Epidemiological evidence on air pollution exposure and COVID-19 events
| First author [ref.] | Study location | Period | Air pollutants exposure | COVID-19 outcome | Findings: quantified results | Comments |
| Short-term exposure | ||||||
| Setti [16] | Italy | 7 Feb to 15 March 2020 | Daily PM10, concentrations higher than the daily limit value (50 µg·m−3) according to the national monitoring system | Daily number of confirmed cases | Positive correlation between the number of cases in each province and the average number of exceedances of PM10 daily limit value (R2=0.98) | Only exceedance data and only one air pollutant (PM10) |
| Bontempi [17] | Italy (Piedmont, Lombardy, 12 cities) | 10 Feb to 27 March 2020 | Daily PM10 concentrations according to the national air quality monitoring system | Daily number of confirmed cases | No evidence of correlations between the presence of high quantities of PM10 and cases on the basis of visual graphs | Lack of statistical test of the correlation |
| Zoran [18] | Italy (Milan) | 1 Jan to 30 April 2020 | Daily average concentrations of O3, NO2 according to the national air quality monitoring system | Daily total number of confirmed cases, new positive cases and total deaths | Positive correlation of O3 (Pearson coefficient=0.64, 0.50, 0.69) but negative correlation of NO2 (Pearson coefficient=−0.55, −0.35, −0.58) with all outcomes | Taking into account humidity, temperature and lockdown (before and after) |
| Zoran [19] | Italy (Milan) | 1 Jan to 30 April 2020 | Daily average concentrations of PM2.5, PM10, daily maximum PM10 according to the national air quality monitoring system and AQI | Daily total number of confirmed cases, new positive cases and total deaths | Positive correlations between daily new cases and daily maximum PM10 (R2=0.51), daily average PM2.5 (R2=0.25) and daily AQI (R2=0.43) | Statistical significance not reported (no p-value) |
| Frontera [20] | Italy | 1 Feb to 31 March 2020 | PM2.5 mean concentration in February (data from Italian Civil Protection Agency) | Total number of cases and deaths | Positive correlation between PM2.5 concentration in February and total number of cases (Pearson coefficient=0.64, p<0.0074) and death numbers (Pearson coefficient=0.53, p<0.032) on 31 March 2020 | No quantification of the correlation and no adjustment for confounders such as population density |
| Conticini [21] | Italy | AQI based on concentration values for up to five key pollutants, including: PM10, PM2.5, O3, SO2 and NO2 | Death rate | Mortality rate in Lombardy and Emilia Romagna (highly polluted by NO2) higher than in the rest of Italy (12% versus 4.5%) | AQI as a proxy of exposure. No statistical tests | |
| Fronza [22] | Europe (47 regional European capitals and 107 major Italian cities) | 10 Feb to 10 April 2020 | Hourly concentrations of PM2.5, PM10, O3 and NH3# | Daily confirmed cases per province and region | Positive correlation between number of cases per million and PM2.5, PM10 and NH3 (0.58≤ r ≤0.68) but negative correlation with O3, (r=−0.44). | Introduction of a binary classifier based on an artificial neural network to explain spatial differences |
| Li [23] | China (Wuhan and Xiao Gan) | 26 Jan to 29 Feb 2020 | AQI and four ambient air pollutants (PM2.5, PM10, NO2 and CO) according to the national air quality monitoring system | Daily number of new cases (incidence) | Incidence correlated with: AQI in both Wuhan (R2=0.13) and Xiao Gan (R2=0.223); PM2.5 and NO2 in both cities (R=0.329 for NO2 in Wuhan; R2=0.117 for PM2.5 in Xiao Gan); PM10 (R2=0.105) | Low values of R2 |
| Jiang [24] | China (Wuhan, Xiao Gan and Huang Gang) | 25 Jan to 29 Feb 2020 | Daily data of eight ambient air pollutants (PM2.5, PM10, SO2, CO, NO2, and 8-h O3) according to the national air quality monitoring system | Daily number of new cases (incidence) | Positive association (RR between PM2.5 1.036 (95% CI 1.032–1.039), 1.059 (1.046–1.072) and 1.144 (1.12–1.169)) and daily incidence in Wuhan, Xiao Gan and Huang Gang | Quantification of the risk |
| Yao [25] | China (63 cities) | 27 Jan to 26 Feb 2020 | Hourly NO2 data according to the national air quality monitoring system | Number of confirmed cases and basic reproduction number (R0) | Positive association of R0 with NO2 in all cities (Chi-squared=10.18, p=0.037) and with 12-day time lag in 11 cities (r>0.51, p<0.005) | Adjustment for temperature and humidity Other confounders that influence R0 (timing of viral introduction, timing of COVID-19 control policies, etc.) not taken into account Only Chi-squared for global test of association |
| Wang [26] | China (72 cities) | 20 Jan to 2 March 2020 | Daily concentrations of PM2.5 and PM10 according to the national air quality monitoring system | Daily confirmed cases | Short-term (lag 7 and 14 days) increase of 10 μg·m−3 in PM2.5 and PM10 associated with daily cases (RR 1.64 (95% CI 1.47–1.82) and 1.47 (1.34–1.61)) | Quantification of the risk controlled for ambient temperature, absolute humidity and migration scale index |
| Zhu [27] | China (120 cities) | 23 Jan to 29 Feb 2020 | Daily concentrations of PM2.5, PM10, SO2, CO, NO2 and O3 according to the national air quality monitoring system | Daily number of confirmed cases | Short-term increase 10-μg·m−3 (lag 0–14) in PM2.5, PM10, NO2 and O3 associated with a 2.24% (95% CI 1.02–3.46), 1.76% (0.89–2.63), 6.94% (2.38–11.51) and 4.76% (1.99–7.52) increase in the daily counts of confirmed cases | Models (GAMs) adjusted for temperature, humidity, wind speed, air pressure and time trend estimating the associations between the moving average (lag 0–7) concentrations of air pollutants Time series analysis allowing to take daily data and lags into account |
| Adhikari [28] | USA (Queens, NY) | 1 March to 20 April 2020 | Daily maximum 8-h O3, daily average PM2.5 according to the national air quality monitoring system | Number of confirmed cases and deaths | Positive association between O3 and cases (10.51% increase (95% CI 7.47–13.63) but negative relationship between PM2.5 and new cases (a one-unit increase in the moving average of PM2.5 associated with a 33.11% (95% CI 31.04–35.22) decrease in the daily new COVID- 19 cases) | Adjusted for meteorological factors, day trends and lagged outcome to account for the potential autocorrelation of the time series of new cases (deaths) |
| Long-term exposure | ||||||
| Fattorini [29] | Italy (regions) | 2010–2019 | Daily data on distribution of NO2, O3, PM2.5 and PM10 and days exceeding regulatory limits during the last 4 years, and during the last decade (2010–2019) with limits exceeded for at least 35 days according to the national air quality monitoring system | Daily number of confirmed cases | Positive correlations in up to 71 provinces between PM2.5, PM10, O3, and NO2 and cases (0.23 ≤R2 ≤0.34) | No adjustments for meteorological factors and population density |
| Wu [30] | USA (all inland counties) | Up to 4 April 2020 | County-level long-term average of PM2.5 between 2000 and 2016 from prediction models using national air quality monitoring system | COVID-19 death rate | A 1 μg·m−3 increase in PM2.5 associated with an 8% increase in the COVID-19 death rate (95% CI 2–15%) | Main analysis adjusted by 20 potential confounding factors including population density, household income, ethnic group and education, median house value, age, sex, BMI, smoking, temperature, relative humidity, number of individuals tested for COVID-19 Possible bias: county-level adjustment factors excluded institutionalised residents |
| Liang [31] | USA (3122 US counties) | 22 Jan 2020 to 29 April 2020 | Long-term (2010–2016) county-level exposures to NO2, PM2.5 and O3 according to the national air quality monitoring system | COVID-19 case-fatality rate and mortality rate | IQR (∼4.6 ppb) increase in NO2 associated with increase of 7.1% (95% CI 1.2–13.4%) and 11.2% (95% CI 3.4–19.5%) in COVID-19 case-fatality and mortality rates No significant association for PM2.5 and O3 | Both single and multipollutant models and controlled for spatial trends and a comprehensive set of potential confounders including state-level test positive rate, county-level healthcare capacity, phase-of-epidemic, population mobility, sociodemographic, socioeconomic status, behaviour risk factors and meteorological factors |
| Travaglio [32] | UK Biobank data sources | 2018 to 2019 | Annual average of daily measurements for NO2, NO and O3 according to the national air quality monitoring system and higher resolution air pollution estimate (<2 km away from self-reported address) | Number of confirmed cases allowing to compute the infectivity rate and deaths | Association between SO2, PM2.5, PM10 and infectivity rate (OR 1.316 (95% CI 1.141–1.521), 1.120 (1.036–1.211) and 1.074 (1.017–1.136)) Smaller association for NO2 association between SO2, NO2, O3 and COVID-19 mortality (OR 1.172 (95% CI 1.005–1.369), 1.200 (1.026–1.414) and 8.503 (2.029–35.626)) | Adjusted for population density and individual-level data from UK Biobank |
| Ogen [33] | Europe (66 administrative regions in four countries: Italy, France, Germany, Spain) | Jan to Feb 2020 | Tropospheric concentrations of NO2 (Sentinel-5P data) taking into account vertical airflow | Number of deaths collected from each country | Data from the Sentinel-5P showed two main NO2 hotspots over Europe: Northern Italy and Madrid metropolitan area, regions in which COVID-19 mortality has been particularly high | Long-term exposure defined as a 2-month period (Jan to Feb 2020) |
| Cole [34] | The Netherlands (355 municipalities) | Up to 5 June 2020 | Annual concentrations of PM2.5, NO2 or SO2, averaged over the period 2015–2019 | Number of cases, hospital admissions and deaths | A 1 μg·m−3 increase in PM2.5 concentrations associated with 9.4 more COVID-19 cases, 3.0 more hospital admissions and 2.3 more deaths | The relationship was observed in rural settings and persisted after controlling for a wide range of explanatory variables and a number of sensitivity and robustness exercises including instrumenting pollution to mitigate potential endogeneity and modelling spatial spill-overs using econometric techniques |
| Pozzer [35] | Worldwide | Up to June 2020 | Chronic exposure to PM2.5 in the years prior to the COVID-19 outbreak estimated on the basis of satellite observations over the year 2019 | Mortality rate ratios attributed to air pollution in the COVID-19 pandemic [34] and the SARS-CoV-1 epidemic [11] | PM2.5 contributes 15% (95% CI 7–33%) to COVID-19 mortality worldwide, 27% (95% CI 13–46%) in East Asia, 19% (95% CI 8–41%) in Europe and 17% (95% CI 6–39%) in North America | Relative risk (or hazard ratio) of excess COVID-19 mortality for USA and SARS-CoV-1 in China (assuming that SARS and COVID-19 mortality are similarly affected by long-term exposure to air pollution) from long-term exposure to air pollution using the exposure–response function of the WHO to estimate the attributable fraction |
COVID-19: coronavirus disease 2019; PMx: particles with a 50% cut-off aerodynamic diameter of x µm; AQI: Air Quality Index; SARS-CoV-1: severe acute respiratory syndrome-coronavirus-1; GAM: generalised additive model; R2: coefficient of determination defined as the proportion of the variance in the dependent variable (COVID-19 outcomes) that is predictable from the independent variable(s); IQR: interquartile range; BMI: body mass index; WHO: World Health Organization. #: according to the ENSEMBLE multi-model that combines the values of other seven models: CHIMERE METNorway, EMEP RIUUK, EURADIM KNMI/TNO, LOTOS-EUROS SMHI, MATCH FMI, SILAM Météo-France and MOCAGE UKMET (www.regional.atmosphere.copernicus.eu).