key: cord-279942-y5io9qzl authors: Chakrabarty, Rajan K.; Beeler, Payton; Liu, Pai; Goswami, Spondita; Harvey, Richard D.; Pervez, Shamsh; van Donkelaar, Aaron; Martin, Randall V. title: Ambient PM2.5 exposure and rapid spread of COVID-19 in the United States date: 2020-11-09 journal: Sci Total Environ DOI: 10.1016/j.scitotenv.2020.143391 sha: doc_id: 279942 cord_uid: y5io9qzl It has been posited that populations being exposed to long-term air pollution are more susceptible to COVID-19. Evidence is emerging that long-term exposure to ambient PM2.5 (particulate matter with aerodynamic diameter 2.5 μm or less) associates with higher COVID-19 mortality rates, but whether it also associates with the speed at which the disease is capable of spreading in a population is unknown. Here, we establish the association between long-term exposure to ambient PM2.5 in the United States (US) and COVID-19 basic reproduction ratio R 0– a dimensionless epidemic measure of the rapidity of disease spread through a population. We inferred state-level R 0 values using a state-of-the-art susceptible, exposed, infected, and recovered (SEIR) model initialized with COVID-19 epidemiological data corresponding to the period March 2–April 30. This period was characterized by a rapid surge in COVID-19 cases across the US states, implementation of strict social distancing measures, and a significant drop in outdoor air pollution. We find that an increase of 1 μg/m3 in PM2.5 levels below current national ambient air quality standards associates with an increase of 0.25 in R 0 (95% CI: 0.048–0.447). A 10% increase in secondary inorganic composition, sulfate-nitrate-ammonium, in PM2.5 associates with ≈10% increase in R 0 by 0.22 (95% CI: 0.083–0.352), and presence of black carbon (soot) in the ambient moderates this relationship. We considered several potential confounding factors in our analysis including gaseous air pollutants, and socio-economical and meteorological conditions. Our results underscore two policy implications – first, regulatory standards need to be better guided by exploring the concentration-response relationships near the lower end of the PM2.5 air quality distribution; and second, pollution regulations need to be continually enforced for combustion emissions that largely determine secondary inorganic aerosol formation. In December 2019, a new strain of coronavirus, SARS-CoV-2, began infecting residents of Wuhan Province, China. (Chinazzi et al. 2020; Li et al. 2020b; Wu et al. 2020a; Xu et al. 2020) In the following months, the disease caused by SARS-CoV-2coronavirus disease 2019 (COVID-19)has spread to nearly every country around the globe, and the situation had rapidly evolved into a global pandemic. (Chinazzi et al. 2020; Gilbert et al. 2020) The viral entry of SARS-CoV-2 has been shown to use Angiotensin-converting enzyme 2 (ACE2) as its co-host receptor. (Wan et al. 2020 ) ACE2 plays a crucial role in lung protection by cleaving and converting angiotensin II (Ang II) to the cardioprotective angiotensin 1-7 (Ang 1-7). (Tikellis and Thomas 2012) Clinical observations of COVID-19 patients (Guan et al. 2020 ) suggest a mechanism involving viral loads of SARS-CoV-2 depleting residual ACE2 activity and impairing host defenses. This causes an imbalance between Ang II and Ang 1-7, resulting in high circulating levels of Ang II, which induces pulmonary vasoconstriction, inflammation, and oxidative stress. (Derouiche 2020) Consequently, depleted lungs, typically in the form of acute lung injury or its most serious form, acute respiratory distress syndrome, manifest in patients with COVID-19. (Li et al. 2020a) A recent study by Italian medical researchers posited a causal link between SARS-CoV-2 infection rate and long-term air pollution exposure in a population.(Frontera et al. 2020) They postulated a -double-hit phenomenon‖: patients chronically exposed to high levels of fine particulate matter (PM 2.5 ; particulate matter with aerodynamic diameter 2.5 micrometers or less) present themselves with an overexpression of ACE2, (Lin et al. 2018 ) which readily facilitates penetration of the viral infection. This in turn depletes ACE2 receptors and gives rise to more severe forms of the disease. Subsequently, the spread of the disease accelerates among the population. J o u r n a l P r e -p r o o f Journal Pre-proof Exposure to PM 2.5 has a well-established association with increased risks and severe outcomes during infectious disease outbreaks, including COVID-19, the 2009 H1N1, and the 1918 Spanish influenza pandemics (Clay et al. 2018 (Clay et al. , 2019 Morales et al. 2017; Wu et al. 2020b ) Exposure to PM 2.5 has been causally linked to occurrences of chronic respiratory disease, infectious respiratory disease, asthma, inflammation, and decreased lung function. ) Recent studies have have strongly associated COVID-19 mortality with long-term air pollution exposure in the US , as well as identifying that long-term meteorological and climatic variables play a minor role in comparison to the amount of susceptible population for fundamentally driving the pandemic dynamics (Baker et al. 2020 ). Yet, no study to our knowledge has examined the association between long-term PM 2.5 exposure and the exponentially fast spread of COVID-19 in a population. (Ferguson et al. 2020) The spread of a disease through a population is estimated using the dimensionless epidemiological parameter the basic reproduction ratio. (Griffin et al. 2011; Liu et al. 2020b; Ridenhour et al. 2018) It is defined as the average number of individuals that an infected individual would infect in an entirely susceptible population. Thus, this parameter is of utmost importance for public health officials and policymakers because it indicates the onset of an outbreak based on the threshold value of 1.0. With increasing >1, disease spread in a population becomes more rapid, and it becomes harder to control the outbreak. Here, we establish the association between and long-term ambient PM 2.5 exposure in the United States of America (US), where the impact of the pandemic has been most severe. Using robust and established statistical analysis, we elucidate the role of PM 2.5 mass concentration and composition in COVID-19's rapid spread across the US. We estimate the state-specific values of using an established epidemic progression model involving the susceptible-exposed-J o u r n a l P r e -p r o o f Inference of the COVID-19 basic reproduction ratio. We inferred state-wise values by fitting the prediction of a susceptible-exposed-infected-recovered (SEIR) model (Li et al. 2020c; Liu et al. 2020a ) to confirmed COVID-19 cases. (Dong et al. 2020) Our epidemic model accounts for the age-stratified disease transmissibility (Wallinga et al. 2006) , and possible large-scale J o u r n a l P r e -p r o o f Journal Pre-proof undocumented transmission (Li et al. 2020c ) of COVID-19 in the US. The detailed model structure and parameterization follow that outlined in recent publications (Li et al. 2020c; Liu et al. 2020a) , except that here we neglected the influence of interstate mobility of COVID-19 carriers on long-term disease progression, considering the rapid decline in domestic traffic after the nationwide implementation of stay-at-home orders. The inference timespan was set between March 02 and April 30 (coinciding the nationwide social distancing period), so as to minimize the influences of other human behaviors and social activities on epidemic dynamics. The daily number of confirmed COVID-19 cases was acquired from a real-time epidemic tracking dashboard published by Johns Hopkins University. (Dong et al. 2020) The fitting procedure can be described as such: First, the daily number of state-wise confirmedand-active COVID-19 cases 31 was tracked and recorded, which can be written as a time-series ( ) (subscript denotes each US state, superscript indicates -reported‖, and is time incrementing with a unit of day between March 02 and April 30). Next, the SEIR prediction was fitted with the ground truth using twenty consecutive days of epidemic size data. Specifically, we took twenty consecutive elements from the ( ) time series, generating a -minibatch‖ written as ( ). Here, is defined as an initial date. A total of 41 minibatches can be generated, where increments between 1 and 41. For each minibatch ( ), we initialize the SEIR model with ( ), and then the model guessed a value for and predicted the epidemic trends on the following nineteen days, written as ̂( ). The values of were inferred in a trialand-error manner, by minimizing the root-mean-square-error between ( ) and (1) Here, represents the basic reproduction ratio values inferred for state within the timespan between and ; and respectively denote the unknown documentation ratio and exposure ratio 34 , which were simultaneously inferred along with . Finally, for each state , we calculated a time-averaged reproduction ratio ̅ by taking the arithmetic mean values of the series. where denotes a constant intercept; ( ) denotes a thin-plate basis smooth function; denotes a zero centered Gaussian noise; stands for degrees of freedom, and was chosen because no high-level of nonlinearity was observed in the data trends. The restricted maximum likelihood method was adopted in the optimization to prevent overfitting. The GAM fitting and analysis were conducted using R software and the -mgcv-1.8-31‖ package. The complete inference results for ̅ (time-averaged between March 2 and April 30) and the PM 2.5 composition profiles corresponding to each US state are tabulated in Supplementary Information (Table S1 -S9). Although real-time values of ̅ were always greater than 1 during the two-month investigation period, the parameter's rapidly declining trend (Supplementary Table S1 ) suggests that the implemented non-pharmaceutical intervention strategies-such as school and business closure-were effective in altering the disease dynamics. Figure 1 shows the correlation between state-wise ̅ and the annual average PM 2.5 concentration. A positive correlation between ̅ and PM 2.5 concentration can be observed from all the past 6-year datasets. The correlation between ̅ and long-term average (between years 2000 and 2017) PM 2.5 exposure profile for the entire continental US are shown in Figure 1 (b) and (c). A qualitative comparison of the color-coded continental US maps reveals the overall spatial pattern of ̅ coinciding with that of the PM 2.5 exposure levels. J o u r n a l P r e -p r o o f Figure 3 shows the results of Johnson-Neyman analysis (Carden et al. 2017) , where SNA fraction is the predictor variable and BC concentration is the moderating variable. These results indicate that the direct relationship between SNA fraction and ̅ is dependent on BC concentration. Figure 3(a) shows the value of ̅ as a function of SNA fraction and BC concentration. Figure 3 shows that in areas where a large portion of PM 2.5 is made up of SNA, an increase in BC concentration leads to an increase in ̅ . However, if SNA makes up a smaller portion of PM 2.5 , an increase in BC concentration leads to a decrease in ̅ . Alternatively, as BC concentration increases, the positive relationship between ̅ and SNA fraction becomes more prominent. shows the 95% confidence interval. Given that we find both population density and SNA fraction to be highly correlated with ̅ , a logical argument might arise regarding the relation of ̅ with SNA simply being a compositional surrogate for population density. Especially since SNA aerosol precursors are closely linked to urbanization, one might argue that it is difficult to delineate the effect of SNA fraction and population density on ̅ . We aim to address this concern and strengthen our argument by citing specific examples wherein confounding relationship of ̅ with population density and SNA with population density do not hold valid. A justified comparison would be between Alabama and Louisiana, two states that are similar climatically, socioeconomically, and demographically (see Table S11 ). In comparing these states, we find that their population densities differ by only 1.04%, but Louisiana has 5% higher fraction of SNA. Despite their similarity in population density, ̅ in Louisiana is 20.7% higher than Alabama. This example shows that when comparing two states that differ only in SNA fraction, the positive association between SNA fraction and ̅ still holds. To the best of our knowledge, this is the first nationwide study to estimate the relationship between long-term exposure to PM 2.5 and the rapidity of COVID-19 spread in the US. Up to 6 µg/m 3 PM 2.5 ambient concentrations, which is below current regulatory standards(EPA 2019), we find that a 1.0 μg/m 3 increment in long-term exposure associates with a 0.25 increase in ̅ . Short-term exposure due to outdoor air pollution dropped significantly during the months of providing new evidence of a relationship of PM 2.5 with ̅ at levels below the current NAAQS (Dirgawati et al. 2019; Makar et al. 2017; Pinault et al. 2016) . With respect to PM 2.5 composition, we find a 10% increase in sulfate-nitrate-ammonium (SNA) ionic fractions to be associated with a 0.218 (≈10%) increase in ̅ (95% CI: 0.083 -0.352). This strong association of ̅ with secondary inorganic fraction in PM 2.5 , which could be is relatively straight forward, partitioning of nitrate to sol phase depends on a number of factors including temperature (Kroll et al. 2020) . NO 2 has also been shown to accelerate the oxidation of SO 2 to form sulfates The inherent ability to accurately quantify the number of COVID-19 cases due to limited testing capacity during the March-April timeframe presents another potential limitation. The large uncertainty represented by the shaded area in Fig. 3 suggests that the variation in ̅ is not fully accounted for by the PM 2.5 exposure profile. Caution must be taken when extrapolating results from this study to generalize the virus spread in other countries and drive public health responses. The basic reproduction ratio is a complex property of an epidemic and is highly sensitive to the underlying model used to estimate it, the specific population demographic, and time period of study. (Ridenhour et al. 2018) Future analysis should include COVID-19 epidemic parameters and PM 2.5 exposure data at finergrain levels, e.g. county-level . Research on disproportionate impacts of air pollutionon on at-risk populations, including minority populations and populations with lower socioeconomic status, during the epidemic is another understudied area. Synergistic analysis of air quality and population demographic information suggest that the highest PM 2.5 concentrations in a given area tended to be measured at locations where populations lived or were more likely to below the poverty line and constituted larger percentages of racial and ethnic minorities. Last but not least, the design limitations of this study calls for detailed research on biological mechanisms responsible for the observed associations. J o u r n a l P r e -p r o o f The epidemic dynamic model and statistical analysis codes can be accessed at https://github.com/pliu1991/COVID19PM25_supporting_data_and_files. Susceptible supply limits the role of climate in the early SARS-CoV-2 pandemic The hidden geometry of complex, network-driven contagion phenomena CAHOST: An excel workbook for facilitating the Johnson-Neyman technique for two-way interactions in multiple regression The effect of travel restrictions on the spread of the 2019 novel coronavirus (COVID-19) outbreak Pollution, infectious disease, and mortality: evidence from the 1918 Spanish influenza pandemic What explains cross-city variation in mortality during the 1918 influenza pandemic? Evidence from 438 US cities Evaluating the Sensitivity of PM2. 5-Mortality Associations to the Spatial and Temporal Scale of Exposure Assessment Oxidative Stress Associated with SARS-Cov-2 (COVID-19) Increases the Se-verity of the Lung Disease-A Systematic Review Long-term exposure to low air pollutant concentrations and the relationship with all-cause mortality and stroke in older men An interactive web-based dashboard to track COVID-19 in real time. The Lancet infectious diseases EPA Impact of nonpharmaceutical interventions (NPIs) to reduce COVID-19 mortality and healthcare demand Preparedness and vulnerability of African countries against importations of COVID-19: a modelling study Joint estimation of the basic reproduction number and generation time parameters for infectious disease outbreaks Clinical characteristics of coronavirus disease 2019 in China The complex chemical effects of COVID-19 shutdowns on air quality Acute lung injury in patients with COVID-19 infection Early transmission dynamics in Wuhan, China, of novel coronavirusinfected pneumonia Substantial undocumented infection facilitates the rapid dissemination of novel coronavirus Instillation of particulate matter 2.5 induced acute lung injury and attenuated the injury recovery in ACE2 knockout mice Dynamic interplay between social distancing duration and intensity in reducing COVID-19 US hospitalizations: A -law of diminishing returns‖