key: cord-0983463-pt9q9kma
authors: Rodriguez-Villamizar, Laura A.; Belalcazar-Ceron, Luis Carlos; Fernández-Niño, Julián Alfredo; Marín-Pineda, Diana Marcela; Rojas-Sánchez, Oscar Alberto; Acuña-Merchán, Lizbeth Alexandra; Ramirez-Garcia, Nathaly; Mangones-Matos, Sonia Cecilia; Vargas-Gonzalez, Jorge Mario; Herrera-Torres, Julián; Agudelo-Castañeda, Dayana Milena; Jiménez, Juan Gabriel Piñeros; Rojas-Roa, Néstor Y.; Herrera-Galindo, Victor Mauricio
title: Air pollution, sociodemographic and health conditions effects on COVID-19 mortality in Colombia: an ecological study
date: 2020-11-26
journal: Sci Total Environ
DOI: 10.1016/j.scitotenv.2020.144020
sha: 8e0e6e2f7dbf941fed96867d5444ba333241b581
doc_id: 983463
cord_uid: pt9q9kma

Objective The present study aimed to determine the association between chronic exposure to fine particulate matter (PM2.5), sociodemographic aspects, and health conditions with COVID-19 mortality in Colombia. Methods We performed an ecological study using data at the municipality level. We used COVID-19 data obtained from government public reports up to and including July 17th, 2020. We defined PM2.5 long-term exposure as the 2014-2018 average of the estimated concentrations at municipalities obtained from the Copernicus Atmospheric Monitoring Service Reanalysis (CAMSRA) model. We fitted a logit-negative binomial hurdle model for the mortality rate adjusting for sociodemographic and health conditions. Results Estimated mortality rate ratios (MRR) for long-term average PM2.5 were not statistically significant in either of the two components of the hurdle model (i.e., the likelihood of reporting at least one death or the count of fatal cases). We found that having 10% or more of the population over 65 years of age (MRR=3.91 95%CI 2.24-6.81), the poverty index (MRR=1.03 95%CI 1.01-1.05), and the prevalence of hypertension over 6% (MRR=1.32 95%CI1.03-1.68) are the main factors associated with death rate at the municipality level. Having higher hospital beds capacity is inversely correlated to mortality. Conclusions There was no evidence of an association between long-term exposure to PM2.5 and COVID-19 mortality rate at the municipality level in Colombia. Demographics, health system capacity, and social conditions did have evidence of an ecological effect on COVID-19 mortality. The use of model-based estimations of long-term PM2.5 exposure includes an undetermined level of uncertainty in the results, and therefore they should be interpreted as preliminary evidence.

The SARS-CoV-2 is a new coronavirus responsible for the human coronavirus disease 2019 (COVID- 19) initially reported in Wuhan, China, in December 2019. The rapid global spread of COVID-19 made the World Organization of Health (WHO) declare it a public health emergency of international concern (1). Up to and including July 20th 2020, 14 ,530,563 cases and 606,741 deaths have been reported in 188 countries (2) .

Approximately 80% of COVID-19 confirmed cases reported mild to moderate disease, and the average case fatality rate is 4.6%, with a wide variation across countries (3) .

Efforts to determine modifiable factors that could increase transmission, exacerbate symptoms, and increase the risk of COVID-19 mortality remain essential to guide public policies. Individual conditions such as age above 65 years and underlying chronic diseases, including diabetes, hypertension, cardiovascular disease, chronic lung disease, kidney failure, and cancer, have shown to increase the risk of mortality (4) (5) (6) (7) (8) .

Environmental factors have also been explored with evidence of COVID-19 airborne transmission (9-11).

Short-term air pollutant concentrations, specifically particular matter (PM), might contribute to the spread of the pandemic by transporting viruses in aerosols (airborne transmission) to longer distances than the usual involved in close contacts transmitted through droplets (12) . In China, Zhu et al. (13) conducted a time-series study with data of 120 cities during January and February of 2020. They found a positive association between the daily count of confirmed cases and concentrations of fine and coarse PM (PM 2.5 and PM 10 , respectively), ozone (O 3 ), nitrogen dioxide (NO 2 ), and sulfur dioxide J o u r n a l P r e -p r o o f (SO 2 ) the two weeks before (lag 0-14). Similar findings for PM 2.5 and O 3 were reported in Italy by Borro et al. (14) and Zoran et al. (15) in studies analyzing data from 110 Italian provinces between February and March and data from Milan comparing correlations before and beyond lockdown. In the United States, Adhikari et al. (16) also found positive short-term associations between air pollutants and confirmed cases in New York.

These studies suggest a significant relationship between air pollution and COVID -19 infection and PM's potential effect in airborne transmission. The role of PM as a potential carrier of the SARS-CoV-2 was analyzed for the pandemic in Italy. The authors compared PM10 concentrations and events in Lombardy (the region with a higher number of cases and deaths) and Piedemont (located near Lombardy with less affectation) before and during the first peak of the pandemic. The results showed that the cities in the Piedemont region had even more PM10 pollution events than the cities in the Lombardy region, suggesting that short-term concentrations of PM10 do not fully explain the spread and severity of the pandemic (17).

Long-term exposure to atmospheric pollution has been hypothesized as a contributing factor for explaining mortality related to COVID-19. This hypothesis is based on the evidence that chronic exposure to air pollutants is associated with chronic inflammatory response and overexpression of inflammatory cytokines and chemokines (18) and with the development of chronic respiratory and cardiovascular diseases (19, 20) . These factors might increase infected people's susceptibility to SARS-CoV-2 and therefore, might mediate the pathway between chronic exposure to air pollution and COVID-19 mortality (See Supplementary material Figure S1 ).

Italy was the first country affected by the pandemic in Europe, with an outbreak and mortality larger than the one observed in the city of Wuhan. The regions in Northern Italy exhibited the higher mortality rates for COVID-19 coinciding with the regions with higher air pollutant concentrations, suggesting that chronic exposure to air pollution might contribute to SARS-Cov-2 lethality (21) . Fattorini and Regoli (22) assessed the correlation between chronicity of exposure to air pollution and COVID-19 mortality by using a regional distribution of the mean concentration NO 2 , PM 10 , PM 2.5, and O 3 from 2016 to 2019, the number of days per year in which the regulatory limits of PM 10 and O3

were exceeded, and the number of years during the last decade (2010-2019) in which limit value of PM 10 was exceeded for at least 35 days. They found significant correlations between all three measurements, supporting early and preliminary evidence on the role of chronic exposure to air pollution on COVID-19 mortality. These studies provided meaningful results; however, they did not control for potential confounding factors involved in the relationship between chronic exposure to air pollution and COVID-19 mortality. Ecological studies conducted in China and United States controlling for a variety of sociodemographic and health conditions showed that chronic air pollution exposure, mainly to NO 2 , PM 2.5 SO 2 , increase the COVID-19 mortality risk by 11.2% (CI95%: 3.4%-19.5%), 15% (CI95%:5% -25%) and 17.2% (CI95%:0.5%-36.9%), respectively (23) (24) (25) . (28) . Despite the known effect of chronic exposure to air pollution on the burden of cardiovascular and respiratory diseases (29, 30) , its potential effect on COVID-19 mortality has not been fully elucidated, particularly in low-and-middleincome countries. This study aimed to determine the association between chronic exposure to PM 2.5 , and COVID-19 mortality in Colombia, South America, using an ecological approach and controlling for potential socioeconomic and health conditions confounders. The purpose of this study is to provide results to assess the hypothesis of the long-term effect of PM 2.5 on COVID-19 mortality in countries with different socioeconomic contexts and pollution levels.

Colombia is a country located in the extreme north of South America, consisting of 32 departments, 1,122 municipalities. According to the National Administrative Department of Statistics (DANE, for its initials in Spanish), the population of Colombia in 2020 is Ninety-two out of 1122 municipalities measure air quality regularly in Colombia (33).

Large cities such as Bogota, Medellin, Bucaramanga, Cali, and Barranquilla have air quality monitoring networks. Medium-size and smaller cities perform periodic manual measurements that are not readily available. Because of the scarcity of surface measurements in the country, we retrieved PM 2.5 surface concentrations from the Copernicus Atmospheric Monitoring Service CAMS Reanalysis (CAMSRA) and CAMS Near Real-Time (CAMSNRT) for this study. CAMSRA uses four-dimensional variational data assimilation techniques, combining satellite observations with a global scale atmospheric model to produce aerosol concentrations and mixing ratios of several gases at the surface and vertical gridded data (34, 35) . CAMSNRT is evaluated every quarter, and evaluation reports are available at the COPERNICUS website (36) . We downloaded surface CAMSRA concentrations over Colombia for PM 2.5 using the ECMWF WebAPI and the Python script provided at this platform. We retrieved monthly average gridded data at a 0.125-degree resolution from January 2014 to December 2018.

We estimated PM 2.5 concentrations at the centroid of each municipality by using a mathematical interpolation from the nearest four retrieved CAMSRA concentrations.

Additionally, in order to evaluate the responsiveness of CAMS-based estimation of PM 2.5 concentrations, as a support for data validation, we evaluated exposure data for the quarantine period (between March 1 and August 31, 2020) using CAMSNRT.

Total population, population by age groups, and area of residence (urban/rural) were retrieved at the municipality level from the estimation of population 2020 based on the J o u r n a l P r e -p r o o f Colombian census DANE 2018. We obtained cartographic information and maps from the DANE Geoportal public website (37) , and the spatial data were created in ArcGIS 10.6.1® using the projection of Colombia in mode Custom Azimuth Equidistant and Datum WGS 1984. We used the Multidimensional Poverty Index as a socioeconomic ecologic measure at the municipality level. The poverty index ranges from 0 to 100 with higher percentages meaning privation of more indicators and dimensions. The higher the index, the higher the socioeconomic deprivation (38).

We obtained data related to the number of confirmed cases and deaths for COVID-19 and the number of RT-PCR tests to confirm positive cases of infected people from the National Institute of Health (INS) website (www.ins.gov.co). The database includes caseby-case information of report date, diagnosis date, date of first symptoms, department and municipality of origin, age, sex, clinical condition, and death date for fatality cases.

Information about the number of tests was available at the department level. We used the crude period prevalence of arterial hypertension, diabetes mellitus, and chronic kidney disease data from the High-Cost Account created by the Ministry of Health and Social Protection; we calculated prevalences for the period between July 1st, 2018 and June 30th, 2019 for the 1,122 municipalities of Colombia. Hospital beds capacity was measured as intermediate and intensive care beds per 100,000 inhabitants for each municipality, as a surrogate of the health system capacity. We obtained the data from the publicly available national registry of healthcare providers (Registro Especial de J o u r n a l P r e -p r o o f Prestadores de Servicios de Salud -REPS https://prestadores.minsalud.gov.co/habilitacion/).

The Colombian municipalities with at least one confirmed case of COVID-19 constituted the analytic sample. We calculated population-time at risk as the total population multiplied by the number of days since the first symptom for the first confirmed case at each municipality. We computed the mortality rate using the population-time at risk as a denominator. We described the geographic distribution of the deaths counts, and explored its fit to a Poisson distribution, using the variance test (VT) and the O 2 test. Based on the mean and variance of the death counts we rejected the null hypothesis in both tests (p>0.01) finding evidence of overdispersion (range: 0-1402) and the presence of inflated zeros (58.5%).

Considering the high number of zeros and that from an epidemiological point of view, the first death of COVID-19 represents a phase of the pandemic for one specific municipality; we decided to fit a hurdle model regression for the death counts. We interpreted Hurdle models as a two-part model integrated into one model. The first part is typically a binary response model (logit), and the second part is usually a truncated-atzero count model (39) . We fit a logit-negative binomial hurdle model and use the population-time at risk as the "offset" variable in the regression models. We used the continuous long-term average of PM 2.5 as the primary independent variable in the hurdle model. We performed a sensitivity analysis, running models using the PM 2.5 average as J o u r n a l P r e -p r o o f categorized variable and as a modeled variable with restricted cubic splines using three knots. We used the Akaike criterion to compare the models. We adjusted the effect of long-term PM 2.5 using the following confounding variables identified in the directed acyclic diagram (DAG, see supplementary material Figure S1 ): percentage of population 65 years or older, percentage of the urban population, population density, poverty index, hospital beds capacity, number of COVID-19 tests at the department level, and prevalence (percentage) of hypertension, diabetes, and chronic renal failure. These variables were used as covariates. We ran the analysis clustered by department to account for potential correlation in municipalities within the same department. We conducted secondary analyses excluding the capital district of Bogotá, which holds the highest count of deaths, excluding Medellin, the city among three capitals for which CAMSRA underestimates land-based concentrations (see supplementary material Figure S3a ), and excluding municipalities with less than ten confirmed cases. Furthermore, to validate CAMS-based estimations, we analyzed their correlation with surface measurements of PM 2.5 during the study period as well as their responsiveness to the quarantine period (between March and August) in Bogota, Barranquilla, and Medellin. We ran all the analyses using STATA 15.

There were 182,140 confirmed cases and 6,288 confirmed deaths for COVID-19 in Figure 2 shows the visual inspection of the relation between the estimated longterm mean of PM 2.5 and the COVID-19 mortality rate logarithm. The patterns did not follow a linear trend, and increased log of mortality rates for COVID-19 are present at the lowest levels of mean PM 2.5 . Using a binomial approach (having or no having deaths), the relation with mean PM 2.5 did not follow a linear trend but a line with different inflection points (See Supplementary material Figure S2 ). Restricted cubic splines of PM 2.5 with three knots identified those points to be 12.6, 19.3, and 26.6 μg/m 3 .

We present the results of our primary analysis using hurdle models in Table 2 . is the other main factor associated with the death rate at the municipality level (RR=1.32; 95% CI 1.03-1.68). On the other hand, having a higher percentage of urban population and higher hospital beds capacity are negatively correlated to mortality ( Table 2) . Once the municipality reaches at least one COVID-19 death, the main factors associated with the mortality rate are the percentage of urban population and the poverty index, which increases the mortality rate in 2% and 3%, respectively (Table 2) . Also, a significant cluster (department) effect was identified in the data (p<0.01).

We found that secondary analysis exhibits similar results to our primary analysis in terms of no evidence of increased risk of COVID-19 mortality rate associated with an increased long-term average of PM 2.5 at the municipality level (Table 3) . Results similar to our primary analysis were also consistent in our sensitivity analysis using different approaches to model PM 2.5 long-term average exposure (See supplementary material Table S1 ).

We present a comparison between the estimated monthly average PM 2. Figure S3b presents the daily average concentrations of CAMSNRT and surface concentrations during the quarantine period, supporting that CAMSNRT responsiveness to changes in surface levels is adequate, with a tendency to underestimate surface levels at higher PM 2.5 surface concentrations.

Our research presents the first ecologic nationwide study conducted in a developing country assessing the association between COVID-19 mortality and long-term exposure to PM 2.5 . Our results did not find evidence of an association between higher There was no evidence of an association between the long-term average of PM 2.5 and the mortality rate for COVID-19 in crude or adjusted models. Our results contrast with the reports of correlational studies conducted in Italy (21, 22) and ecological studies in China (24) and the United States (23), which found positive associations between PM 2.5 and COVID-19 mortality after adjusting for four and 20 potential confounders, respectively.

These studies supported the hypothesis that the effect of long-term exposure to PM 2.5 on COVID-19 mortality is largely mediated by comorbidities linked to chronic PM-related inflammation (21, 40) . In this regard, it has been proposed that chronic exposure to PM 2.5 causes alveolar ACE-2 receptor overexpression, which may increase viral load in patients exposed to pollutants (41) . Our findings revealed a significant effect of aging and poverty Another possible explanation for our findings is that long-term exposure to PM 2.5 has less impact on biological susceptibility to COVID-19 complications and deaths compared to the effect of other air pollutants such as nitrogen dioxide (NO 2 ). Multipollutant models in Colombia have identified a stronger short-term effect of NO 2 on respiratory and cardiovascular morbidity compared to other pollutants (42) . A country-wide crosssectional study in the United States using multipollutant models for the effect of PM 2.5 , NO 2, and O 3 found a solid positive association between NO 2 and COVID-19 fatality and mortality but did not find significant associations with PM 2.5 and O 3 (25) . The authors discussed that divergent results with the previous US nationwide study (23) are probably due to the use of multi-pollutant models and the adjustment for spatial trends, which might have confounded the findings. Unfortunately, we did not count on reliable NO 2 and O 3 long-term exposure estimations, so we did not assess this effect in multi-pollutant models.

We found an independent and significant effect of the older age, the poverty index, and the prevalence of hypertension (over 6%) associated with the COVID-19 mortality rate.

Several studies reported similar findings related to age and chronic diseases (4) (5) (6) (7) (8) . In Italy Conticini et al. (21) discussed that factors such as the age structure of the affected population, the great differences between the Italian regional health systems, the capacity of intensive care units in the region, and prevention policies adopted by the government had played a major role in the spread of and mortality for SARS-CoV-2, presumably more than long-term air pollution itself. The effect of poverty on COVID-19 mortality is less described in the literature, but it represents a major risk condition in developing countries, probably related to unstable employment and income, lower health literacy, and limited access to preventive health services (43, 44) . A few recent ecological studies in the US at the county level have reported a correlation between COVID-19 mortality rate and some social disparities such as poverty status and non-English speaking households and other ethnic minorities (45, 46) .

The strengths of this study include the use of nationwide public government data at the municipality level and the adjustment for nine sociodemographic and health conditions using a hurdle model. The main limitation of this study is the lack of empirical data for the long-term estimation of PM 2.5 exposure. The estimation of PM 2.5 concentration in this study comes from the CAMSRA model, which has been evaluated using independent measurements available in different world regions at the surface level and in the tropospheric column. These evaluations show that CAMSRA successfully reproduces levels and trends of aerosols and gases (47) . A recent research conducted to evaluate the performance of CAMSRA over the cities of Bogota, Medellin, and Barranquilla for PM 2.5 , CO, and NO 2 concentrations comparing measurements from the air quality J o u r n a l P r e -p r o o f monitoring networks with retrieved CAMSRA concentrations showed that CAMSRA is able to reproduce PM 2.5 levels and trends in these three cities ( Figure S3a) . However, the model largely underestimates NO 2 and CO concentrations (48) . Additionally, we compared the daily average concentrations of CAMSNRT with surface concentrations in the same cities during the quarantine period ( Figure S3b) , and the results also indicate that CAMSNRT adequately reproduces the trends and levels of surface PM 2.5 , with a tendency to underestimate surface levels at higher PM 2.5 concentrations during the dry season.

Although elevated levels of PM 2.5 are observed in urban areas, PM 2.5 distribution in Colombia shows that even medium-size and small municipalities have similar or even higher concentrations of PM 2.5 . This behavior coincides with aerosols' geographical distribution reported in previous studies for Colombia (30, 49) . These studies indicate that biomass burning is a critical source of PM 2.5 in Colombia and that both large and Our study has other limitations. First, the ecological study's nature precludes the extrapolation of inferences from the empirical evaluation of hypotheses based on clusters (i.e., municipalities) to the individual level. Therefore, the absence of a relationship between long-term exposure to PM 2.5 and mortality among patients diagnosed with SARS-CoV-2 should at best be regarded as provisional. Second, in the absence of reliable NO 2 and O 3 long-term exposure estimations, we could not incorporate them into the analysis or evaluate the independent association of PM 2.5 and mortality in the context of multi-pollutant models. Third, mortality data reflect fatal cases among patients with a confirmatory diagnosis of the infection, excluding deaths among undiagnosed individuals (due to low testing rates or unreliable test results) and those occurring outside of hospitals. Systematic differences in municipalities' capability to comprehensively and correctly identify and register deaths attributable to the infection could have biased our estimate of effect. Although this issue could not be directly addressed in the analysis, adjusting for testing rates and hospital beds capacity should have partially corrected for differential readiness of municipal health systems to cope with the epidemic.

There was no evidence of an association between long-term exposure to PM 2.5 and mortality rate for COVID-19 at the municipality level in Colombia. Demographics, health system capacity, and social conditions did show an ecological effect on COVID-19 mortality. The use of model-based data to estimate the long-term PM 2.5 exposure is an 

Hospitalized Patients With 2019 Novel Coronavirus-Infected Pneumonia in Wuhan, China

Clinical predictors of mortality due to COVID-19 based on an analysis of data of 150 patients from Wuhan, China

Hypertension and its severity or mortality in Coronavirus Disease 2019 (COVID-19): a pooled analysis

COVID-19 and diabetes: Knowledge in progress

Clinical course and risk factors for mortality of adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study

Kidney disease is associated with in-hospital death of patients with COVID-19 Kidney In

COVID-19 Outbreak Associated with Air Conditioning in Restaurant

Airborne Transmission Route of COVID-19: Why 2 Meters/6 Feet of Inter-Personal Distance Could Not Be Enough

Airborne transmission of SARS-CoV-2

Identifying airborne transmission as the dominant rout for the spread of COVID-19

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

Evidence-base considerations exploring relations between SARS-CoV-2 pandemic and air polltuion: involvement of PM2.5-mediated up-regulation of the viral receptor ACE-2

Assessing the relationship between surface levels of PM2.5 and PM10 particulate matter impact on COVID-19 in

Short-term effects of ambient ozone, OM2.5, and meteorological factors on COVID-19 confirmed cases and deaths in Queens

First data analysis about possible COVID-19 virus airborne diffusion due to air particulate matter (PM): the case of Lombardy (Italy)

Air pollution and health

Cardiovascular effects of air pollution

Can atmospheric pollution be considered a cofactor in extremely high level of SARS-CoV-2 lethality in Northern Italy?

Role of the chronic air pollution levels in the Covid-19 outbreak risk in Italy

Exposure to air pollution and COVID-19 mortality in the United States: A nationwide cross-sectional study. medRxiv

Association of particulate matter pollution and case fatality rate of COVID-19 in 49 Chinese cities

Urban Air Pollution May Enhance COVID-19 Case-Fatality and Mortality Rates in the United States. medRxiv. 2020. 26. World Health Organization. WHO Coronavirus Disease (COVID-19) Dashboard

South America prepares for the impact of COVID-19

Database on Source Apportionment Studies for Particulate Matter in the Air (PM10 and PM2.5)

Long term exposure to ambient air pollution and incidence of acute coronary events: prospective cohort study and meta-analysis in 11 European cohorts from the ESCAPE Project

Estimating the air quality and health impacts of biomass burning in northern South America using a chemical transport model

Instituto Nacional de Salud (Colombia). Coronavirus (COVID -2019) en Colombia 2020

The CAMS interim reanalysis of carbon monoxide, ozone and aerosol for 2003-2015

European Centre for Medium-Range Weather Forecasts. The new CAMS global reanalysis of atmospheric composition

Validation report of the CAMS near-real time global atmospheric composition service

Medida de la pobreza multidimensional municipal con informacion censal 2020

Regression Analysis of Count Data. Cambridge: Cambridge University Pres

Effects of short-and long-term exposures to particulate matter on inflammatory marker levels in the general population

Severe air pollution links to higher mortality in COVID-19 patients: the "double-hit" hypothesis

Short-term joint effects of ambient air pollutants on emergency department visits for respiratory and circulatory diseases in Colombia

The COVID-19 Pandemic: Effects on Low-and Middle-Income Countries

On the Macroeconomic and Social Impact of the Coronavirus Pandemic in Latin America and the Developing World

Spatial Disparities in Coronavirus Incidence and Mortality in the United States: An Ecological Analysis as of

Social determinants of COVID-19 mortality at the county level. medRxiv

Evaluation of the CAMS global atmospheric trace gas reanalysis 2003-2016 using aircraft campaign observations

Evaluación espacial y temporal de la calidad del aire en Colombia a partir de los datos del servicio de monitoreo atmosférico de Copernicus (CAMS) y monitoreos en superficie

Universidad Nacional de Colombia

Belalcazar-Cerón LC. Spatial and temporal assessment of particulate matter using AOD data from MODIS and surface measurements in the ambient air of Colombia