90 Days of COVID-19 Social Distancing and Its Impacts on Air Quality and Health in Sao Paulo, Brazil

The COVID-19 pandemic has imposed a unique situation for humanity, reaching up to 5623 deaths in Sao Paulo city during the analyzed period of this study. Due to the measures for social distancing, an improvement of air quality was observed worldwide. In view of this scenario, we investigated the air quality improvement related to PM10, PM2.5, and NO2 concentrations during 90 days of quarantine compared to an equivalent period in 2019. We found a significant drop in air pollution of 45% of PM10, 46% of PM2.5, and 58% of NO2, and using a relative-risk function, we estimated that this significant air quality improvement avoided, respectively, 78, 337, and 387 premature deaths, respectively, and prevented approximately US $720 million on health costs. Moreover, we estimated that 5623 deaths by COVID-19 represent an economic health loss of US $10.5 billion. Both health and economic gains associated with air pollution reductions give a positive perspective of the efforts towards keeping air pollution reduced even after the pandemic, highlighting the importance of improving the strategies of air pollution mitigation actions, as well as the crucial role of adopting efficient measures to protect human health both during and after the COVID-19 global health crisis.


INTRODUCTION
The coronavirus disease  pandemic caused by spreading rapidly a severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has imposed a unique situation for the humanity since the Spanish flu pandemic, transmitted by the H1N1 influenza A virus, which devastated at least 50 million people in 1918. Considered the deadliest pandemic in modern history, its impacts exceed the limit of the 18's decade due to subsequent outbreaks caused by new strains of H1N1 virus, such as the most recent in 2009, formerly known as swine flu (Brida et al., 2020;Tu et al., 2020;Martini et al., 2019;Dhama et al., 2012;WHO, 2010;CDC, 2009;Bradley et al., 2011;Chowell et al., 2006).
Likewise, in the past two decades, the coronaviruses have also shown a continuing potential threat to public health since the emergence of severe acute respiratory syndrome coronavirus (SARS-CoV) and Middle East respiratory syndrome coronavirus (MERS-CoV). SARS-CoV and MERS-CoV can cause severe respiratory distress syndrome in humans and despite the SARS-CoV-2 being highly pathogenic and more deadly to humans, SARS epidemic (2003 -2004) and MERS outbreak (2012) were responsible for the deaths of thousands of people worldwide, as shown in Fig. 1 (Brida et al., 2020;Tu et al., 2020;Martini et al., 2019;Dhama et al., 2012;WHO, 2010;CDC, 2009;Bradley et al., 2011;Chowell et al., 2006). Since its emergence in Wuhan, the capital of Hubei province at the end of 2019, the SARS-CoV-2 has quickly spread worldwide, characterizing a pandemic. On January 30 th , 2020, the Emergency Committee of the World Health Organization (WHO) declared a global health emergency based on the increasing rates of case reporting on different continents (Velavan and Meyer, 2020).
The first case in Latin America was confirmed on February 2020, in Sao Paulo city, Brazil.
The confirmed patient had travelled to Lombardy, in northern Italy, and the same origin was confirmed for the virus, by genome sequencing (Faria, 2020). Since then, an exponential increase of confirmed cases was observed in the following weeks. No specific drugs or vaccines are available, and health systems are overburdened in every city, especially in Sao Paulo city, which reported 124,105 confirmed cases of COVID-19 and 5,623 deaths up to June 14 th , being considered the epicenter of the pandemic in Brazil and in South America.
In addition to the worrying symptoms of pneumonia, acute respiratory distress syndrome (ARDS) and multiple organ dysfunction, which can lead to death, mainly in people considered to be at risk groups, such as the elderly and people with chronic non-communicable diseases and respiratory diseases, (Singhal, 2020;Ahmed et al., 2020;Rothan and Byrareddy, 2020;Lake, 2020), some studies have suggested that is a higher susceptibility of COVID-19 infection of populations exposed to high concentrations of air pollution (Conticini et al., 2020;Frontera et al., 2020;Tavella et al., 2020). Wu et al. (2020) conducted a cross-sectional study in order to investigate whether long-term average exposure to fine particulate matter (PM2.5) is associated with an increased risk of COVID-19 death in the United States. They found that an increase of only 1 μg / m 3 in PM2.5 is associated with an 8 % increase in the COVID-19 death rate (95 % confidence interval: 2 %, 15 %) (Wu et al., 2020). Considering that, in heavily populated and polluted places, the care to prevent the COVID-19 progress should be more severe.
Sao Paulo, one of the largest urban centers of the world, stands out not only for its economic and industrial performance but also its high levels of atmospheric pollutants and greenhouse gases emissions mainly from motor vehicles, responsible for air quality degradation and negative impacts on public health (CETESB, 2018;IBGE, 2019;Andrade et al., 2017).
According to measurements carried out by the Sao Paulo's Environmental Agency, vehicles are responsible for 96.7 % of carbon monoxide (CO), 63.9 % of nitrogen oxides (NOx), and 40 % of particulate matter with less than 10 µm (PM10) emissions (CETESB, 2018).
Considering only the particulate matter with a diameter less than 2.5 microns (PM2.5), 5,012 premature deaths occur per year in Sao Paulo city (Abe and Miraglia, 2016). In addition, Leirião et al. (2020) recently reported the strong influence of diesel-fueled heavy-duty vehicles in air quality. Merely during the Brazilian truck-driver strike in 2018, the reduction in the PM10 concentration resulted in the prevention of between six and eight deaths, which implies between 321 and 442 avoided deaths in a year scenario only in Sao Paulo.
Several factors, mainly of anthropogenic origin, are related to increased risk of disease outbreaks of vector-borne and zoonotic diseases, such as mega-trends in human population growth, ecosystems reduction and fragmentation and land use modification and climate change (Rizzoli et al., 2019;Allen et al., 2017;Guo et al., 2019;Loh et al., 2015). Therefore, the relationship between a pandemic, as the current COVID-19, global environmental change and human health is overly complex and deep.
In this context, this pandemic has also been a great opportunity to discuss the effects of anthropogenic activities on air quality and its implications on public health. Therefore, we aimed to investigate the air quality improvement during the quarantine period, as well as to compare the associated avoided deaths to COVID-19 burden deaths, considering the relative risk and the economic outcomes in Sao Paulo megacity.

METHODS
The study was performed in Sao Paulo (23°32′50′′S -46°38′09′′W), the most populous city in According to these initial events, we chose to investigate the air quality in the city of Sao Paulo, starting on March 16 th and breaking the analysis period into 13 weeks, ending at June 14 th .
As a reference for the comparison, the days in the equivalent period in 2019 (March 16 th -June 14 th , 2019) were selected. This set of days was designated as the "control period." Daily records of precipitation, wind speed and means of air temperature were obtained from Sao Paulo Environmental State Agency Air Quality Information System -QUALAR (CETESB, 2020) and from Brazilian Agro-meteorological Monitoring System (AGRITEMPO, 2020). These meteorological parameters were used as a criterion to exclude the days of the control period that differed from the range of the meteorological conditions of quarantine period.
The pollutants selected for analysis were nitrogen dioxide (NO2), particulate matter with less than 10 μm (PM10) and particulate matter with less than 2.5μm (PM2.5). Hourly atmospheric concentrations data were obtained from records of Sao Paulo Environmental State Agency Air Quality Information System -QUALAR (CETESB, 2020) for both periods. CETESB has 17 automatic monitoring stations in the municipality of Sao Paulo. Not all stations measure all pollutants. Therefore, PM10, PM2.5 and NO2 concentrations in both periods were collected, according to the availability of data from 15 monitoring stations (Fig. 2), and compiled on their respective week, as previously established, to calculate the pollutants' weekly average. The statistical analyses were performed using GraphPad Prism® 5.0 software for analysis and graphing. The data were analyzed for normality by the D'Agostino-Pearson test. As the pollutants' concentrations behave as nonparametric variables, we used the Mann-Whitney test or Kruskal-Wallis test followed by the Dunn's post hoc test to compare the weekly average of quarantine and control periods.
The images of NO2 were detected through the Tropospheric Monitoring Instruments (TROPOMI) on-board ESA's Sentinel-5 satellite. The extraction and processing of the images were performed by Sentinel Hub EO Browser, Google Earth Pro and ImageJ. This latter software was used to convert the images from RGB to 8-bit grayscale and to quantify the mean gray value, as an approach to estimate the NO2 reduction (Lopez et al., 2019;Ciobotaru et al., 2019).
The coefficients were used to estimate the probability of mortality associated to the exposure of the pollutants, named as relative risk (RR), obtained by Eq. (1)

RR = exp [ᵝ (∆x)]
(1) where, RR is the relative risk of all-cause mortality due to air pollution; β exposure-response coefficients related to each pollutant; ΔX is the decrease in the pollutant concentrations (µg / m 3 ), the difference of pollutant concentrations between quarantine period and equivalent period in 2018.
The RR values were used to calculate the attributable fraction (AF), the fraction of deaths attributable to the risk factor of PM10, PM2.5 or NO2 variation, defined in Eq.
(2), in order to estimate the avoided daily mortality during the quarantine period, based on the mean mortality on control period.
The daily mean of the all-cause mortality on control period were calculated and multiplied by the attributable factors in order to estimate the all-cause mortality avoided per day during the quarantine. The economic valuation of health outcomes was performed by the value of statistical life (VSL) estimation, a well-known approach (Alberini et al., 2006;Ding et al., 2019;Ran et al., 2018;Robinson, 2017;Wolfe et al., 2019). Considering the avoided deaths related to air pollutants' emissions reduction, an economic gain is expected. However, due to the deaths caused by COVID-19, a considerable economic loss is expected.
In order to estimate the economic impact of the quarantine and to calculate the trade-off between mortality and economic costs, data of confirmed deaths caused by COVID-19, collected from governmental records (CVE, 2020) was multiplied by VSL established value. The same calculus was done for avoided deaths related to air pollutants' emissions.
VSL can be defined as a measure of willingness to pay for reducing the risk of dying, thus consists of the monetary value of postponed deaths (Alberini et al., 2006;Wolfe et al., 2019). For each avoided death and deaths caused by COVID-19, the VSL assumed was US$ 1.88 million, previously proposed by OECD (Roy and Braathen, 2017).

Air quality improvement during 90 days of COVID-19 social distancing
As a first-step approach, we verified the daily records of precipitation, wind speed and mean temperature. Comparing both periods, we observed that wind speed and mean temperature presented similar ranges. For precipitation, we detected a discrepancy ( Table 1). Table 1. Summary of the daily records of precipitation, wind speed and mean temperature during quarantine and control periods.

Number of observations (N); Standard deviation (SD).
Considering that precipitation is an important factor that influences the dilution and dispersion of pollutants (Chou et al., 2005;Pellon de Miranda et al., 2004;Ramis and Santos, 2013;Yu, 2012), 12 days of control period, with precipitation above 10 mm, were not considered in this study. After excluding these outlier days, we can observe that the means and ranges of meteorological variables were similar over the weeks of the quarantine and control period ( Table   2). Table 2. Weekly averages of meteorological variables during quarantine and control periods.
Data were expressed as mean ± standard deviation (SD).
Taking into account the months that comprise the analysis period, we observed an upward trend over the months in both years. Despite the significative reductions were only observed for PM10 and PM2.5 for April, we can observe a consistent decrease in 2020 concentrations of all pollutants compared to 2019 values (Fig. 3). However, from this initial analysis, we can conclude that the division of the analysis period on a weekly scale could provide more detailed information about the variation of pollutants.
During 13 weeks of social distancing, we observed that the pollutants showed a gradual increase in concentrations over the weeks, similarly to the monthly analysis. For PM10, the average concentrations ranged between 16.9 (week 2) and 32.38 µg/m 3 (week 13). PM2.5 presented average concentrations varying between 9.05 (week 4) and 17.15 µg/m 3 (week 12). And for NO2, the average concentrations ranged between 17.21 (week 1) and 40.71 µg/m 3 (week 11) ( Fig. 4). The average of social distancing also fluctuated, presenting the lowest index at week 1 (46.5 %) and the highest between week 2 and 4 (55.4; 56.9 and 53.3 %, respectively). In the remaining weeks, the average of social distancing ranged between 51.7 and 48.7 %.
The best air quality indexes for all pollutants were observed at the weeks 2 and 4 with social distancing above 53.3 %; and the worst concentrations were detected at the weeks 12 and 13, with social distancing less than 50 %.
From week 4 on, we also found a slackening in the social distancing accompanied by an increase in the average of pollutants concentrations, quite evident for PM10 and NO2. The remaining weeks recorded at least 2 days, reaching 6 consecutive days at week 13, with less than 50 % of social distancing. Some factors has contributed for the reduction in social distancing, such as popular manifestations against this measure which occurred in several days, differences in federal and local governments recommendations concerning Covid-19 prevention actions, and mostly important the poor and working class who do not have alternatives in this matter, can also explain the reduction in its index and consequently the fluctuation of pollutants' concentrations.
As shown in Fig. 5, the comparison between the PM10 concentrations during quarantine and control periods revealed 6 weeks with a reduction of at least 25 %, the highest reaching -45 % in week 3 (p < 0.001). For PM2.5, we found 8 weeks with a reduction of at least 29 %, reaching -46 % in week 13 (p < 0.001). The air quality improvement in terms of NO2, showed 9 weeks with a reduction of at least 33 %, reaching -58 % in week 3 (p < 0.01).

Associated health economics outcomes
Based on Based on the results presented in Fig. 5, we estimated the potential health and economic benefits related to pollutants' reductions achieved during the analyzed weeks of quarantine. As it can be seen from the Attributable Factors (AF) ranged were 0.2 % -0.8 % for PM10, 0.9 % -3.2 % for PM2.5 and 4 % -6.9 % for NO2. As to be expected, due to NO2 has presented the greatest levels of reduction compared to the control period, its weekly RR and AF values was also higher in relation to other analyzed pollutants.  . 7). Given the monetary value for the observed health outcomes from analyzed improvement of pollutants concentrations by VSL approach, a well-known and widely used method (Markandya et al., 2018;Roy and Braathen, 2017), we applied the economic valuation of deaths caused by COVID-19 and of the avoided deaths due to the reduction on short-term exposure to PM10, PM2.5 and NO2 (Table 4). During the analyzed weeks of quarantine period in Sao Paulo city, the deaths caused by COVID-19 accounted for an economic health loss around US$ 10.5 billion. On the other hand, we can consider a potential economic health benefit in terms of the prevented premature deaths related to the observed improvement in air quality. Thus, we estimated that the expressive number of avoided deaths due to PM10, PM2.5 and NO2 reductions can represent an economy of US$ 1.5 billion. On the one hand, we estimated the economic health loss associated with deaths related to COVID-19 in Sao Paulo city, which was shown exceedingly raised, even before reaching the important goal of flatten the curve of coronavirus infections, that makes us wonder about the efficiency of the applied public policies and the commitment of the strategies of control the pandemic and population's commitment to comply with social distancing. On the other hand, we observed that the epidemic control actions were closely related to the improvement of air quality, due to the major source of air pollution in the region, the emissions from vehicles' exhaust, which were significantly reduced. An impact assessment evaluation can consider the exposure to a multi-pollutant mixture. Adding pollutant-specific effects may be justified when levels of the specific pollutants are clearly not correlated (WHO, 2000). In this research, we estimated the health effects concerning a reduction in PM10, PM2.5 and NO2 concentrations. Our results showed that air quality improvement contributed with a substantial fraction of the avoided deaths associated to a significant monetized health co-benefits, signaling that the governments have the capability to improve air quality through policy measures.

DISCUSSION
The quarantine and cities in shutting down have also caused a supplementary effect, the improvement of environmental quality, particularly in relation to air quality. Satellite images have captured a dramatic reduction in air pollution over China during February 2020, accounting for over 30 % less in NO2 levels, compared to early January when power plants were operating at normal levels. In addition, the study also estimated a reduction of CO2 emissions by 25 % (Isaifan, 2020). Xu et al. (2020) showed that, from January to March 2020, the decrease in NO2 concentration was of great significance and much higher than other pollutants, such as PM10 and PM2.5. Moreover, Anjum (2020) also showed from satellite images a remarkable reduction of air pollution in Italy, France and USA. Our results corroborate these findings since we also showed a significant reduction in NO2 concentration in Sao Paulo city which was greater than the other analyzed pollutants, confirmed by the abrupt decline of 62.8 % in tropospheric NO2 concentrations also observed by satellite images (Fig. 6).
Human activities do not change only the atmosphere with the emission of pollutants, or only the climate system with the emission of greenhouse gases, the consequences are systemic, since they can affect directly human health, socio-economic and political stability. Emissions of air pollutants, related to the exacerbation of prevalence of cardiorespiratory diseases, type 2 diabetes, mental disorders; land use modification associated with deforestation that modifying disease host / vector communities, elevating disease risk and driving wildlife zoonotic diseases emergence; and mega-trends in climate accompanied by extreme weather conditions makes the population becomes increasingly vulnerable. These conditions have resulted in several socioeconomic impacts, such as morbidities, mortality, migration, poverty exacerbation, violent conflict, affecting people of all ages and all nationalities (Rizzoli et al., 2019;Grande et al., 2020;Watts et al., 2018).
Furthermore, in addition to the reduced number of deaths due to air pollution, we should consider that there are other benefits attributable to the reduction in air pollution itself and could also have positive benefits in reducing preventable non communicable diseases (Chen and Bloom, 2019;Neira et al., 2018). In this sense, a more comprehensive analysis should contemplate other health outcomes. In a strongly linked and integrated world, the impacts of this disease go way beyond mortality.
Unemployment and several companies shutting down operations or announcing dismissals, as well as the change in consumption habits are influencing stock markets and contributing to a financial crisis worldwide. Fernandes (2020) analyzed different scenarios of the economic shock posed by the current COVID-19 crisis (and a confidence interval), expressed as a percentage of GDP for each analyzed country. For Brazil, a mild scenario considered a drop of -3.9 % of the GDP.
We found that decrease of PM10, PM2.5 and NO2 concentration levels could respectively prevent 78, 337 and 387 premature deaths, which an economy of US$ 1.5 billion, providing an evidence for policy making by the quantification and monetization of air-pollution-related health effects. This is a tool for planning purposes, once the monetized value of avoided premature mortality typically dominates the calculated benefits of air pollution regulations. In

CONCLUSIONS
Our study showed that during 90 days of COVID-19 social distancing in Sao Paulo city, Brazil, 5,623 deaths occurred due to this new disease estimated in an economic health loss of US$ 10.5 billion. In opposite, we observed a significant air quality improvement. During the analyzed weeks, the decrease of PM2.5, PM10 and NO2 levels respectively reached up 45 %, 46 % and 58 % reduced in comparison to control period. The positive impact on air quality in terms of PM2.5, PM10 and NO2 respectively provided avoided 78, 337 and 387 premature deaths and prevented up US$ 1.5 billion on health costs. These results highlight the importance of continuing to enforce existing air pollution regulations and measures to protect human health both during and after the COVID-19 global health crisis.

ACKNOWLEDGMENTS
The present study was financed by the Coordenação