Since January 2020 Elsevier has created a COVID-19 resource centre with free information in English and Mandarin on the novel coronavirus COVID- 19. The COVID-19 resource centre is hosted on Elsevier Connect, the company's public news and information website. Elsevier hereby grants permission to make all its COVID-19-related research that is available on the COVID-19 resource centre - including this research content - immediately available in PubMed Central and other publicly funded repositories, such as the WHO COVID database with rights for unrestricted research re-use and analyses in any form or by any means with acknowledgement of the original source. These permissions are granted for free by Elsevier for as long as the COVID-19 resource centre remains active. Environmental Pollution 286 (2021) 117220 Contents lists available at ScienceDirect Environmental Pollution journal homepage: www.elsevier.com/locate/envpol Trade-offs between short-term mortality attributable to NO2 and O3 changes during the COVID-19 lockdown across major Spanish cities☆ Hicham Achebak a, b, *, Hervé Petetin c, Marcos Quijal-Zamorano a, Dene Bowdalo c, Carlos Pérez García-Pando c, d, Joan Ballester a a Climate and Health Program, Barcelona Institute for Global Health, Barcelona, Spain b Centre for Demographic Studies, Autonomous University of Barcelona, Barcelona, Spain c Barcelona Supercomputing Center, Barcelona, Spain d ICREA, Catalan Institution for Research and Advanced Studies, Barcelona, Spain A R T I C L E I N F O A B S T R A C T Keywords: The emergence of the COVID-19 pandemic forced most countries to put in place lockdown measures to slow COVID-19 lockdown down the transmission of the virus. These lockdowns have led to temporal improvements in air quality. Here, we NO2 evaluate the changes in NO2 and O3 levels along with the associated impact upon premature mortality during the O3 COVID-19 lockdown and deconfinement periods along the first epidemic wave across the provincial capital cities Mortality Spain of Spain. We first quantify the change in pollutants solely due to the lockdown as the difference between business-as-usual (BAU) pollution levels, estimated with a machine learning-based meteorological normalization technique, and observed concentrations. Second, instead of using exposure-response functions between the pollutants and mortality reported in the literature, we fit conditional quasi-Poisson regression models to estimate city-specific associations between daily pollutant levels and non-accidental mortality during the period 2010–2018. Significant relative risk values are observed at lag 1 for NO2 (1.0047 [95% CI: 1.0014 to 1.0081]) and at lag 0 for O3 (1.0039 [1.0013 to 1.0065]). On average NO2 changed by − 51% (intercity range − 65.7 to − 30.9%) and − 36.4% (− 53.7 to − 11.6%), and O3 by − 1.1% (− 20.2 to 23.8%) and 0.6% (− 12.4 to 23.0%), during the lockdown (57 days) and deconfinement (42 days) periods, respectively. We obtain a reduction in attributable mortality associated with NO2 changes of − 119 (95% CI: -273 to − 24) deaths over the lockdown, and of − 53 (− 114 to − 10) deaths over the deconfinement. This was partially compensated by an increase in the attributable number of deaths, 14 (− 72 to 99) during the lockdown, and 8 (− 27 to 50) during the deconfinement, associated with the rise in O3 levels in the most populous cities during the analysed period, despite the overall small average reductions. Our study shows that the potential trade-offs between multiple air pollutants should be taken into account when evaluating the health impacts of environmental exposures. 1. Introduction industry, tourism), leading to substantial reductions in anthropogenic emissions of air pollutants (Guevara et al., 2021) and temporal im­ Exposure to outdoor air pollution is a major environmental risk provements in air quality, at least for some pollutants (Bauwens et al., factor contributing to a range of adverse health outcomes (Kampa and 2020), which is expected to have an immediate public health benefit Castanas, 2008; Shah and Balkhair, 2011), with millions of premature (Isaifan, 2020). annual deaths across the globe (Cohen et al., 2017; European Environ­ In Europe, given the large contribution of road transport to NOX ment Agen, 2019). The emergence of the COVID-19 pandemic has forced emissions, and its short lifetime, the COVID-19 lockdowns have led to societies to put in place lockdown measures to slow down the trans­ unprecedented reductions in ambient NO2 concentrations, reaching up mission of the virus. Lockdowns have severely impacted the activity of to about − 50% in urban areas (Petetin et al., 2020; Barré et al., 2020; key socio-economic sectors (e.g., aviation, road transport, energy, Grange et al., 2020). The magnitude and sign of the changes in ☆ This paper has been recommended for acceptance by Da Chen. * Corresponding author. Barcelona Institute for Global Health (ISGlobal), Carrer Doctor Aiguader 88, 08003, Barcelona, Catalonia, Spain. E-mail address: hicham.achebak@isglobal.org (H. Achebak). https://doi.org/10.1016/j.envpol.2021.117220 Received 17 February 2021; Received in revised form 1 April 2021; Accepted 20 April 2021 Available online 4 May 2021 0269-7491/© 2021 Elsevier Ltd. All rights reserved. H. Achebak et al. Environmental Pollution 286 (2021) 117220 secondary pollutants such as O3, as well as a large fraction of PM2.5, are stations between 1 January 2010 and 21 June 2020. We computed 24-hr more heterogeneous partly due to the highly nonlinear atmospheric mean pollutant concentrations only if at least 75% of the hourly data chemistry involved. In the case of O3, the changes depend upon local were available. When more than one air quality station was available for factors including the amount of NOX, the amount and reactivity of the a city, the daily values were averaged across stations. NO2 and O3 data VOCs, oxidant levels, and meteorology (Kroll et al., 2020) along with were respectively missing in 4.3% and 5.3% of days on average. variations in imported O3 from other regions. Generally, O3 decreases when reducing NOX in VOC-rich environments like rural areas. In 2.2. Association between daily NO2, O3 and mortality contrast, O3 can increase in NOX-saturated environments such as large urban areas when reducing NOX due to a reduced titration by NO. An A two-stage analysis approach was developed to assess the short- increase of 20%–30% in the daily O3 levels has been reported on average term effects of NO2 and O3 on mortality. In the first stage, a time- across a selection of the main European cities between February and stratified case-crossover design with conditional quasi-Poisson regres­ July 2020 (Grange et al., 2020). Another study including most of the sion model (Armstrong et al., 2014) was applied to estimate the available European measurement stations (Ordóñez et al., 2020), city-specific associations between daily NO2 and O3 levels and showed either enhancements or reductions of the daily maximum 8-h non-accidental mortality (ICD-10: A00-R99) during the period running average O3 in urban background and traffic sites and re­ 2010–2018, summarized as relative risk (RR). These associations were ductions in rural sites between March and April. characterized by using an unconstrained distributed lag model (DLM) In this paper, we evaluate the changes in NO2 and O3 levels due to the with lags up to 3 days, which allows the description of the delayed ef­ COVID-19 lockdown and the associated predicted impact upon prema­ fects of air pollutants on mortality. The regression model also included a ture mortality across the provincial capital cities of Spain, which is stratum defined by a three-way interaction term of year, calendar among the countries most severely affected by COVID-19. We focus on month, and day of the week to control for the seasonal and long-term the first epidemic wave, discriminating among three distinct phases of trends and the weekly cycle of mortality; a linear term to adjust for the lockdown between 15 March and 10 May 2020 (57 days), and the relative humidity; and a distributed lag non-linear model (DLNM) deconfinement between 11 May and 21 June (42 days). Recent studies (Gasparrini et al., 2010) to account for the non-linear and delayed ef­ have assessed the potential reduction in mortality due to improved air fects of temperature. The DLNM is based on the definition of a quality during the lockdown periods in China (Chen et al., 2020; Giani cross-basis, a bi-dimensional space of functions describing simulta­ et al., 2020; Nie et al., 2021) and Europe (Giani et al., 2020), showing a neously the shape of the relationship along the exposure range (expo­ substantial number of avoided predicted fatalities from PM2.5 and NO2 sure-response function) and its distributed lag effects (lag-response that, in case of China, could have surpassed the number of reported function). Specifically, the exposure-response function was modelled deaths from COVID-19 (Chen et al., 2020). through a quadratic B-spline, with one internal knot placed at the 75th Our methodological framework attempts to overcome some of the percentile of the daily temperature distribution (Achebak et al., 2019), simplifications and assumptions that may be partly biasing some esti­ and the lag-response function was modelled through a natural cubic mates provided in previous studies (Achebak et al., 2020). On the one B-spline, with an intercept and three internal knots placed at equally side, quantifying the magnitude and the sign of the change in pollutants spaced values in the log scale of the lag period 0–14. The quasi-Poisson solely due to the lockdown in each location is challenging, in particular regression model was as follows: for O3 due its strong sensitivity to meteorological variability. We Log(mortality) = intercept + stratum + dlm(pollution, lag = 0–3) + effectively cancel out the effects of meteorological variability upon the dlnm(temperature, lag = 0–14) + humidity. changes in pollutant concentrations during the lockdown based on a The modelling choices were thoroughly tested in sensitivity analyses machine learning meteorological normalization technique (Petetin and, in addition to the model described above, two-pollutant models et al., 2020; Barré et al., 2020; Grange et al., 2020). On the other side, we were also fitted to test the robustness of the associations between NO2 purposely avoid the use the exposure-response functions between air and O3 and mortality (Supplementary Information, pp 3–4). pollutants and mortality reported in the literature to transform changes In a second stage, a multivariate random-effects meta-analysis was in the pollution levels into changes in mortality. Such an approach can performed to estimate the average association between air pollutants lead to biases due to the lack of consistency in the periods, locations and and mortality across cities, and to derive the best linear unbiased pre­ sources of data between the calibration of the exposure-response models dictions of the pollutant-mortality relationship in each location. in the former studies and the estimation of pollution reductions in the latter ones. Instead, we fit epidemiological models based on historical 2.3. Business-as-usual NO2 and O3 levels during the lockdown and health and air pollution data in each major Spanish city, and then we deconfinement periods estimate the meteorology-normalized reduction in pollutant concen­ trations at the same locations to consistently compute the change in Quantifying the impact of the lockdown on air pollution requires attributable deaths during the lockdown and deconfinement periods. estimating the business-as-usual (BAU) pollutant concentrations that would have been observed in the absence of the mobility restrictions. 2. Data and methods Given the strong influence of meteorology upon the variability of pollutant concentrations, the use of climatological concentrations ob­ 2.1. Data collection tained from historical data may provide biased BAU concentrations, especially if specific meteorological anomalies prevail locally. Here, we We collected health and environmental data for 47 provincial capital implemented machine learning (ML) models following the methodology cities in mainland Spain and the Balearic Islands. We obtained daily described and validated in Petetin et al., 2020 to obtain the daily BAU counts of death from the Spanish National Institute of Statistics (INE), NO2 and O3 concentrations and their associated uncertainties during the and daily mean observations of temperature (◦ C) and relative humidity lockdown and deconfinement. ML models fed with ERA5 reanalysis (%) from weather stations data available through the European Climate meteorological data (daily minimum, mean and maximum 2-m tem­ Assessment and Dataset (ECA&D) (Royal Netherlands Meteoro, 2020), perature, surface wind speed, normalized 10-m zonal and meridional both covering the period from 2010 to 2018. Mortality data had no wind speed components, surface pressure, total cloud cover, net solar missing values, whereas temperature and relative humidity were radiation at the surface, downward solar radiation at the surface, missing in 1.6% and 4.5% of days on average, respectively. We also downward UV radiation at the surface, and the boundary layer height) obtained hourly quality-assured NO2 and O3 observations from the EEA and time variables (date index, Julian date, day of the week) were Air Quality eReporting (European Environment Agen, 2020) urban trained with historical 2017–2019 data and used to predict NO2 and O3 2 H. Achebak et al. Environmental Pollution 286 (2021) 117220 daily concentrations in 2020. As demonstrated in Petetin Petetin et al., machine learning-based BAU pollution estimates, assuming random 2020, these models successfully estimate weather-normalized changes daily BAU level uncertainties across the reported periods. We note that in pollutant concentrations during the COVID-19 lockdown. We evalu­ our approach assumes no change in the risk of dying from air pollution ated the ML-based predictions in all cities during 2020 and before the when computing attributable deaths during the COVID-19 period, which COVID-19 lockdown, considering the normalized mean bias (nMB), the may actually not be the case. For example, if COVID-19 killed the normalized root mean square error (nRMSE) and the Pearson correlation elderly, the population as a whole will become less vulnerable to air coefficient (PCC). Although the predictive skills show some variability pollution, i.e. the RR will decrease. Moreover, “air pollution deaths” and from one city to another (Tables S1 and S2 in Supplementary Material), “COVID-19 deaths” may at least partly overlap. an overall good performance is found for both NO2 and O3, with The changes in mortality are shown for the three distinct phases of nMB/nRMSE/PCC of +5%/28%/0.86 (N = 3463 points) and the lockdown. During the first phase of the lockdown, from 15 March to +6%/22%/0.86 (N = 3625 points), respectively. 29 March, the citizens had to stay at home, except for purchasing essential goods (i.e., food and medicines), working or attending emer­ gencies, and nonessential shops and businesses (including bars and 2.4. Quantification of the changes in mortality attributable to NO2 and O3 restaurants) had to close. Due to the persistent rise in infections and during the lockdown and deconfinement periods hospitalisations, more severe restrictions were adopted by the author­ ities in a second phase of the lockdown from 30 March to 12 April The RR of mortality associated with NO2 and O3 levels in each city (partially coinciding with Easter holidays), during which only essential was used to transform the daily (i) observed and (ii) BAU NO2 and O3 activities such as food trade, healthcare services, and some industries time-series during the lockdown and deconfinement periods into daily were authorised. A third and last phase of the shutdown began on 13 number of deaths attributable respectively to NO2 and O3, following a April, when some nonessential economic sectors, including construction method described elsewhere (Gasparrini and Leone, 2014). Then, the and industry, were allowed to return to work. The deconfinement period change in mortality attributable to NO2 and O3 during the lockdown and lasted from 11 May to 21 June, and it was also characterised by different deconfinement results from the sum of the difference between the stages that, unlike the lockdown period, were not homogeneously number of deaths attributable to the observed NO2 and O3 and the implemented across the country given that the evolution of the epidemic number of deaths attributable to BAU NO2 and O3, respectively. The differed among territories. baseline mortality used to quantify attributable deaths was from the average annual cycle corresponding to the period 2010–2018. We pro­ vide 95% confidence intervals for the number of deaths that account for the uncertainties associated to both the fitted regression model and the Fig. 1. Geographical distribution of study locations and the corresponding overall pollution-mortality association, 2010–2018. The study locations are represented with the corresponding average daily mean pollution level. Error bars are 95% CIs. 3 H. Achebak et al. Environmental Pollution 286 (2021) 117220 3. Results between pollutants and because the data used in this approach are limited to one or a few cities, which does not ensure sufficient statistical We analysed data from 47 major cities representing about 32% of the power to consistently characterise the associations. Note that the city- total Spanish population. During the period 2010–2018, the overall specific estimates of RR of mortality were not statistically significant average concentrations of NO2 and O3 across the 47 cities (Fig. 1) were for O3 (Figs. S3 and S4 in the Supplementary Material), except for 22.0 μg/m3 (SD 13.5) and 55.3 μg/m3 (SD 21.8), respectively. As ex­ Madrid, which is, in turn, the most populous city in Spain and a major pected, the highest concentrations of NO2 are observed in winter and the contributor to the total mortality burden due to air pollution in the lowest ones in summer, whereas the opposite pattern is observed for O3. country (Tables S3 and S4 in the Supplementary Material). The fractions Daily NO2 concentrations are negatively correlated with temperature of mortality attributable to NO2 and O3 over the period 2010–2018 (PCC = − 0.19) and O3 (PCC = − 0.53), but positively correlated with were, respectively, 1.4% (95% CI: 0.4 to 2.5) and 1.9% (95% CI: -0.5 to humidity (PCC = 0.10). By contrast, daily O3 concentrations are posi­ 4.2), which translates into 1737 (95% CI: 451 to 3006) and 2281 (95% tively correlated with temperature (PCC = 0.45) and negatively corre­ CI: -573 to 5056) premature annual deaths across the 47 cities. lated with humidity (PCC = − 0.50). The average mortality from non- Overall, applying the weather normalization technique, we found external causes was 7 (SD 12.2) daily deaths, ranging from 1 in the that the levels of NO2 decreased on average by − 51.0% (range across rural city of Teruel to 74 in the largest city, Madrid. cities from − 65.7to − 30.9) during the lockdown, and by − 36.4% (range Fig. 1 shows the pooled RR values of death at 10 μg/m3 of NO2 and − 53.7 to − 11.6) during the deconfinement, which is equivalent to re­ O3 over lags 0–3 days. Significant RR values were observed at lag 1 for ductions of − 9.8 μg/m3 (range − 18.3 to − 3.3) and − 6.0 μg/m3 (range NO2 (1.0047 [95% CI: 1.0014 to 1.0081]) and at lag 0 for O3 (1.0039 − 16.9 to − 1.3), respectively (Fig. 2a, and Table S5 in the Supplementary [1.0013 to 1.0065]). These risk estimates were found to be robust across Material). These changes in air pollutants are consistent with the recent different modelling choices related to the control of the temperature and literature (Petetin et al., 2020; Barré et al., 2020; Grange et al., 2020; the humidity, but some degree of sensitivity was found if the model was Ordóñez et al., 2020). At least part of the spatial variability in NO2 re­ adjusted for other pollutants (Figs. S1 and S2 in the Supplementary ductions can be attributed to the differences in city size, which implies Material). However, results from two-pollutant models should be differences in the relative contribution of road transport to total NOX interpreted with particular caution because of multicollinearity issues emissions, which is typically higher in larger cities (Petetin et al., 2020; Fig. 2. Air pollution levels and attributable mortality during the lockdown and deconfinement periods in Spain. OBS = Observed; BAU = Business-as-usual; DIF = Difference (OBS-BAU). Error bars are 95% CIs. Solid error bars represent CIs taking into account both the uncertainties of the fitted regression model and of the estimated BAU levels. For reference, dashed error bars represent CIs without considering the uncertainty in BAU levels. 4 H. Achebak et al. Environmental Pollution 286 (2021) 117220 Ordóñez et al., 2020). The strongest reductions in NO2 occurred during [95% CI: 1.0059 to 1.0085] for NO2, and 1.0043 [95% CI: the second phase of the lockdown, when the most stringent measures to 1.0034–1.0052] for O3) and a multilocation analysis in 398 cities (Meng lower the transmission of the virus were in place. The reduction in NO2 et al., 2021) for NO2 (1.0046 [95% CI: 1.0036 to 1.0057]). concentrations was observed in all the cities included in the study During the lockdown and deconfinement periods we found a strong (Table S6 in the Supplementary Material), and it was associated with a reduction in NO2 levels (− 45% on average) that was associated with a significant decrease in the mortality attributable to NO2 of − 119 (95% significant decrease in the number of deaths, although some of these CI: -273 to − 24) deaths over the lockdown (57 days), and of − 53 (− 114 deaths may have occurred anyway a few days, weeks or months later to − 10) deaths over the deconfinement (42 days) (Fig. 2b, and Table S7 among the frailest individuals. This reduction in mortality from the NO2 in the Supplementary Material). drop was however partially compensated by a small increase in the By contrast, the changes in O3 levels were much lower and quite number of deaths associated with the rise in O3 levels during the ana­ variable across cities, including both increases and decreases during the lysed period in the most populated cities (Madrid, Barcelona), despite analysed periods (Table S8 in the Supplementary Material). The stron­ the average reduction of − 0.5% across all the cities. Our study shows gest relative changes in O3 occurred during the most stringent lockdown that trade-offs between multiple air pollutants should be taken into ac­ period (first half of April) (Fig. 2c, and Table S9 in the Supplementary count when evaluating the health impacts of environmental exposures. Material), when O3 was found to substantially increase, although only in The overall positive impact on mortality from lower NO2 levels will a subset of cities, including Madrid and Barcelona. On average, O3 likely continue, although to a lesser extent, as lockdown measures are decreased by − 1.1% (inter-city range of − 20.2 to 23.8) during the imposed again due to new peaks of the epidemic and/or due to the lockdown and increased by 0.6% (range − 12.4 to 23.0) during the foreseen slowdown of the economy and the levels of social and industrial deconfinement, which is equivalent to − 0.7 μg/m3 (range − 12.1 to activity (e.g. transport). While NO2 reductions will likely not be as 11.5) and 0.4 μg/m3 (range − 8.9 to 10.5), respectively (Fig. 2c, and intense, their duration could be relatively long. Anticipating the Table S9 in the Supplementary Material). These changes in O3 concen­ magnitude and even the sign of response of O3 is more complex as it trations were associated with an increase in premature mortality of 14 depends strongly not only on the location and chemical sensitivity (− 72 to 99) deaths during the lockdown, and of 8 (− 27 to 50) deaths regime, but also on seasonal weather variations. For example, our study during the deconfinement (Fig. 2d, and Table S10 in the Supplementary period does not include the month of July 2020, when O3 levels are Material). typically at their maximum in Spain. In this study, we focused only on NO2 and O3 because of the higher 4. Discussion availability and density of data, both temporally (during the historical and the COVID-19 period) and spatially (covering almost all the capital We applied a weather normalization technique to quantify the cities of the Spanish provinces), compared to other pollutants such as variation in NO2 and O3 during the lockdown and deconfinement pe­ PM10, PM2.5, SO2 or CO. The number of missing values in the time series riods, and fitted epidemiological models to estimate the corresponding is particularly important when assessing the exposure-response associ­ predicted change in mortality across 47 provincial capital cities in Spain. ation between air pollution and health outcomes. Indeed, one of the Overall, the ML-based weather normalization technique used in this aspects that makes our study unique and robust is that we used data that study to estimate the BAU NO2 and O3 concentrations showed a good is consistent between the historical period (2010–2018), used to performance, despite several potential sources of uncertainty. This in­ describe the daily epidemiological associations, and the COVID-19 cludes (1) the relatively coarse spatial resolution of the ERA5 meteo­ period, for which we predicted the reduction in attributable mortality. rological data that may not always represent adequately the conditions In future studies, we plan to consider air pollution data from sources prevailing at the urban level, (2) the presence of trends that are here other than monitoring stations (i.e., gridded data from satellite obser­ taken into account in a simplified way, and (3) the influence of other vations, simulation models), to extend the present analysis to the entire drivers on pollutant concentrations, especially for O3 (e.g. deposition, geographical domain of Spain, thus including both rural and urban long-range transport, inter-annual variability of background O3 levels). areas. By fitting epidemiological models in each city, we found that, on average, an increase of 10 μg/m3 in daily NO2 and O3 were associated 5. Conclusion with a RR of death of 1.0047 (95% CI: 1.0014 to 1.0081) and 1.0039 (1.0013–1.0065), respectively. The magnitude of the association for Our findings provide useful insight on the potential short-term health NO2 is much smaller than the one reported in a previous study in Spain benefits of reducing air pollutant emissions from road traffic and eco­ (1.012 [95% CI: 1.010 to 1.014]) (Linares et al., 2018). This difference nomic activities. With permanent reductions in emissions, the positive could be explained by the different modelling approach, as well as the effects could be even greater on the long-term. This may even include possible measurement error in the latter study, arising from the use of air less severe health impacts from epidemics causing respiratory infections pollution data from air quality stations located in the provincial capital such as COVID-19, as the diseases caused by long-term exposure to air cities and mortality data aggregated for the whole provinces. This pollution are in turn risk factors for COVID-19 severity and mortality. mismatch is likely to have biased the exposure-response associations because NO2 shows high spatial variability, and therefore, measure­ Declaration of competing interest ments from air monitoring stations in the provincial capital city might not be representative of the exposure of the whole population in the The authors declare that they have no known competing financial province. The ideal approach would have been to restrict the analysis to interests or personal relationships that could have appeared to influence the capital cities as we did here. On the other side, the estimation ob­ the work reported in this paper. tained for O3 is higher than the one found by Vicedo-Cabrera and col­ leagues for Spain (1.0006 [95% CI: 0.9992 to 1.0019]) (Vicedo-Cabrera Acknowledgments et al., 2020) in a multi-country analysis, which, in turn, could be explained by the differences in the definition of the exposure variable. In HA gratefully acknowledges funding from the Secretariat for Uni­ our study, the RR of mortality is given by the average of the 24 hourly versities and Research of the Ministry of Business and Knowledge of the values of O3 at lag 0, while in the study by Vicedo-Cabrera and col­ Government of Catalonia (grant numbers B00391 [FI-2018], 100180 leagues (Vicedo-Cabrera et al., 2020) is given by the daily maximum of [FI-2019] and B200139 [FI-2020]). HP acknowledges funding from the the 8-h averages at lags 0–1. Our results are more in line with a recent European Union’s Horizon 2020 research and innovation programme systematic review and meta-analysis (Orellano et al., 2020) (1.0072 under grant agreement No H2020-MSCA-COFUND-2016-754433 (Marie 5 H. Achebak et al. Environmental Pollution 286 (2021) 117220 Skłodowska-Curie fellowship). CPG-P acknowledges support by the Achebak, H., Petetin, H., Quijal-Zamorano, M., Bowdalo, D., García-Pando, C.P., Ballester, J., 2020. Reduction in air pollution and attributable mortality due to European Research Council (grant no. 773051, FRAGMENT), the AXA COVID-19 lockdown. Lancet Planet Health 4, e268. Research Fund, and the Spanish Ministry of Science, Innovation and Armstrong, B.G., Gasparrini, A., Tobias, A., 2014. Conditional Poisson models: a flexible Universities (RYC-2015-18690 and CGL2017-88911-R). JB gratefully alternative to conditional logistic case cross-over analysis. BMC Med. Res. Methodol. acknowledges funding from the European Union’s Horizon 2020 14, 122. Barré, J., Petetin, H., Colette, A., et al., 2020. Estimating lockdown induced European research and innovation programme under grant agreement No 865564 NO2 changes. Atmos. Chem. Phys. https://doi.org/10.5194/acp-2020-995. 956396 (MSCA ITN project EDIPI), and from the Ministry of Science and Bauwens, M., Compernolle, S., Stavrakou, T., et al., 2020. Impact of coronavirus Innovation (MCIU) under grant agreements No RYC2018-025446-I outbreak on NO2 pollution assessed using TROPOMI and OMI observations. Geophys. Res. Lett. 47 https://doi.org/10.1029/2020GL087978. (programme Ramón y Cajal) and EUR2019-103822 (project EURO- Chen, K., Wang, M., Huang, C., Kinney, P.L., Anastas, P.T., 2020. Air pollution reduction ADAPT). All authors acknowledge funding from the European Union’s and mortality benefit during the COVID-19 outbreak in China. Lancet Planet Health Horizon 2020 research and innovation programme under grant agree­ 4, e210–e212. Cohen, A.J., Brauer, M., Burnett, R., et al., 2017. Estimates and 25-year trends of the ment No 865564 (European Research Council Consolidator Grant global burden of disease attributable to ambient air pollution: an analysis of data EARLY-ADAPT). MQ-Z and JB acknowledge funding from the European from the Global Burden of Diseases Study 2015. Lancet 389, 1907–1918. Union’s Horizon 2020 research and innovation programme under grant European Environment Agency, 2019. Air Quality in Europe 2019. https://www.eea.eu ropa.eu/publications/air-quality-in-europe-2019. agreement No 727852 (H2020 project Blue-Action). ISGlobal acknowl­ European Environment Agency. Air quality e-reporting database. https://www.eea.euro edges support from the Spanish Ministry of Science and Innovation pa.eu/data-and-maps/data/aqereporting-8. (Accessed 15 November 2020). through the “Centro de Excelencia Severo Ochoa 2019–2023” Program Gasparrini, A., Leone, M., 2014. Attributable risk from distributed lag models. BMC Med. Res. Methodol. 14, 1–8. (CEX2018-000806-S), and support from the Generalitat de Catalunya Gasparrini, A., Armstrong, B., Kenward, M.G., 2010. Distributed lag non-linear models. through the CERCA Program. BSC co-authors acknowledge the BSC-CNS Stat. Med. 29, 2224–2234. “Centro de Excelencia Severo Ochoa 2015–2019” Program (SEV-2015- Giani, P., Castruccio, S., Anav, A., Howard, D., Hu, W., Crippa, P., 2020. Short-term and 0493), and PRACE for awarding us access to Marenostrum Supercom­ long-term health impacts of air pollution reductions from COVID-19 lockdowns in China and Europe: a modelling study. Lancet Planet Health 4, e474–e482. puter at BSC. We acknowledge the E-OBS dataset from the EU-FP6 Grange, S.K., Lee, J.D., Drysdale, W.S., et al., 2020. COVID-19 lockdowns highlight a risk project UERRA (http://www.uerra.eu) and the Copernicus Climate of increasing ozone pollution in European urban areas. Atmos. Chem. Phys. https:// Change Service, and the data providers in the ECA&D project doi.org/10.5194/acp-2020-1171. Guevara, M., Jorba, O., Soret, A., et al., 2021. Time-resolved emission reductions for (https://www.ecad.eu). atmospheric chemistry modelling in Europe during the COVID-19 lockdowns. Atmos. Chem. Phys. 21, 773–797. Appendix A. Supplementary data Isaifan, R.J., 2020. The dramatic impact of Coronavirus outbreak on air quality: has it saved as much as it has killed so far? Glob J Environ Sci Manag 6, 275–288. Kampa, M., Castanas, E., 2008. Human health effects of air pollution. Environ. Pollut. Supplementary data to this article can be found online at https://doi. 151, 362–367. org/10.1016/j.envpol.2021.117220. Kroll, J.H., Heald, C.L., Cappa, C.D., et al., 2020. The complex chemical effects of COVID- 19 shutdowns on air quality. Nat. Chem. 12, 777–779. Linares, C., Falcón, I., Ortiz, C., Díaz, J., 2018. An approach estimating the short-term Author contributions effect of NO2 on daily mortality in Spanish cities. Environ. Int. 116, 18–28. Meng, X., Liu, C., Chen, R., et al., 2021. Short term associations of ambient nitrogen dioxide with daily total, cardiovascular, and respiratory mortality: multilocation HA designed the study, did the epidemiological analysis, interpreted analysis in 398 cities. BMJ 372, n534. the results, and wrote the original draft. HP contributed to the study Nie, D., Shen, F., Wang, J., et al., 2021. Changes of air quality and its associated health design, implemented the machine-learning meteorology-normalization and economic burden in 31 provincial capital cities in China during COVID-19 model for air pollution data and edited the manuscript. MQ-Z contrib­ pandemic. Atmos. Res. 249, 105328. Ordóñez, C., Garrido-Perez, J.M., García-Herrera, R., 2020. Early spring near-surface uted to the study design and processed mortality and temperature data. ozone in Europe during the COVID-19 shutdown: meteorological effects outweigh DB contributed to the study design and processed air pollution data. emission changes. Sci. Total Environ. 747, 141322. CPG-P contributed to the study design and edited the manuscript. JB Orellano, P., Reynoso, J., Quaranta, N., Bardach, A., Ciapponi, A., 2020. Short-term exposure to particulate matter (PM10 and PM2.5), nitrogen dioxide (NO2), and ozone provided the mortality data, contributed to the study design and edited (O3) and all-cause and cause-specific mortality: systematic review and meta-analysis. the manuscript. All authors contributed to the submitted version of the Environ. Int. 142, 105876. manuscript and approved the final version. Petetin, H., Bowdalo, D., Soret, A., et al., 2020. Meteorology-normalized impact of the COVID-19 lockdown upon NO2 pollution in Spain. Atmos. Chem. Phys. 20, 11119–11141. References Royal Netherlands Meteorological Institute (KNMI). European climate assessment & dataset (ECA&D). https://www.ecad.eu/dailydata/. (Accessed 15 November 2020). Achebak, H., Devolder, D., Ballester, J., 2019. Trends in temperature-related age-specific Shah, P.S., Balkhair, T., 2011. Air pollution and birth outcomes: a systematic review. and sex-specific mortality from cardiovascular diseases in Spain: a national time- Environ. Int. 37, 498–516. series analysis. Lancet Planet Health 3, e297–306. Vicedo-Cabrera, A.M., Sera, F., Liu, C., et al., 2020. Short term association between ozone and mortality: global two stage time series study in 406 locations in 20 countries. BMJ 368, m108. 6