General findings and uncertainties
This study juxtaposes two risk assessment approaches combining four concentration scenarios with two counterfactual choices to put lung cancer deaths attributable to ambient air pollution into the context of risk assessment methods and concepts. In line with previous studies, we found that the sum of the unit risk-based attributable deaths across single carcinogens identifies only a fraction of the total burden captured with the excess rate-based epidemiological approach for PM10 (see Supplementary Materials). To guarantee comparability with previous assessments, we used PM10 instead of PM2.5 as the marker or air pollution (e.g., Röösli et al. 2003). In line with those studies and the Global Burden of Disease (Cohen et al. 2017), we used attributable cases instead of years of life lost (Héroux et al. 2015, 2017; Morfeld and Erren 2017).
Our quantitative comparison of the toxicology-based paradigm with the epidemiology-based assessment of attributable deaths reveals interesting differences in the (implicit) acceptance of risk underlying these two approaches. As shown in Tables 1 and 2, none of the PM10 scenarios fully complies with tolerating any risk level for five carcinogens.
The number of attributable deaths differs both in relative and in absolute terms under a range of alternative methodological assumptions to be discussed in more detail below.
First, our two counterfactual PM10 concentrations (7.5 vs. 3.3 μg/m3) highlight the strong influence of this parameter. Although it is appropriate to disclose attributable deaths down to very low counterfactual levels, it should be well communicated that the apparent increase in the attributable burden is caused by the alternative counterfactual value rather than by changes in the toxicity of air pollution.
Second, the values we choose for the relative risk determine the excess rate in the epidemiological approach. Ideally, the relative risk estimate would originate from Switzerland, but this is not available. We used the worldwide PM2.5 relative risk estimate for lung cancer incidence from the meta-analysis of Huang et al. (2017). We selected this relative risk because it is (1) from the most recent meta-analysis, (2) specific for incidence (not mixed with mortality) and (3) based on a higher number of studies than the European estimates. This choice results in a number of deaths rather similar to the one estimated in a study commissioned by the Swiss Federal Office for Spatial Development (ECOPLAN and INFRAS 2014). For public authorities, methodological consistencies facilitate the communication of results over time. However, one could also argue for other choices from the identified nine relative risk estimates published in three international meta-analyses (Raaschou-Nielsen et al. 2013; Hamra et al. 2014; Huang et al. 2017). Depending on the choice of relative risk, the attributable annual lung cancer deaths for the scenario A2 (255 in our study) range from 98 to 1079 (see Supplementary Materials). Smoking cannot explain this heterogeneity in the relative risk estimates because the studies used for the calculation of the relative risk estimate adjusted for smoking (among other factors). Whereas public authorities may prefer using the same relative risks for all consecutive studies to better compare results and trends, it is inevitable that new and possibly more appropriate risk estimates get published and, thus, used in risk assessments. Therefore, there is a need for proper communication strategies to explain the meaning of uncertainties and “conflicting results,” which are driven by methodological choices rather than by changes in the toxicity of air pollution.
Third, the choice of the lung cancer incidence impacts the excess rate. We used average incidence data from the period 2011–2015 rather than some theoretical “baseline incidence” before exposure to ambient air pollution. The latter is not available, but we conjecture this uncertainty to be of minor influence given that lung cancer incidence is most strongly driven by smoking, which tended to become less prevalent over the past decades.
Fourth, the choice of unit risk factors determines the result of the health assessment in the toxicological approach. Most unit risks are based on occupational studies (see Supplementary Materials). Transferability of the risk estimates to the general population involves uncertainties. On the one hand, this implies extrapolation of risk functions with unknown errors from much higher occupational exposures down to ambient air concentrations. On the other hand, the higher proportion of vulnerable persons in the general population or the higher toxicity of metals in acid ambient aerosols (Nordberg et al. 1985) may result in the underestimation of risks, if one relies on occupational studies alone. Similarly, the combined interaction of multiple carcinogens or between carcinogens and other pollutants is not captured in the occupational studies (Kawaguchi et al. 2006; Berenbaum 1985); thus, the health burden might be underestimated.
Fifth, the inclusion of additional carcinogens would increase the number of attributed deaths. Furthermore, some of the considered carcinogens are markers of larger groups of substances. If we had included the effect of the whole group, the resulting health burden would have been higher (see Supplementary Materials). We conclude that the restriction to five carcinogens explains part of the strong difference between the PM10 and carcinogen-based attributable deaths of lung cancer. PM10 captures not only all particle-bound carcinogens but also various interactions between these substances as well as, to some extent, interactions with correlated exposures to gases.
Sixth, derived population-weighted mean concentrations of PM10 and carcinogens might have some uncertainty, because they are based on a limited number of monitoring stations (up to ten in our study), but the stations are representative for most populous areas. Alternatively, PM10 can rely on comprehensively validated hybrid maps using spatial models, based on a range of monitoring stations, emission data and spatial information. For 2010 (scenarios A2, B2 and C1), the estimated concentration from the model was only 3% higher than the one from the NABEL stations used in our analyses; thus, our study is not sensitive to this methodological choice. A further non-quantifiable uncertainty relates to the selected year(s) to derive the exposure. Lung cancer has a long latency period, i.e., the incidence is a result of “past long-term exposure.” We used data from 2010; thus, the implicit assumption is that these values also stand for the longer-term exposure. However, the PM10 population-weighted concentration decreased strongly between 1991 and 2015 from over 30 to approximately 15 µg/m3. Similarly, concentrations of carcinogens were also reduced by varying proportions. Although the size of these temporal uncertainties is unknown, we expect all scenarios to be similarly affected; thus, comparisons across approaches and scenarios remain valid.
Seventh, we assumed that the survival rate of lung cancer cases was zero. The 10-year survival rate of Swiss lung cancer patients between 1998 and 2012 was on average 10% (11% for women and 9% for men) (Arndt et al. 2016). If we applied a nonzero survival rate, one would have obtained a proportionally lower number of attributable deaths. However, survival data for periods beyond 10 years—relevant for our risk assessment—are not available. If lung cancer is ultimately considered non-curable, our assumption may result in a negligible bias.
Policy implications
A major motivation of this study related to the question, whether the current regulatory framework of PM, with its science-based air quality standards, remains an adequate choice despite PM now being accepted as a carcinogen. As shown in our assessment, all risk models of the toxicological approach for five carcinogens correspond to accepting much less lung cancer deaths in Switzerland than the ones attributed to PM10. However, although the approach to define “acceptable” cases is apparently much stricter, we see a range of advantages in maintaining air quality standards versus replacing it with the risk-level framework commonly used for single carcinogens.
First and foremost, PM10 is not only a carcinogen but causes a range of non-cancer morbidities and related premature deaths such as cardiovascular and respiratory diseases (WHO-Europe 2013). Furthermore, other types of cancer beyond lung cancer have been associated with PM exposure, e.g., sinonasal cancer (WHO-Europe 2000, p. 202), oral cancer (Chu et al. 2018) and possibly breast cancer (Andersen et al. 2017; White et al. 2018; Cheng et al. 2019). Indeed, the list of identified health effects of PM is constantly increasing. Under a policy framework of “acceptable” risk levels, e.g., 1 in 1,000,000, the “acceptable” target concentration would constantly change, namely decrease, with every additional outcome considered to be causally related to PM. Such “moving targets” are not only difficult to communicate to policymakers and the population at large, but also pose a major challenge for the agencies in charge of clean air development plans. In addition, “moving targets” jeopardize the proper communication of progress in clean air policy. Indeed, a policy framework defining the number of “acceptable” cases instead of setting ambient air quality standards, as used for all criteria pollutants, would force policymakers to define the number of “acceptable” cases for each air pollutant and each of the many health outcomes to then derive the related clean air target value (Thurston et al. 2017).
For carcinogens not regulated with limit values, we rather recommend agencies to continue the “as low as possible” policy. In line with this notion, the Swiss Federal Commission for Air Hygiene (EKL in German) recommended in 2013 to reduce airborne elemental carbon, as a marker of diesel exhaust, to 20% of the levels observed at that time, within 10 years (EKL 2013). Based on Table 2, this recommendation approximately corresponds to accepting around five deaths per year and it only complies with a level of risk of 1 in 1,000,000.
As shown in our assessment, air quality standards for PM provide a transparent base to estimate premature deaths under a broad range of policy scenarios. We consider of particular interest our scenario using 11 µg/m3 as a counterfactual PM10 concentration to comply with the newly adopted annual PM2.5 limit value. PM10 concentrations are substantially determined by the PM2.5 values, and over the past decades, clean air policies reduced ambient concentrations of both particle fractions in parallel. However, the OPAC annual air quality standards of PM2.5 (10 μg/m3) are de facto more stringent than the related PM10 target (20 μg/m3). Indeed, whereas all Swiss monitoring sites comply with the latter, PM2.5 concentrations remain above the limit values at several sites. Once PM2.5 values comply at all sites, including hot spots, the population-weighted mean PM2.5 is expected to be close to 8.3 μg/m3 and PM10 concentrations approximately at 11 μg/m3, assuming that 73.5% of PM10 consist of PM2.5 (BAFU 2019).
Our findings may also guide the upcoming revision of the WHO Air Quality Guidelines (WHO-Europe 2016), where the lack of an apparent PM threshold of no adverse effect and its definition as a carcinogen cannot be ignored either. According to the above arguments, we consider the promotion of fixed air quality guideline values appealing and appropriate. A major challenge of the WHO Air Quality Guideline does not relate to the science-based derivation of such limit values, but to globally convince governments to adopt these values in national regulations, to enforce clean air strategies (Kutlar Joss et al. 2017), to communicate health benefits of clean air policies (Henschel et al. 2012) and to provide guidance in the interpretation of the burden of ambient air pollution given its mixture of many pollutants (Héroux et al. 2015).
Conclusions
Our comparison of the epidemiological and toxicological approach to assess the lung cancer burden in the whole population has shown that the epidemiological approach using a marker of air pollutants, e.g., PM, can better cover the exposure of the whole population than a limited selection of single carcinogenic air pollutants. Thus, applying a toxicological approach for only five inhalable particle-bound carcinogens with a risk level of 1 in 1,000,000, 1 in 100,000 and 1 in 10,000 for each carcinogen resulted in a number of lung cancer deaths that is smaller than the more comprehensive epidemiology-based derivation for PM10. Whereas single carcinogens may be regulated under an “acceptable” number of cases risk framework, our study emphasizes the advantage of air quality limit values to regulate complex mixtures of particulates or particle-bound pollutants such as PM, irrespective of their carcinogenicity or the absence of thresholds of no effect. Setting science-based ambient standards at a fixed level as promoted by the WHO Air Quality Guidelines remains a pragmatic, transparent and efficient tool to guide effects-oriented clean air policymaking and to monitor its success.