Long-term exposure to air pollution and severe COVID-19 in Catalonia: a population-based cohort study

Ambient air pollution is a major contributor to global morbidity and mortality, with well-established effects on cardiovascular and respiratory diseases. However, evidence on how long-term exposure to air pollution influences the incidence or severity of acute respiratory infections, including COVID-19, remains limited. COVID-19 severity is influenced by age, sex, and chronic comorbidities, and mechanistic evidence suggests air pollutants can impair lung defenses and potentially upregulate SARS-CoV-2 receptors in the lung. Early ecological studies reported associations between ambient air pollution and increased COVID-19 hospitalization and mortality, but such designs have important limitations. Individual-level cohort studies are required to provide more robust estimates. Prior individual-level studies have shown positive associations, particularly for PM2.5, but results for NO2 are less consistent and several gaps remain due to limited sample sizes, few events, heterogeneity of estimates, and lack of multipollutant models. This study aimed to address these gaps by analyzing a large, population-based cohort in Catalonia to investigate associations between long-term exposure to PM2.5, NO2, O3, and black carbon (BC) and severe COVID-19 outcomes (hospitalization, ICU admission, death) and hospital length of stay during 2020.

Early ecological studies suggested links between higher ambient air pollution and worse COVID-19 outcomes, but they suffered from ecological fallacy and confounding. Subsequent individual-level cohorts of COVID-19 cases in Ontario (Canada) and California (US) reported positive associations for PM2.5 with hospitalization and mortality, while findings for NO2 were inconsistent or null in several studies. Evidence for BC was scarce and generally null in smaller or selected cohorts. A population-based study from Rome found modest positive associations for long-term PM2.5 and NO2 with COVID-19 mortality. Overall, heterogeneity in outcome definitions, exposure assessment, confounder adjustment, limited sample sizes, small numbers of events, and the rarity of multipollutant models have hindered firm conclusions, motivating a large, general-population analysis with comprehensive modeling.

Study design and population: A population-based cohort (COVAIR-CAT) of adults (≥18 years) residing in Catalonia, Spain, was assembled via deterministic record linkage across public health administrative databases. From 5,127,059 adults covered in 2015, those alive and residing in Catalonia on March 1, 2020, were included (n=4,669,011); exclusions for loss to follow-up, inconsistent dates, missing address or exposure yielded 4,660,502 individuals followed until December 31, 2020. Ethics approval was obtained (CEIM-PS MAR, no. 2020/9610). Data sources: Individual sociodemographics, migration, and vital status came from the Catalan Central Registry of Insured Persons. Comorbidities and healthcare use were obtained from primary care (CMDB-AP), urgent care (CMDB-URG), and acute hospital discharge (CMBD-AH) databases using ICD-9/ICD-10 codes. SARS-CoV-2 RT-qPCR and rapid antigen testing data were from the Catalonia surveillance system (SUVEC). Area-level covariates derived from the 2011 Census and other sources included urbanicity, socioeconomic indices, Gini and deprivation indices, proportion of non-Spanish residents, and distance to the nearest primary care center. A weekly test-positive proportion (TPP) at the healthcare management area level was used as a pandemic indicator. Outcomes: Primary outcome was COVID-19-related hospitalization; secondary outcomes were COVID-19-related ICU admission, death, and hospital length of stay (LOS). COVID-19-related events were those occurring within 30 days of the first COVID-19 diagnosis (laboratory-confirmed or clinically coded), with hospitalizations occurring up to 10 days before diagnosis also considered to capture first-wave testing limitations. Main analyses excluded diagnoses in nursing homes. Exposures: Individual-level annual averages for 2019 were assigned at the residential address for PM2.5, NO2, and warm-season O3 using COVAIR-CAT spatiotemporal models at 250 m resolution (cross-validated R2: 0.61 PM2.5; 0.77 NO2; 0.87 O3) developed using meteorological/monitoring data and machine learning (Random Forest-based selection). Complementary analyses used ELAPSE 2010 land-use regression estimates for PM2.5, NO2, O3, and BC. Due to registry disruptions, 2019 exposures were assigned to the address at the start of 2021 or latest available. Covariates: Age, sex, individual income group (drug co-payment), and an individual health risk group index (multimorbidity/complexity) were included. Smoking status (non-, former, active), body mass index, and specific chronic conditions (e.g., COPD, diabetes, obesity, dyslipidemia, hypertension, cardiovascular disorders) were obtained from clinical databases. Area-level covariates included small-area socioeconomic index, deprivation, Gini, proportion of non-Spanish residents, urbanicity, distance to primary care, and weekly TPP; health region was included as strata. Statistical analysis: Cox proportional hazards models estimated hazard ratios (HR) for associations between interquartile range (IQR) increases in air pollutants and COVID-19-related hospitalization, ICU admission, and death in single- and two-pollutant models, using March 1, 2020 as time origin and censoring at death, 30 days post-diagnosis, out-migration, or end of follow-up. Competing risk of death for hospitalization/ICU outcomes was handled via cause-specific hazards (censoring deaths). Negative binomial regression estimated incidence rate ratios (IRR) for hospital LOS among hospitalized individuals. Sequential adjustment models: Model 1 (age spline, sex), Model 2 (+ smoking, income, health risk group), Model 3 (+ area-level SES, proportion non-Spanish, distance to primary care, urbanicity strata, weekly TPP), Model 4 main (+ health region strata). Sensitivity analyses included adding potential mediators (chronic comorbidities), additional SES indices, multiple imputation for smoking/BMI, restricting to lab-confirmed cases, including nursing home diagnoses, restricting to non-movers, and several ad hoc checks (e.g., alternative censoring date, distance to hospital, population density, missing indicator for smoking, restricting to diagnosed/testing subsets, and cause-specific hospitalizations). Wave-specific analyses stratified by Wave 1 (Mar 1–Jun 20, 2020) and Wave 2 (Jun 21–Dec 31, 2020). Nonlinearity was explored using penalized splines (3 df). Analyses were performed in R (version 4.1.2).

- Cohort: 4,660,502 adults included. In 2020 there were 340,608 COVID-19 diagnoses (64% lab-confirmed), 47,174 hospitalizations (14% of cases), 4,699 ICU admissions (1.4%), and 10,001 deaths (3%); 37% of deaths occurred outside hospital. Median hospital LOS was 7 days (IQR 4–14). - Exposure levels (2019, COVAIR-CAT): PM2.5 mean 13.9 µg/m³ (SD 2.2), NO2 26.2 µg/m³ (SD 10.3), O3 (warm season) 91.6 µg/m³ (SD 8.2). - Single-pollutant main models (per IQR increase; PM2.5=3.2 µg/m³; NO2=16.1 µg/m³): • PM2.5: Hospitalization HR 1.19 (95% CI, 1.16–1.21); ICU HR 1.16 (1.09–1.24); Death HR 1.13 (1.07–1.19); LOS IRR 1.06 (1.04–1.08). • NO2: Hospitalization HR 1.25 (1.22–1.29); ICU HR 1.42 (1.30–1.55); Death HR 1.18 (1.10–1.27); LOS IRR 1.06 (1.03–1.09). • O3: Associations negative in single-pollutant models for severe outcomes. - Two-pollutant models: • NO2 remained positively associated with hospital and ICU admission after adjustment for PM2.5 (e.g., hospitalization HR 1.12, 95% CI 1.08–1.17; ICU HR 1.51, 95% CI 1.33–1.72 in table; variant adjusted for PM2.5 reported). • PM2.5 remained positively associated with hospital admission and LOS after adjustment for NO2 (hospitalization HR 1.12, 95% CI 1.08–1.15; LOS IRR 1.07, 95% CI 1.04–1.10). • When co-adjusted for NO2, O3 associations became null or positive (e.g., ICU HR 1.10, 95% CI 1.02–1.18; Death HR 1.01, 95% CI 0.95–1.07). O3 was positively associated with LOS in two-pollutant models. - Black carbon (ELAPSE 2010): Positive associations with hospitalizations HR 1.19 (95% CI, 1.16–1.22), ICU admissions HR 1.19 (1.10–1.28), deaths HR 1.06 (1.00–1.13), and LOS IRR 1.04 (1.02–1.07). - Wave-specific effects (single-pollutant): Stronger associations in Wave 1 than Wave 2 for hospitalization and ICU admission. For hospitalization HR per IQR: NO2 Wave 1 1.32 (1.27–1.37) vs Wave 2 1.16 (1.11–1.22); PM2.5 Wave 1 1.25 (1.21–1.28) vs Wave 2 1.11 (1.07–1.14). Death associations were positive in both waves, with some variation. - Shape of association: No clear evidence of nonlinearity for PM2.5 or NO2 with hospitalization, ICU, or death within common exposure ranges. - Robustness: Results were broadly consistent across sensitivity analyses, alternative exposure models (COVAIR-CAT 2018; ELAPSE 2010), and outcome definitions, though inclusion of nursing home cases affected death estimates.

The study demonstrates that long-term exposure to PM2.5 and NO2 is positively associated with severe COVID-19 outcomes—including hospitalization, ICU admission, death—and with longer hospital stays in a large, general-population cohort. Multipollutant models confirmed robustness: NO2 retained associations with hospital and ICU admissions after adjusting for PM2.5, while PM2.5 retained associations with hospitalization and LOS after adjusting for NO2. O3 findings were sensitive to adjustment for NO2, likely due to strong negative correlation (r ≈ −0.82), complicating interpretation. Black carbon was also positively associated with severity metrics. Findings are largely consistent with prior cohorts for PM2.5 and help reconcile mixed evidence for NO2, potentially due to improved power, general-population design (capturing both infection risk and progression to severity), refined exposure assessment, and extensive confounder control. Stronger associations during the first wave may reflect greater susceptibility, healthcare system strain, or unmeasured contextual factors early in the pandemic. Potential biological pathways include pollution-induced impairment of respiratory defenses, upregulation of SARS-CoV-2 entry receptors (e.g., ACE2) from particulate exposure, and altered immune responses (e.g., reduced type II interferon and antibody responses). The positive association with hospital LOS underscores air pollution’s contribution to healthcare burden beyond incidence of severe events. Overall, the results support air quality improvements as a strategy to reduce the burden of severe acute respiratory infections.

In a population-based cohort of over 4.6 million adults in Catalonia, higher long-term exposures to PM2.5, NO2, and BC were associated with increased risks of COVID-19-related hospitalization, ICU admission, and death, and with longer hospital length of stay. Associations were robust across multiple sensitivity analyses, exposure models, and outcome definitions, and persisted in multipollutant models. These findings strengthen the evidence that reducing ambient air pollution can mitigate the severity and healthcare burden of COVID-19 and potentially other acute respiratory infections. Future research should elucidate biological mechanisms (including receptor regulation and immune modulation), assess impacts in later pandemic phases with vaccination and variants, refine causal mediation by comorbidities, and evaluate policy interventions targeting pollution reductions and their effects on infectious disease severity.

- Temporal context: Analyses cover 2020 before widespread vaccination and variants of concern; estimates may differ in later phases of the pandemic. - Confounding: Lack of individual-level data on race/ethnicity, migration status, physical activity, and occupation may result in residual confounding despite adjustment for income and area-level indicators. - Outcome misclassification: Use of clinical diagnoses during limited testing periods could introduce misclassification; however, sensitivity analyses restricted to lab-confirmed cases yielded similar results. - Selection considerations: Exclusion of nursing home diagnoses in main analyses enhances general-population applicability but limits generalizability to institutionalized populations; inclusion in sensitivity analyses affected death estimates. - Exposure assignment: 2019 exposures were assigned to the address available at start of 2021 or last known, potentially inducing exposure misclassification due to moves; a non-mover sensitivity analysis was performed. - O3 interpretation: High negative correlation with NO2 complicates interpretation of O3 effects in single- versus two-pollutant models. - Mediation vs adjustment: Health risk index and comorbidity adjustments may overlap with mediating pathways, complicating causal interpretation, though estimates changed minimally when adding comorbidities.