Outdoor Air Pollutant Concentration and COVID-19 Infection in Wuhan, China

COVID-19 infection, first reported in Wuhan, China in December 2019, has become a global pandemic, causing significantly high infections and mortalities in Italy, the UK, the US, and other parts of the world. Based on the statistics reported by John Hopkins University, 4.7M people worldwide and 84,054 people in China have been confirmed positive and infected with COVID-19, as of 18 May 2020. Motivated by the previous studies which show that the exposures to air pollutants may increase the risk of influenza infection, our study examines if such exposures will also affect Covid-19 infection. To the best of our understanding, we are the first group in the world to rigorously explore the effects of outdoor air pollutant concentrations, meteorological conditions and their interactions, and lockdown interventions, on Covid-19 infection in China. Since the number of confirmed cases is likely to be under-reported due to the lack of testing capacity, the change in confirmed case definition, and the undiscovered and unreported asymptotic cases, we use the rate of change in the daily number of confirmed infection cases instead as our dependent variable. Even if the number of reported infections is under-reported, the rate of change will still accurately reflect the relative change in infection, provided that intra-city mobility, and the change in testing capacity and case definition. Hence, the effects of the lockdown policy and the inter-city and intra-city mobility, and the change in testing capacity and case definition are all taken into account in our statistical modelling. Furthermore, we adopt the generalized linear regression models covering both the Negative Binomial Regression and the Poisson Regression. These two regression models, when combined with different time-lags (to reflect the COVID-19 incubation period and delay due to official confirmation) in air pollutant exposure (PM 2.5 ), are used to fit the COVID-19 infection model. Our statistical study has shown that higher PM 2.5 concentration is significantly correlated with a higher rate of change in the daily number of confirmed infection cases in Wuhan, China ( p < 0.05). We also determine that a higher dew point interacting with a higher PM 2.5 concentration is correlated with a higher rate of change in the daily number of confirmed infection cases, while a higher UV index and a higher PM 2.5 concentration are correlated with a lower rate of change. Furthermore, we find that PM 2.5 concentration eight days ago has the strongest predictive power for COVID-19 Infection. Our study bears significance to the understanding of the effect of air pollutant (PM 2.5 ) on COVID-19 infection, the interaction effects of both the air pollutant concentration (PM 2.5 ) and the meteorological conditions on the rate of change in infection, as well as the insights into whether lockdown should have an effect on COVID-19 infection.

intra-city mobility, and the change in testing capacity and case definition. Hence, the effects of the lockdown policy and the inter-city and intra-city mobility, and the change in testing capacity and case definition are all taken into account in our statistical modelling. Furthermore, we adopt the generalized linear regression models covering both the Negative Binomial Regression and the Poisson Regression. These two regression models, when combined with different time-lags (to reflect the COVID-19 incubation period and delay due to official confirmation) in air pollutant exposure (PM2.5), are used to fit the COVID-19 infection model. Our statistical study has shown that higher PM2.5 concentration is significantly correlated with a higher rate of change in the daily number of confirmed infection cases in Wuhan, China (p < 0.05). We also determine that a higher dew point interacting with a higher PM2.5 concentration is correlated with a higher rate of change in the daily number of confirmed infection cases, while a higher UV index and a higher PM2.5 concentration are correlated with a lower rate of change. Furthermore, we find that PM2. Previous studies on COVID-19 infection have examined a number of key factors, including demographics, meteorology, and lockdown measures, to determine whether they are strongly . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted May 26, 2020. . https://doi.org/10.1101/2020.05.19.20106484 doi: medRxiv preprint correlated with COVID infection. [2][3][4][5] To the best of our understanding, this is the first rigorous study that investigates the effects of outdoor air pollution concentrations and lockdown on Covid-19 infections in China. Previous studies have suggested that the exposures to air pollutants with and/or without interacting with meteorological conditions may increase the risk of influenza infection. [6][7][8] Such relationships have also been observed in SARS and MERS. 9,10 More recently, research studies have suggested that meteorological conditions are associated with the spread of 11 In the US and Europe, the long-term air pollution exposure has been identified as a predictor of COVID-19 mortality. 12,13 Some studies have started hypothesizing that air pollution is a significant attribute to COVID-19 infection in China and Italy. 14-16 A recent study has suggested that air pollution is associated with COVID-19 infection after the lockdown has been

Unit of Analysis and Data Collection
Our study examines COVID-infection and its relation to air pollutants concentration in Wuhan from the period of 1 January to 20 March 2020. This was the period when COVID-19 was first announced officially in China, the lockdown measures were strictly exercised in Wuhan and other . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 26, 2020. . https://doi.org/10.1101/2020.05.19.20106484 doi: medRxiv preprint parts of China, and the number of confirmed cases peaked and dropped subsequently. We examine the relationship between air pollutant concentration (PM2.5) and lockdown vs. COVID-19 Infection in Wuhan. Data is collected on a daily basis at the city level. Our data are sourced from the following websites: Daily confirmed COVID-19 cases are collected from a popular online platform which aggregates the cases reported by the Chinese national/provincial health authorities. 18 The air pollutants concentration data are collected from the Chinese National Environmental Monitoring Center. 19 The meteorological data, including the temperature, the dew point (the temperature to achieve a relative humidity of 100%), 20 the ultraviolet (UV) index, the precipitation (including rain and snow), and the wind speed, are collected from the US National Climatic Data Center and a weather data API owned by Apple, Inc. 21,22 The mobility data, including the inter-and the intra-city movements index, are collected from Baidu, Inc. 23

Statistical Analysis
As suggested in previous studies, the number of confirmed cases is likely to be under-reported due to the lack of testing capacity, the change in confirmed case definition, and the undiscovered and unreported asymptotic cases. 5,24,25 Hence, we adjust the confirmed infection case data as follows: First, since the number of reported infections might not be reliable, we use the rate of change in our statistical analysis, i.e., the rate of change in the daily number of confirmed cases as compared to that of the previous day (Eq (1)), to reflect the relative variation in COVID-19 infection during the study period. Even if the number of reported infections is incorrect (under-reported), the rate of change will still accurately reflect the relative change in infection, provided the trend of underreporting remains the same. However, the rate of change can still be distorted by China's public health interventions such as the lockdown policy and the change in testing capacity. 26 Therefore, the effects of both the lockdown policy and the change in testing capacity are taken into account in our statistical analysis. In addition, the change in confirmed case definition can distort the actual epidemic curve. 25 Hence, in our infection case modelling, we take into account the most significant discontinuity in the infection curve contributed by the change in definition on confirmed case during the study period.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

1) / The Daily Number of Daily Confirmed Infection Cases in Day t-1
Eq (1) Regression models are constructed to examine the relationship between the outdoor air pollution concentration and the COVID-19 Infection. The dependent and independent variables are listed as follows.

COVID-19 Infection:
We use an adjusted dependent variable, namely, the rate of change in the daily number of confirmed infection cases.  Table 1 for the list of regression models). The best-fit model is selected based on Akaike Information Criterion (AIC). 6  . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 26, 2020.

Result
The  Table 2.
. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

(which was not certified by peer review)
The copyright holder for this preprint this version posted May 26, 2020.   2. *p-value < 0.1, **p-value < 0.05, ***p-value < 0.01 The statistical relationship between the air pollutant concentration (PM2.5) and the rate of change in the daily number of confirmed infection cases is statistically significant and positive (p < 0.05).
This suggests that a higher PM2.5 concentration is associated with a higher rate of change in the