Mortality attributable to ambient fine particulate matter and nitrogen dioxide in Switzerland in 2019: Use of two-pollutant effect estimates

Introduction: Air pollution health risk assessments have traditionally used single-pollutant effect estimates for one proxy ambient air pollutant such as PM2.5. Two-pollutant effect estimates, i.e. adjusted for another correlated pollutant, theoretically enable the aggregation of pollutant-specific health effects minimizing double-counting. Our study aimed at estimating the adult mortality in Switzerland in 2019 attributable to PM2.5 from a single-pollutant effect estimate and to the sum of PM2.5 and NO2 from two-pollutant estimates; comparing the re-sults with those from alternative global, European and Swiss effect estimates.Methods: For the single-pollutant approach, we used a PM2.5 summary estimate of European cohorts from the project ELAPSE, recommended by the European Respiratory Society and International Society for Environmental Epidemiology (ERS-ISEE). To derive the two-pollutant effect estimates, we applied ELAPSE-based conversion factors to ERS-ISEE PM2.5 and NO2 single-pollutant effect estimates. Additionally, we used World Health Or-ganization 2021 Air Quality Guidelines as counterfactual scenario, exposure model data from 2019 and Swiss lifetables.Results: The single-pollutant effect estimate for PM2.5 (1.118 [1.060; 1.179] per 10 mu g/m3) resulted in 2240 deaths (21,593 years of life lost). Using our derived two-pollutant effect estimates (1.023 [1.012; 1.035] per 10 mu g/m3 PM2.5 adjusted for NO2 and 1.040 [1.023; 1.058] per 10 mu g/m3 NO2 adjusted for PM2.5), we found 1977 deaths (19,071 years of life lost) attributable to PM2.5 and NO2 together (23% from PM2.5). Deaths using alternative effect estimates ranged from 1042 to 5059.Discussion: Estimated premature mortality attributable to PM2.5 alone was higher than to both PM2.5 and NO2 combined. Furthermore, the proportion of deaths from PM2.5 was lower than from NO2 in the two-pollutant approach. These seemingly paradoxical results, also found in some alternative estimates, are due to statistical imprecisions of underlying correction methods. Therefore, using two-pollutant effect estimates can lead to interpretation challenges in terms of causality.