the derivation of smoking attributable death, where it is generally assumed but impossible to prove that “attributable death” is at the same time “preventable death”.

In the light of all the uncertainties and limitations, researchers should not lose sight of the general patterns and perspectives. Given the current level of evidence of the association between air pollution and health, the precautionary principle may provide excellent guide to rigorously implement clean air strategies. Horton 65 well described the precautionary principle as follows:

“We must act on facts, and on the most accurate interpretation of them, using the best scientific information. That does not mean that we must sit back until we have 100% evidence about everything. Where the state of the health of the people is at stake, the risks can be so high and the costs of corrective action so great, that prevention is better than cure. We must analyse the possible benefits and costs of action and inaction. Where there are significant risks of damage to the public health, we should be prepared to take action to diminish those risks, even when the scientific knowledge is not conclusive, if the balance of likely costs and benefits justifies it” 65.

Currently available impact assessment studies clearly suggest that public health largely benefits from better air quality. These studies are rather selective underestimates as they are strongly driven by mortality but do not fully include quantification of the impact on morbidity, and their consequences on quality of life among the diseased and the caregivers. It is also of note that key clean air strategies converge into abatement of climate change 66. In general, energy consumption is very closely related to both air pollution and greenhouse gases. The dominant causes of both problems are the excessive and inefficient combustion of fossil fuel. Thus, for many policy options, the benefit of air pollution abatement will go far beyond what prudent health-impact assessments may derive. From a climate change and air pollution perspective, improved energy efficiency and a strong and decisive departure from the “fossil fuel” combustion society is a science-based must.