Since the HRAPIE project report was published, a number of policy-relevant questions have come to our attention, raised by readers of the report. The following paragraphs provide some further perspectives on the HRAPIE recommendations, in response to the issues raised.
Use of thresholds, lower levels, counterfactuals, etc., in quantifying health impacts
The HRAPIE report recommended that quantification of long-term effects of PM2.5 on mortality be undertaken at all concentrations, whereas for NO2, quantification was recommended only above annual mean concentrations >20 µg/m3. This has raised some confusion.
We point out that following the recommendation for estimating impacts of PM2.5 down to zero concentrations, higher impacts are estimated than in the GBD 2010 study (Lim et al. 2012). This is because the GBD 2010 study estimated impacts down to a counterfactual concentration below which—because of a lack of data—the risk could not be quantified; this counterfactual for ambient air pollution is different than zero and was chosen as an alternative exposure distribution to which the current exposure distribution is compared (Burnett et al. 2014). The selection of counterfactual levels for risk factors, including for ambient particulate air pollution, were informed by: (1) a counterfactual population distribution of exposure that is theoretically possible and would result in the lowest population disease burden—the theoretical minimum risk exposure distribution (TMRED), and (2) the availability of convincing evidence supporting a continuous reduction in risk of disease extending from current levels down to the counterfactual TMRED (Murray et al. 2012). On this basis, the TMRED for ambient PM2.5 was defined by a uniform distribution with lower and upper bounds at the minimum and 5th percentile of the PM2.5 exposure distribution of the American Cancer Society (ACS) Cancer Prevention II cohort study of 5.8 and 8.8 µg/m3, respectively (Krewski et al. 2009; Lim et al. 2012; Burnett et al. 2014). This approach assumed that long-term exposure to PM2.5 less than 5.8 µg/m3 confers no excess risk, and that although excess risk may extend below the 5th percentile of the distribution, estimates in that range are statistically unstable and therefore highly uncertain. Uncertainty from the counterfactual, the prevalence of exposure, and the exposure–response function was all propagated into the final risk factor uncertainty, and the final uncertainty reflected that of the estimated age- and sex-specific mortality rates as well. For future work under GBD 2013 the same approach is used but includes information from 9 cohort studies with minimum exposure concentrations less than or equal to the 5th percentile in the ACS, i.e. 8.8 µg/m3. Whenever impacts of policy measures are estimated using the HRAPIE CRFs, results will not differ greatly from those obtained using the GBD approach, as for the medium to long term, concentrations in most, if not all, areas in Europe will not likely go below 5.8 µg/m3. We also point out that under the HRAPIE project, due to wider availability of risk estimates and greater precision of background national data for all-cause mortality in Europe, natural all-cause mortality was chosen as the outcome for quantification. This is different to what was chosen as outcome as part of the GBD project, where cause-specific mortality was used because patterns of causes of death vary considerably globally.
The lower limit of 20 µg/m3 (annual) for NO2 was motivated by reference to the Naess et al. (2007) Norwegian study, and by unpublished analyses of the data presented in the Cesaroni et al. (2013) paper from Rome. These studies assessed NO2 effects in single-pollutant models. The Naess et al. (2007) paper shows a generally linear relationship between NO2 and mortality, among 71–90 year olds, in the 20–60 µg/m3 range. Figure 1 in that paper actually shows a steeper CRF in the 0–20 µg/m3 range, with wider confidence intervals due to the smaller numbers of participants at such low exposures. The Cesaroni et al. (2013) paper continued to show a linear decline below 20 µg/m3, although again with wider confidence intervals. The evidence presented does not suggest that the effect is zero at 20 µg/m3, just that the size of the effect is less certain below 20 µg/m3. We also note that a recent cohort study conducted among ca. 52,000 adults in Copenhagen, Denmark (Raaschou-Nielsen et al. 2012) has shown a significant, almost linear concentration–response relationship between long-term NO2 concentration (chosen by the authors as an indicator of urban air pollution dominated by traffic exhaust) and mortality [for cardiovascular disease (CVD), ischemic heart disease and all causes] throughout the observed range of NO2 concentrations, which in the large majority of subjects was below 20 µg/m3 (minimum 10.5 µg/m3, median 15.1 µg/m3, maximum 59.6 µg/m3). This study was included in the Hoek et al. (2013) meta-analysis, but we did not explicitly consider it when discussing lower limits of quantification in the HRAPIE project. All-cause mortality increased by 8 % per 10 µg/m3 NO2 long-term exposure at the residence address in the study by Raaschou-Nielsen et al. (2012), so slightly more than estimated in the Hoek et al. meta-analysis. Therefore, the HRAPIE recommendation to calculate the impacts of long-term NO2 exposure on mortality for levels over 20 µg/m3, ignoring potential impacts at lower concentrations, may be too conservative.
Age dependency of CRFs
The HRAPIE report briefly discussed reasons why RRs for factors such as smoking and CVD decline with age, and how that might impact calculations. Air pollution studies have not generally investigated age dependency of RRs in any detail, but there is some specific information to suggest that this is also true for relationships between air pollution and all-cause and cause-specific mortality. Over a 10-year follow-up period, an analysis of the ACS-1 study (Enstrom 2005) showed a RR for PM2.5 and all-cause mortality twice as high among subjects aged 43–64 years at baseline than in subjects aged 65–99 years at baseline (Table 6 of the paper). The study by Naess et al. (2007) showed, for both men and women, a clearly higher RR of CVD as well as chronic obstructive pulmonary disease (COPD) mortality in subjects aged 51–70 years at baseline than in subjects aged 71–90 years at baseline (Tables 3 and 4 of the paper). In the Harvard Six Cities Study, as reanalysed by Krewski et al. (2000), the RR for all-cause mortality in relation to an 18.6 µg/m3 increase in PM2.5 was 2.11 (0.88–5.07) for those <40 years old at baseline, 1.66 (1.17–2.35) for those 41–55 years old, and 1.17 (0.98–1.40) for those >55 years old at baseline. The same reanalysis showed that, in the ACS study, for those with high school education or less, the RR for all-cause mortality in relation to a 24.5 µg/m3 increase in PM2.5 was 1.51 (1.00–2.27) for those <50 years old at baseline, 1.27 (1.02–1.60) for those 50–60 years old, and 1.28 (1.14–1.43) for those >60 years old at baseline. Early work had already shown that impacts on life expectancy may be over-estimated when RRs observed at certain specific ages in cohort studies are applied to subjects at higher ages with high baseline mortality (Brunekreef, 1997). Most current risk assessments assume that the excess relative risk among adults does not vary with age, but the GBD 2010 estimates incorporated age-dependency of the air pollution relative risk(s) such that the age-specific excess relative risk for cardiovascular mortality declined with increasing age leading to lower estimates of attributable burden (Lim et al. 2012; Burnett et al. 2014).
Cessation lag between reduced long-term exposure to PM2.5 and mortality
The distribution of the cessation lag is relevant both for the size of the effect over a defined time period and when calculating the economic value of the effect due to discounting the value over time. The HRAPIE report stated that findings from the follow-up of the Harvard Six Cities study suggest that mortality effects may be partially reversible, over a time period possibly as short as a year. Ideally, for health impact calculations, a range of possible delays between reduced exposure and a reduced impact on mortality would be used. The HRAPIE report noted that long-term exposure to PM2.5 is linked with lung cancer mortality. Delays for this are likely to be measured in decades, although lung cancer mortality is a small proportion of the total. Other organizations such as the United States Environment Protection Agency (US EPA) have recommended a distribution of different lags (US EPA 2011). Readers are referred to this and other documents (Hurley et al. 2005; COMEAP 2010) for a more detailed discussion.
Application of HRAPIE recommendations and implications for European Union air policy
An impact assessment accompanying the EU policy package was developed by the EC and provides the results of the implementation of the HRAPIE recommendations in a cost–benefit context (EC 2013). According to the EC’s impact assessment, over 406,000 premature deaths were estimated to be related to long-term PM2.5 and short-term O3 exposure in year 2010. New evidence on impacts from chronic O3 exposure is not included but would add around 5 % to this total (EC 2013). The evidence of health impacts from NO2 exposure was considered, but as there was a lack of agreement regarding the extent to which the exposure data used by the EC properly reflected exposure of the population (Holland 2014), no quantification was made of NO2 health impacts. Further work is needed to characterize the link between estimated NO2 exposure and health outcomes as provided in the recommendations of the HRAPIE report.
According to the EC assessment, the mortality associated with PM, the most important pollutant, has been reduced by around 20 % between 2000 and 2010 (EC 2013). Modelled trends in pollutant levels show that under a business-as-usual scenario (baseline projection) the impacts of air pollution will continue to decrease by 2020, where they will amount to an estimated 340,000 premature deaths. The progress in further reducing the health impacts from air pollution is expected to be considerably slower beyond 2020. On average across the EU, baseline projection suggests a decline of the loss of statistical life expectancy attributable to the exposure to PM2.5 from 8.5 months in 2005 to 5.3 months in 2025. Depending on the valuation methodology [Value of Statistical Life (VSL) or Value of a Life-Year (VOLY)], the health-related external costs from air pollution ranged between €330 billion and €940 billion in 2010, and would be reduced in the baseline to €210–730 billion in 2030 (2005 € prices).
The corresponding benefits of the proposed air policy package can be monetized, resulting in about €40–140 billion per year in 2030, while the costs of pollution abatement to implement the package are estimated to reach € 3.4 billion per year in 2030. The impact assessment states that the monetized benefits will therefore be about 12–40 times higher than the costs (EC 2013).
The policy package does not propose changes to the existing air quality standards in the ambient air quality directive at this stage. The EC acknowledges that the current standards are insufficient for the protection of public health, particularly in reference to the WHO air quality guidelines. The focus will be on a full attainment of current air quality standards by 2020. The new policy proposes stricter national emission ceilings and new source legislations which, according to the EC, are expected to pave the way for tightened standards in the ambient air quality directive at a later stage. A 5-year policy review cycle is being considered with a first review taking place no later than 2020, at which time the scope for tightening the air quality standards will be considered by the EC (EC 2013).
The scientific evidence is rapidly expanding, reaffirming and strengthening previously reported associations as well as revealing new health outcomes. Special efforts are required from the scientific community and the policy-makers to engage in a dialogue and enable proper interpretation and synthesis of the scientific evidence for use in policy formulation. The HRAPIE project report, summarized in this paper, illustrates the complexities involved in recommending suitable CRFs, with accompanying recommendations on methods, baseline rates and appropriate strategies for combining results. These complexities need to be acknowledged, and sufficient resources should be made available to enable proper synthesis and interpretation of the evidence in future reviews of air pollution health effects.
The REVIHAAP and HRAPIE projects provide scientific arguments for taking decisive actions to improve air quality to further reduce the burden of disease associated with air pollution in Europe. They further recommend that the EC ensures that the evidence on the health effects of air pollutants and the implications for its air quality policy is reviewed regularly. The material developed as part of these projects is equally relevant to all Member States of the EU, in their development and implementation of effective strategies to reduce air pollution and its significant impacts on public health. The two projects described above also provide an important input to the development of air quality policies by the parties of the United Nations Economic Commission for Europe (UNECE) Convention on Long-range Transboundary Air Pollution (LRTAP), which are outside the EU, especially Member States from the eastern part of the WHO European Region.