key: cord-0972880-uykqnipz
authors: Travaglio, Marco; Yu, Yizhou; Popovic, Rebeka; Selley, Liza; Leal, Nuno Santos; Martins, Luis Miguel
title: Links between air pollution and COVID-19 in England()
date: 2020-10-19
journal: Environ Pollut
DOI: 10.1016/j.envpol.2020.115859
sha: 0fb8ab7825ac4dd85cce792901eda929c1aa3be8
doc_id: 972880
cord_uid: uykqnipz

In December 2019, a novel disease, coronavirus disease 19 (COVID-19), emerged in Wuhan, People’s Republic of China. COVID-19 is caused by a novel coronavirus (SARS-CoV-2) presumed to have jumped species from another mammal to humans. This virus has caused a rapidly spreading global pandemic. To date, over 300,000 cases of COVID-19 have been reported in England and over 40,000 patients have died. While progress has been achieved in managing this disease, the factors in addition to age that affect the severity and mortality of COVID-19 have not been clearly identified. Recent studies of COVID-19 in several countries identified links between air pollution and death rates. Here, we explored potential links between major fossil fuel-related air pollutants and SARS-CoV-2 mortality in England. We compared current SARS-CoV-2 cases and deaths from public databases to both regional and subregional air pollution data monitored at multiple sites across England. After controlling for population density, age and median income, we show positive relationships between air pollutant concentrations, particularly nitrogen oxides, and COVID-19 mortality and infectivity. Using detailed UK Biobank data, we further show that PM(2.5) was a major contributor to COVID-19 cases in England, as an increase of 1 m(3) in the long-term average of PM(2.5) was associated with a 12% increase in COVID-19 cases. The relationship between air pollution and COVID-19 withstands variations in the temporal scale of assessments (single-year vs 5-year average) and remains significant after adjusting for socioeconomic, demographic and health-related variables. We conclude that a small increase in air pollution leads to a large increase in the COVID-19 infectivity and mortality rate in England. This study provides a framework to guide both health and emissions policies in countries affected by this pandemic.

shows that in England specifically, ambient nitrogen oxides concentrations exceed these 154 limits in 89% of designated air quality assessment zones (DEFRA, 2019) . In addition, 155

England experienced the highest excess all-cause mortality rate in Europe in the first five 156 months of 2020 compared with 2015-19, making it one of the world's most affected countries 157 by the COVID-19 pandemic, according to recent data (Raleigh, 2020) . These observations 158

indicate that England provides a unique setting in which to examine the link between air 159 pollution and COVID- 19. 160 In this study, we first investigated potential links between regional and subregional variations 161 in air pollution and population-level COVID-19-related deaths and cases in England by 162 employing coarse and fine resolution methods. Next, we addressed these associations 

Our study utilised regional-level, subregional-level and individual-level information to 181 estimate the relationship between air pollution and COVID-19 in England. For our initial 182 regional analysis, the number of patients infected with SARS-CoV-2 in England was 183 obtained from Public Health England (PHE) and analysed according to the following and Southwest England. Region-level data on the cumulative number of SARS-CoV-2-186 related deaths in England was retrieved from the National Health Service (NHS) ( Table 1) . 187

This source provides one of the most comprehensive region-specific records of COVID- 19 pollutants, we restricted our regional analysis to three major air pollutants, namely, nitrogen 217 dioxide, nitrogen oxide and ozone, across the prespecified regions ( reported in µg/m 3 , except for ozone, whose metric is the number of days on which the daily 240 max 8-hr concentration is greater than 120 µg/m 3 . A detailed quality report regarding this 241 We assigned an average of annual pollutant concentrations from the PCM data to each study 261 subject, on the basis of a six-digit postcode. Individual-level data were collected from the UK 262

Biobank on April 26, 2020. This dataset contained information on individuals that tested 263 positive for COVID-19. No COVID-19 test data were available for UKB assessment centers 264 in Scotland and Wales, thus data from these centers were not included. 265 266

Heatmap legends were generated using GraphPad Prism 8 (www.graphpad.com), and regions 268 are labelled with the mapped colour values. 269

In our regional exploratory analysis, we fitted generalised linear models to our data using 272 COVID-19 deaths and cases as the outcomes and nitrogen oxide, nitrogen dioxide and ozone 273 as the exposures of interest, adding the corresponding population density values as a 274 confounding variable. Population density (person/km 2 ) data correspond to subnational mid-275

year population estimates of the resident population in England and excludes visitors or 276 short-term immigrants (< 12 months). We modelled both the number of cases and deaths 277 using negative binomial regression analyses since the response variables are overdispersed 278 count data. We used the same model type for our subregional analysis, adding mean annual 279 earnings and median age as confounding factors. 280

For the UK Biobank models, we fitted a binomial regression model because the response 281 variable, COVID-positive or -negative, is defined as either 0 or 1. 282

Methods for assessing the fit of the model included residual analyses, diagnostic tests, and 283 information criterion fit statistics. The goodness of fit of each regression model was 284 determined using the log-likelihood and Akaike Information Criterion (AIC) statistics. 285

For all models, we calculated the odds or risk ratios and their 95% confidence intervals to 286 quantify the effects of the independent variables on the response variables. The models were 287 built using the MASS package (www.stats.ox.ac.uk/pub/MASS4/) in R. The comparison 288 tables were generated using the Stargazer package (Hlavac, 2018 

We analysed the associations between cumulative numbers of COVID-19 cases and deaths 299 with the concentrations of three major air pollutants recorded between 2018 and 2019, when 300 no COVID-19 cases were reported. Due to differences in data availability for each air 301 pollutant, we only included annual mean values of daily measurements, which was the most 302 consistent aggregation type reported for all air pollutants described in this analysis. We 303 started by analysing publicly available data from seven regions in England (Table 1) England are associated with increased numbers of COVID-19 infections and mortality. We 317 applied a negative binomial regression model to estimate the association between each air 318 pollutant with the cumulative number of both COVID-19 cases and deaths at the regional 319 level (Supplementary Tables 2 and 3 ). The model was chosen based on the data type (count 320 data) and log likelihood and AIC scores (Akaike, 1998) . Population density, a confounding 321 factor, was added to this model as an independent variable to account for differences in the 322 number of inhabitants across regions. The levels of nitrogen oxide and nitrogen dioxide are 323 significant predictors of COVID-19 cases (p < 0.05), independent of the population density 324 (Supplementary Table 2 ). We next applied a similar method to assess the association with the 325 number of COVID-19 deaths (Supplementary Table 3 ). Ozone, nitrogen oxide and nitrogen 326 dioxide levels are significantly associated with COVID-19 deaths, together with the 327 population density. 328

Taken together, the negative binomial regression models of both COVID-19 cases and deaths 329 (Supplementary Tables 2 and 3) show that nitrogen dioxide, nitrogen oxide and ozone levels 330 are significant predictors of COVID-19-related death, after accounting for the population 331 density. This study provides the first evidence that SARS-CoV-2 cases and deaths are 332 associated with regional variations in air pollution across England. 

We next used information from the UK Biobank to further assess whether people exposed to 376 increased pollution levels are more likely to contract SARS-CoV-2 at the individual scale. 377

This resource contains data from more than half a million UK volunteers recorded across 378 multiple years. We obtained COVID-19 data reported by the UK Biobank up to and including 379

April 26, 2020. This dataset contained COVID-19 test results for 1,464 participants, of whom 380 664 were diagnosed as positive for COVID-19. The location of each subject included in the 381 analysis is shown in Figure 4A . Compared to the local authority case model, the UK Biobank 382 analysis provides a higher resolution air pollution estimate (less than 2 km away from their 383 self-reported address) and includes potentially asymptomatic cases. 384

In our model, we accounted for a list of confounding variables (Supplementary Table 1 Figure 4B ). That is, we found that a single unit 392 increase in PM 2.5 levels was associated with a statistically significant 12% increase in 393 COVID-19 cases, regardless of the primary exposure measure (i.e., single year or multiyear 394 exposure). For PM 10 , a one-unit increase was associated with approximately 8% more obtained from the subregional models, where PM was not found to predict the number of 397 cases, which may be related to the lack of individual-level data. Nonetheless, both our 398 subregional and individual-level models suggest that the levels of nitrogen oxides and dioxide 399 were positively associated with COVID-19 infectivity, with an odds ratio of approximately 400 1.03 for both the single-year and multi-year model ( Figure 4B ). Based on our results, we 401 predict that an increase of only 1 g/m3 in the long-term average of nitrogen dioxide levels 402

increased COVID-19 cases by 4.5% [95% CI: 5.99% -3.05%] while a similar increase in 403 nitrogen oxides was associated with approximately 2% more cases [95% CI: 2.92% -1.35%]. 404

Conversely, ozone levels are not significant predictors of infectivity at the individual level, 405 although they were significantly associated with deaths and cases at the subregional level 406 (Figures 3 and 4B ). In addition to air pollution, we observed an association between current 407 smokers and a lower likelihood of COVID-19 positivity than previous and non-smokers. 408

However, according to our model, population density and predisposing health factors, such as 409 age, sex, diabetes and a previous history of cancer and lung problems, are not predictors of 410 the probability of being infected (Supplementary Table 6 Here, we identified associations between air pollution and COVID-19 deaths and cases in 415

England, expanding on previous evidence linking high mortality rates in Europe with 416 increased toxic exposure to air pollutants (Conticini et al., 2020; Ogen, 2020). Air pollution 417 exposure and health impact estimates have been suggested to mainly depend on the resolution 418 at which they are evaluated (Stroh et al., 2007) . Therefore, we first calculated the effects of 419 air pollution on COVID-19 mortality and spread using regional, coarse resolution data, and 420 then high-resolution, individual-level observations obtained from the UK Biobank. By 421 employing finer resolution grids, we found statistically significant evidence that an increase 422 in the long-term average of PM 2.5 is associated with the largest increase in COVID-19 423 infectivity in England. 424 425 According to our initial findings, regional variations in nitrogen oxide and ozone 426 concentrations in England predict the numbers of COVID-19 cases and deaths, independent 427 of the population density. However, overall uncertainties for modelled exposure estimates at 428 the regional scale (Stroh et al., 2007) led us to obtain increased spatial resolution. Using long-term average of nitrogen oxides and dioxide levels was associated with a 1.5% and 2.5% 431 increase in COVID-19 related mortality, respectively. Notably, these findings are consistent 432 with studies conducted during the previous SARS outbreak, where long-term exposure to air 433 pollutants predicted adverse outcomes in patients infected with SARS in China (Cui et al., 434 2003) . Although nitrogen oxides are key ozone precursors, the relationship between these 435 gases and ozone is nonlinear in ozone chemistry (Kelly and Gunst, 1990) . Therefore another study from Northern Europe where levels of PM 2.5 were found to be strongly 491 associated with COVID-19 incidence, after adjusting for multiple demographic and health-492 related confounders (Andree, 2020). However, this report is the first study to employ 493 individual-level data to assess the relationship between air pollution exposure and COVID-494 19, after controlling for individual characteristics such as age and underlying health 495 conditions obtained by the UK Biobank. By modelling air quality estimates based on the 496 nearest available measurements to individuals' residences, our modelling strategy helped to 2007). Furthermore, although considerable anecdotal evidence suggests that air quality is 499 associated with COVID-19 outcome (Conticini et al., 2020; Ogen, 2020; Wu et al., 2020) , 500 most studies to date have been unable to accurately quantify the number of COVID-19 cases 501 due to limited testing capacity. In England, government guidelines have prioritised testing for 502 symptomatic COVID-19 patients, meaning that official figures do not include the growing 503 number of people who are asymptomatic or are self-isolating at home due to mild COVID-19 504 symptoms. In contrast, all UK Biobank participants included in this study were subjected to 505 COVID-19 testing since the beginning of the pandemic. Because a large proportion of 506 COVID-19 infections are asymptomatic (Day, 2020; Nishiura et al., 2020) , the UK Biobank 507 model provides greater sensitivity to the analysis of infection rates compared to ecological 508 models. We suggest that these differences may partly explain the conflicting results for PM 2.5 509 and PM 10 observed between our subregional and individual-level models. 510 511 Despite some notable advantages in inferring the relationship between COVID-19 infectivity 512 and air quality, it is important to acknowledge that our individual-level analysis presented 513 some limitations. First, our assessment of exposure remains inherently limited because the 514 degree to which ambient monitoring stations represent the exposure of the subject is 515 imperfect. For instance, we were unable to assess microclimate differences in exposure or 516 details regarding the subjects' activity and location, such as the time spent in traffic and 517 indoors. Therefore, questions remain concerning the generalisability of the above findings, as 518 microenvironmental (e.g., work, home, school, etc.) and behavioural factors (e.g., mobility) 519 profoundly affect an individual's exposure to air pollution (Ozkaynak et al., 2013) . Though 520 the incorporation of these factors is problematic in the midst of a pandemic, future work 521 should address the confounding effects of additional variables to obtain more accurate PM 522 exposure estimates (Pansini and Fornacca, 2020; Zhu et al., 2020b) . Second, it has become 523 clear over the course of the pandemic that confounding factors in addition to those considered 524 in the current study, such as ethnicity, are also associated with COVID-19 infectivity and 525 mortality rates (Brandt et al., 2020; Dutton, 2020) . A caveat to the use of UK Biobank is the 526 limited representation of ethnic minority groups because the large majority of the participants 527 are of white ethnicity. A recent report by the ONS suggested that the correlation between air 528 pollution and COVID-19 mortality in England becomes weaker once ethnicity is controlled 529 for as a confounding variable (Dutton, 2020) . This finding suggests that either air pollution 530 leads to disproportionate outcomes in ethnic minority groups or that the estimated between the distribution of ethnic groups in England and highly polluted areas. Therefore, 533 our results should be interpreted in the context of our modelling design and future studies 534 should address the relationship between COVID-19 and air pollution after taking into account 535 the confounding effect of ethnicity. 

Clean Air Strategy 2019: executive summary

Information Theory and an Extension of the Maximum Likelihood Principle

Selected Papers of Hirotugu Akaike

Incidence of COVID-19 and Connections with Air Pollution Exposure: Evidence from the 626 Netherlands. medRxiv

Response of human alveolar macrophages to ultrafine, fine, and 628 coarse urban air pollution particles

Valuation of air pollution 630 externalities: comparative assessment of economic damage and emission reduction under COVID-19 lockdown

Air pollution, racial disparities, and COVID-19 mortality

COVID-19: immunopathology and its implications for therapy

Air Pollution Exposure and Covid-19 in Dutch Municipalities

Can atmospheric pollution be considered a co-factor in extremely high level of 638 SARS-CoV-2 lethality in Northern Italy? Environmental Pollution

Evaluating the Sensitivity of

Associations to the Spatial and Temporal Scale of Exposure Assessment

Air pollution and case fatality of SARS in 643 the People's Republic of China: an ecologic study

Covid-19: four fifths of cases are asymptomatic, China figures indicate

Air Pollution in the UK

COVID-19 as a factor influencing air pollution?

Coronavirus (COVID-19) related mortality rates and the effects of air pollution in England. Office for 648 National Statistics

Nitrogen dioxide and mortality: review and meta-analysis of long-term studies

COVID-19: air pollution remains low as people stay at home

The Influence of COVID-19 on Air Quality in India: A Boon or Inutile

Concentrated ambient air particles induce mild pulmonary inflammation in healthy 656 human volunteers

Outdoor air pollution and asthma

The spatial variation of O3, NO, NO2 and 659 NO x and the relation between them in two Swedish cities

The short-term impacts of COVID-19 lockdown on urban air pollution in China

stargazer: Well-Formatted Regression and Summary Statistics Tables

Clinical features of patients infected with 2019 novel coronavirus in Wuhan

Response of ozone to changes in hydrocarbon and nitrogen oxide concentrations in outdoor 668 smog chambers filled with Los Angeles air

Vascular responses to long-and short-term exposure to fine particulate matter: MESA Air 671 (Multi-Ethnic Study of Atherosclerosis and Air Pollution)

COVID-19 and hypertension

The Lancet Commission on pollution and health

Characterizing changes in surface ozone levels in metropolitan and rural 681 areas in the United States for

Urban Air Pollution May Enhance COVID-19 Case-Fatality and Mortality Rates in the United States. medRxiv

Aggregates of ultrafine 686 particles impair phagocytosis of microorganisms by human alveolar macrophages

Long-term analysis of NO, NO2 and O3

Long-term 690 exposure to air pollution and incidence of cardiovascular events in women

COVID-19 pandemic and environmental pollution: A blessing in disguise? Sci 692

Estimation of the asymptomatic ratio of novel coronavirus infections

Assessing nitrogen dioxide (NO2) levels as a contributing factor to coronavirus (COVID-19) fatality. Sci 697 Total Environ 726

Air pollution exposure prediction approaches used in air 699 pollution epidemiology studies

Quantifying the impact of current and future concentrations of air pollutants on respiratory disease risk in England. Environ 702 Health 16

Initial evidence of higher morbidity and mortality due to SARS-CoV-2 in regions with 704 lower air quality. medRxiv

UK's record on pandemic deaths

Deposition rates of viruses and bacteria above the 707 atmospheric boundary layer

The 709 pathophysiology of enhanced susceptibility to murine cytomegalovirus respiratory infection during short-term exposure to 5 710 ppm nitrogen dioxide. The American review of respiratory disease

Is the short term 712 limit value for sulphur dioxide exposure safe? Effects of controlled chamber exposure investigated with bronchoalveolar 713 lavage

Inflammatory Cell Response in Bronchoalveolar Lavage Fluid after Nitrogen-Dioxide Exposure of 716

Healthy-Subjects -a Dose-Response Study

Brake dust exposure exacerbates inflammation and transiently compromises phagocytosis in 719 macrophages

SARS-Cov-2 RNA Found on Particulate Matter of Bergamo in 722 Northern Italy: First Preliminary Evidence. medRxiv

Effect of restricted emissions during COVID-19 on 724 air quality in India

) and the oxidative potential of particulates 727 and diabetes prevalence in a large national health survey

A study of spatial resolution in pollution exposure modelling

UK biobank: an open 732 access resource for identifying the causes of a wide range of complex diseases of middle and old age

Estimates of the severity of coronavirus 738 disease 2019: a model-based analysis

Clinical Characteristics of 138 Hospitalized Patients With 2019 Novel Coronavirus-Infected Pneumonia 741 in Wuhan, China

Relation of long-term exposure to air pollution to 746 brachial artery flow-mediated dilation and reactive hyperemia

OpenSAFELY: factors associated with COVID-19-related hospital death in the linked 751 electronic health records of 17 million adult NHS patients. medRxiv

Exposure to air pollution and COVID-19 mortality in 753 the United States. medRxiv

COVID-19: what has been learned and to be learned about the novel 755 coronavirus disease

A Novel Coronavirus 758 from Patients with Pneumonia in China

Association between short-term exposure to air pollution and COVID-19 760 infection: Evidence from China

We are grateful to all the staff members with critical functions in administration, operations 618 and logistics at the MRC Toxicology Unit during the present crisis. 619 620