We conducted a cross-sectional, county-based study in Germany for the first COVID-19 outbreak, March – May 2020, to analyze the association between long-term (2009 – 2019) exposures of nitrogen dioxide (NO2), nitric oxide (NO), ozone (O3), and particulate matter (aerodynamic diameter <10 µm (PM10), aerodynamic diameter < 2.5 µm (PM2.5)) with COVID-19 cases, deaths, incidence, the numbers of occupied ICU beds, occupied mechanical ventilators on the ICU, and COVID-19 mortality.
Ethical approval was obtained from the ethical commission of the Charité (EA2/038/21; head: Prof. Dr. Kaschina). Patient consent was waived, because no individual patient data were collected and data analysis was performed anonymously.
Study area and COVID-19 situation in Germany
Since February 1, 2020, any suspected COVID-19 case had to be reported to the national health authorities, the Robert Koch Institute (RKI). Local public health departments at county level assessed COVID-19 cases and fatalities daily.
Test capacities increased fast within the first weeks and had a sufficient level since March and could be performed throughout Germany if needed. The first prominent COVID-19 case was announced January 27 by the RKI. However, retrospectively, it was noted that many COVID-19 cases had occurred as early as January 1, 2020. On March 4th major events with more than 1,000 participants were forbidden. On March 22 all federal states in Germany announced social distancing, prohibition of gatherings of more than 2 people, with the exception of household members and closure of schools and daycare facilities.
From April 1 the German Interdisciplinary Association for Intensive Care and Emergency Medicine (DIVI) implemented a registry for all ICU beds and mechanical ventilation capacities on ICUs within Germany. The aim was to facilitate patient care in case of insufficient ICU bed capacities. Within the DIVI registry, each hospital reported how many COVID-19 patients needed ICU treatment and mechanical ventilation daily. As of April 16, reporting to the DIVI registry was mandatory for all German hospitals. Due to the centralized information sharing made possible by the DIVI registry, ICU and ventilator capacities never fell short during the outbreak in Germany in Spring 2020.
The first outbreak of COVID-19 cases started on March 1, 2020 and had the highest numbers of new COVID-19 infections registered in the week from March 30 – April 3, with more than 30,000 new cases reported daily. The highest number of COVID-19 patients needing ICU care were reported on April 18, with 2,928 patients. Declining numbers of COVID-19 infections resulted in a cautious opening of playgrounds, zoos, and churches on April 30. Private gatherings with people from another household were allowed from May 6, schools reopened on May 11 and borders to surrounding countries were gradually reopened from May 13, 2020. Slight variations in reopening existed across states in Germany but were generally within a day or two of those dates given here.
COVID-19 data
COVID-19 cases, deaths, and first documented case (date) in each county were obtained from the open source database of the Robert Koch Institute [11]. We included COVID-19 cases and deaths from March 4, 2020, the time point after which large events were prohibited and private meetings were restricted to members of the own household, to avoid the influence of cluster events. Data were included up to May 16, after which lock down restriction began to be lifted. Incidence and mortality rate per 100,000 inhabitants were calculated at county level.
To sub-classify COVID-19 patients with a severe course of the illness, we extracted the number of occupied ICU beds and mechanical ventilators on the ICU from the DIVI registry [12]. The hospital-based reported data were allocated to the appropriate county in Germany. Data were included from April 16, the start of mandatory reporting, until May 16, when lock down restrictions were lifted. Data were calculated per 100,000 inhabitants. For additional information on data pre-processing see the supplemental material, section S1.
Air pollution data
Long-term air pollution data were collected from 2010 through 2019. The preparation of the air pollution data to provide concentrations at the level of county are provided in detail in Caseiro et al. (2021) and described briefly below [13].
Hourly concentrations of NO2, NO, O3, PM2.5, and PM10 at background stations were downloaded from Airbase [14]. Metrics, e.g., annual mean concentration, were calculated at the spatial level of county in Germany, corresponding to the Nomenclature of Territorial Unit for Statistics level 3 (NUTS-3). An averaging strategy which favours more remote stations was used when more than one station was present in the county. To gap-fill where air quality monitoring data was not available, a relationship between Copernicus Atmospheric Monitoring Service (CAMS) global reanalysis data [15] and the monitoring data was developed and used to estimate missing data. See Table 1 for federal state level averages or Caseiro et al. (2021) for access to the complete open-access dataset [13].
Basic population data, socio-economic and health-related parameters
Population density, age (fraction of people older than 64) and sex (fraction of female) distribution were extracted from the open source database of the Federal Statistical Office of Germany [16] for the year 2019 (detailed description see supplement information (SI) Figure S1).
Socio-economic and health data were extracted from a representative epidemiological and health-monitoring survey (November 2014 – July 2015; 24,016 subjects) conducted in Germany by the RKI [17]. Extracted data included the fraction of people suffering hypertension, coronary heart disease (CHD), diabetes, asthma, chronic kidney disease, obesity (body mass index above 30), and daily smoking behavior. Furthermore, socio-economic parameters including the fraction of people born in non-EU states, people with low socio-economic status, and school attendance of 10 years or less were also extracted. These data points were only available at the state level.
Statistics
From 402 counties within Germany, we included 392 counties in our analysis, 10 counties were excluded due to no DIVI data (no reporting hospitals).
We calculated negative binominal models to estimate the association between long-term air pollutant exposure and COVID-19 parameters. We fit single-pollutant and tri-pollutant models to estimate the effect of each pollutant without and with control for co-pollutants. Since we found high correlations between NO2 and NO (Pearson Correlation 0.879, p-value < 0.001) and between PM10 and PM2.5 (Pearson Correlation 0.621, p-value < 0.001) we performed a tri-pollutant model including NO2, O3 and PM2.5 to avoid collinearity.
We adjusted our models by the following parameters: days since first COVID-19 case, age > 64 years, sex distribution, and population density. In the next step, we adjusted our models for the potential health and socio-economic confounders.
To improve the validity of our model we conducted a number of sensitivity analyses, including limiting the analysis to include only counties with modelled air quality data below 20% (i.e. measured data representing over 80%), different periods for incidence and mortality analysis (from January 1 until May 16 2020; from April 16 until May 16 2020), and case fatality rate.
We did not adjust for test capacity or availability of health care services, since shortages in these areas was not an issue in Germany [18].
We did not perform zero-inflated negative binominal models since we had no zeros in COVID-19 cases and only 19 zeros in COVID-19 deaths per county. We had no zeros in ICU beds occupied and 17 zeros for required mechanical ventilation.
Results of the negative binomial models are presented as main effect estimates with 95% confidence intervals. For the count component, the results indicate the change in percentage of COVID-19 cases, deaths, incidence, occupied ICU beds, required mechanical ventilation in ICU and COVID-19 mortality also with a 95% confidence interval. Calculations were performed with SPSS, Version 26 (Copyright IBM, Inc., Chicago, IL 60606, USA).