Particulate matter air pollution and COVID-19 infection, severity, and mortality: A systematic review and meta-analysis

Ecological evidence links ambient particulate matter ≤2.5 mm (PM2.5) and the rate of COVID-19 infections, severity, and deaths. However, such studies are unable to account for individual-level differences in major confounders like socioeconomic status and often rely on imprecise measures of PM2.5. We conducted a systematic review of case-control and cohort studies, which rely on individual-level data, searching Medline, Embase, and the WHO COVID-19 database up to 30 June 2022. Study quality was evaluated using the Newcastle-Ottawa Scale. Results were pooled with a random effects meta-analysis, with Egger's regression, funnel plots, and leave-one-out/trim-and-fill sensitivity analyses to account for publication bias. N = 18 studies met inclusion criteria. A 10 μg/m3 increase in PM2.5 was associated with 66 % (95 % CI: 1.31–2.11) greater odds of COVID-19 infection (N = 7) and 127 % (95 % CI: 1.41–3.66) odds of severe illness (hospitalisation, ICU admission, or requiring respiratory support) (N = 6). Pooled mortality results (N = 5) indicated increased deaths due to PM2.5 but were non-significant (OR 1.40; 0.94 to 2.10). Most studies were rated “good” quality (14/18 studies), though there were numerous methodological issues; few used individual-level data to adjust for socioeconomic status (4/18 studies), instead using area-based indicators (11/18 studies) or no such adjustments (3/18 studies). Most severity (9/10 studies) and mortality studies (5/6 studies) were based on people already diagnosed COVID-19, potentially introducing collider bias. There was evidence of publication bias in studies of infection (p = 0.012) but not severity (p = 0.132) or mortality (p = 0.100). While methodological limits and evidence of bias require cautious interpretation of the findings, we found compelling evidence that PM2.5 increases the risk of COVID-19 infection and severe disease, and weaker evidence of an increase in mortality risk.


Introduction
A considerable body of ecological evidence suggests fine particulate matter air pollution ≤2.5 μm (PM 2.5 ) increases the risk of COVID-19 infection, severity, and death (Katoto et al., 2021;Copat et al., 2020;Bourdrel et al., 2021). However, this evidence relies on comparisons of geographic units, which do not account for individual-level differences and often misclassify exposures due to poor precision/resolution in PM 2.5 estimates (Bourdrel et al., 2021). Associations between PM 2.5 and COVID-19 may therefore be spurious, confounded by socioeconomic differences that influence exposure to air pollution and COVID-19 risks (Blakely et al., 2004).
Nevertheless, there are several reasons to suspect that PM 2.5 increases COVID-19 risks. PM 2.5 increases expression of Angiotensin-Converting Enzyme 2 (ACE2), which the COVID-19 spike protein uses to bind to and enter host cells (Bourdrel et al., 2021;Rebuli et al., 2021). Though there is limited evidence for ambient PM 2.5 , studies of cigarette smoking suggest it inhibits cell defence against infections (Rebuli et al., 2021). Exsmokers have higher expression of ACE2 receptors than non-smokers (Suzuki et al., 2023;Smith et al., 2020), which suggests that high PM 2.5 exposures may increase long-term COVID-19 vulnerability. PM 2.5 and COVID-19 may also independently worsen cardiopulmonary health, and their compounded effects may increase the likelihood of severe disease and death (Bourdrel et al., 2021;Landrigan et al., 2018).
This systematic review builds on previous reviews (Katoto et al., 2021;Copat et al., 2020;Bourdrel et al., 2021) by focusing on studies using individual-level data that can provide more precise exposure estimates and better account for confounders. We address the following questions: 1. Is ambient PM 2.5 exposure predictive of COVID-19 infection, severe COVID-19 disease, and COVID-19 mortality? 2. Do discrete high PM 2.5 events like wildfires predict COVID-19 infection, severe COVID-19 disease, and COVID-19 mortality?

Methods
This review is registered on PROSPERO (Sheppard et al., 2022) and is reported according to PRISMA 2020 guidelines (Page et al., 2021). A completed PRISMA checklist is available on a public repository (Gao et al., 2022).

Inclusion/exclusion criteria
To be eligible for inclusion, studies had to analyse individual-level data on the association between PM 2.5 and COVID-19 infection, severity, or mortality using either a case-control or cohort design. Studies needed to present original research in an English-language peer-reviewed journal no later than 30 June 2022.
Studies were ineligible if they used ecological, cross-sectional, caseseries, animal, or in-vitro designs; studies with a mixture of methods that included either case-control or cohort design were considered eligible. Common ecological studies such as time series studies were excluded because PM 2.5 levels are typically lower after a surge in infections due to reduced economic activity (Silva et al., 2022). This could introduce spurious correlations between short-term variations in PM 2.5 and COVID-19 outcomes. Studies using the comparable approach, case-crossover analysis, were excluded for the same reason. Although advanced methods, such as bidirectional or time-stratified controlled time, can make these models less prone to bias, there can still be temporal confounding effects due to the possible dynamic associations between exposure and outcomes (Wu et al., 2021;Braeye and Hens, 2020). Hypothesis, review, editorial, commentary, and opinion pieces were excluded, as were pre-prints and conference presentations. Studies not using PM 2.5 or only examining indoor air pollution or tobacco smoke as the pollutant exposure were excluded.

Search strategy and screening
We searched Medline, Embase and the World Health Organization COVID-19 database using terms listed in the Appendix A. In addition, we screened the reference lists of grey literature and previous systematic reviews on similar topics for studies meeting the inclusion criteria. Two study authors (NS & TL) independently screened abstracts and full-texts for eligibility. Disagreements were resolved between screening authors or, failing that, by a third author (MC).

Data extraction and quality assessment
Two authors (NS & TL) independently extracted data and assessed study quality, and a third author (MC) settled disagreements. Data extraction focused on characteristics of the study sample/population, operationalisation of PM 2.5 measurement, and COVID-19 outcomes. Effect size and direction, coefficient type (e.g., Hazard Ratio, Odds Ratio), and confidence intervals were tabulated.
Quality was assessed using the Newcastle-Ottawa Scale (NOS) (Wells et al., 2013). In the Comparability domain, we awarded two stars for studies controlling for individual-level socioeconomic factors and any additional factor, and one star for those adjusting for individual-level socioeconomic factors without any other factor or adjusting broader socioeconomic proxies (area-level socioeconomic indicator or race/ethnicity) but not individuallevel socioeconomic indicators.
NOS scores were converted to Agency for Health Research and Quality (AHRQ) standards using the rubric in Shamsrizi et al. (2020): • Good quality: 3 or 4 stars in Selection domain AND 1 or 2 stars in Comparability domain AND 2 or 3 stars in Outcome domain • Fair quality: 2 stars in Selection domain AND 1 or 2 stars in Comparability AND 2 or 3 stars in Outcome domain • Poor quality: 0 or 1 star in selection domain OR 0 stars in Comparability domain OR 0 or 1 stars in Outcome domain

Meta-analysis
Results were pooled using a random-effects meta-analysis with the metafor (Viechtbauer, 2010) and metaviz (Kossmeier et al., 2019) packages in R (R Core Team, 2022). A Meta-analysis Of Observational Studies in Epidemiology (MOOSE) (Stroup et al., 2000) checklist, along with meta-analysis code and data, is available in our public repository (Gao et al., 2022). Pooling was limited to studies rated "good" or "fair"; sensitivity analysis included all studies regardless of quality. Assuming inherent variance due to differences in populations and methods, we used random effects models and report the I 2 statistic for heterogeneity. All outcomes were converted to Odds Ratios for synthesis. Egger's regression and funnel plots tested for publication bias. While not specified in the original protocol on PROSPERO, we added trim-and-fill and leave-one-out sensitivity analyses to test the robustness of results.
Where studies reported multiple outcomes, we prioritised the following: lengthiest PM 2.5 measurement; most comprehensive measure of outcomes (e.g., serology and self-reported symptoms rather than one or the other; any escalation of care needs rather than just hospitalisation or ICU admittance); complete rather than restricted samples/populations (e.g., analysis of the entire Ontario population rather than only test-takers in Sundaram et al. (2021)); models adjusting for socioeconomic factors. For the two studies by Mendy et al., we used only the more recent, larger study (Mendy et al., 2021a) since it included all participants from the earlier one (Mendy et al., 2021b). We prioritised continuous PM 2.5 measures; to account for the one study which used only a categorical PM 2.5 measure (Sundaram et al., 2021), we took the estimates between the reference category (2-6 μg/m 3 ) and third group (8-9 μg/m 3 ) as reflective of the effect of a 5 μg/m 3 difference in PM 2.5 . This approach to outcome selection was not specified in the protocol as outcome reporting preferences of studies were unforeseeable.

Search results
Search results are in Fig. 1 below. The initial literature search of Medline, Embase and the WHO COVID-19 database yielded 1442 studies, which was reduced to 18 after screening. One study was excluded even though it met the inclusion criteria because it reported only statistically significant results rather than all results regardless of significance (Rzymski et al., 2022a). A full list of screened studies along with reasons for exclusion is available in our public repository (Gao et al., 2022).

Study characteristics
All 18 included studies used a cohort design and focused on background ambient PM 2.5 ; none were case-control studies. No study investigated discrete, large-scale PM 2.5 exposures, meaning we were unable to address our second research question.
Half the studies used North American data (N = 9), mostly from the US (N = 6), followed by Canada (N = 2) and Mexico (N = 1). The remainder mostly used European data (N = 8), primarily the UK (N = 4), followed by Italy (N = 2), Spain, and Poland (N = 1 each). The last study used Chinese data.

Study quality
Study quality is summarised in Table 1. Most (13 of 18 studies) were rated "good". More detail is available in the Critical appraisal document on our public repository (Gao et al., 2022).
Several methodological limitations in the included studies are worth noting. Only four of the 18 studies included individual-level adjustments for socioeconomic factors (e.g., education, household income, insurance status) (Bozack et al., 2022;Chadeau-Hyam et al., 2020;Elliott et al., 2021;Kogevinas et al., 2021). Of the remaining 14, three did not adjust for any socioeconomic factors (Bergamaschi et al., 2022;Li et al., 2022;Rzymski et al., 2022b). Three of the seven infection studies only included participants with a COVID-19 test (Chadeau-Hyam et al., 2020;Scalsky et al., 2022;Travaglio et al., 2021), while the remainder either used entire cohorts regardless of whether there was a record of a COVID-19 test (Sidell et al., 2022;Veronesi et al., 2022) or conducted analyses of the entire cohort as well as just those tested (Sundaram et al., 2021;Kogevinas et al., 2021). Similarly, all but one study examining severity (Kogevinas et al., 2021) and mortality (Elliott et al., 2021) were limited to cohorts who were diagnosed with COVID-19, while three were restricted to patients hospitalised with COVID-19 (Bozack et al., 2022;Li et al., 2022;Rzymski et al., 2022b). Restricted cohorts present a risk of collider bias, as PM 2.5 exposure could influence both whether an individual sought testing for COVID-19 or was COVID-19 positive, resulting in distorted associations (Griffith et al., 2020).
Other methodological issues were not captured by the NOS tool. Three studies (Sundaram et al., 2021;Chadeau-Hyam et al., 2020;Elliott et al., 2021) included multiple predictors of interest within a single model rather than build models around PM 2.5 as an exposure. As   these are not designed to account for how independent variables may interact (e.g., as mediators or colliders), the statistical associations are less reliable (Westreich and Greenland, 2013). Some studies reported resolutions up to 100m 2 (Kogevinas et al., 2021;Scalsky et al., 2022), others used entire cities (Li et al., 2022) or monitoring stations spaced tens of kilometres apart (Sidell et al., 2022;Chen et al., 2022a). Several did not specify PM 2.5 resolution. The timeframe of PM 2.5 measurement also varied considerably, from just the week prior to inclusion/recruitment (Rzymski et al., 2022b) up to ten years (Mendy et al., 2021a;Mendy et al., 2021b) and nearly two decades (López-Feldman et al., 2021).
All studies using UK data relied on the UK Biobank. While this sample is large at around 500,000 people, it is not considered representative of the UK population due to low participation rates and a skew towards older persons (Keyes and Westreich, 2019;Swanson and The, 2012). Additionally, three used PM 2.5 estimates from 2010 (Chadeau-Hyam et al., 2020;Elliott et al., 2021;Scalsky et al., 2022)a decade oldand all UK Biobank studies based exposure on participant residences at enrolment, which was between 2006 and 2010. This does not account for change of address or the steady decline in PM 2.5 in the interim (Department for Environment Food &amp and Rural Affairs, 2022).
Most studies used single-pollutant models, i.e., PM 2.5 without any other air pollutants (N = 12). The remainder were evenly split between only multi-pollutant models (Sundaram et al., 2021;Chadeau-Hyam et al., 2020;Elliott et al., 2021) and both single and multi-pollutant models (Li et al., 2022;Veronesi et al., 2022;Chen et al., 2022a). Rzymski et al. (2022b) used a dichotomous indicator of PM 2.5 based on whether the mean or maximum exceeded 20 μg/m 3 in the week before admission to hospital with COVID-19. All others used continuous mean PM 2.5 , while Mendy et al. (2021a) also used the maximum.

PM 2.5 exposure and COVID-19 infection
Seven studies examined PM 2.5 and COVID-19 infection, which are summarised in Supplementary Table 1. All were rated "good" quality and reported a significant and positive association. Pooled results indicated a 10 μg/m 3 increase in PM 2.5 was associated with a 66 % increase in the odds of COVID-19 infection (95 % CI: 1.31 to 2.11) (see Fig. 2), with 83 % of the variance attributable to heterogeneity (p < 0.001). Egger's regression suggested publication bias (p = 0.012). Trim-and-fill points could not be applied, though results from leave-one-out sensitivity analysis remained significant with estimates ranging from 1.48 to 1.78 (see Supplementary Figs. 1 and 2).
Associations were not consistent across analyses, though no negative association was identified, i.e., all were positive or null. Kogevinas et al. (2021) did not find an association with infection determined solely by serological tests within the subsample who agreed to take a test (n = 3922). However, there was a significant association within both the serological test subsample and the full sample (n = 9088) when infection was determined by combining serological tests and self-report indicators. The difference may be attributable to limited sensitivity of the serological tests, leading to false-negatives; only 70 % of cases identified through self-reported indicators had detectable COVID-19 antibodies. Sundaram et al. (2021) categorised PM 2.5 exposures into five ordinal categories, which exhibited J-shaped curve with COVID-19 infection; compared to the lowest exposure group (2-6 μg/m 3 ), COVID-19 infection risk Fig. 2. Meta-analysis forest plot of three outcomes: COVID-19 infection (pink), severity (blue), and mortality (orange). Note: OR represents change in odds of outcome associated with every 10 μg/m 3 increase in ambient PM 2.5 exposure. The size of the square represents relative meta-analytic weight of each study. was lower (6-7 μg/m 3 ), similar (7-8 μg/m 3 ), and then increasingly higher in the next two groups (8-9 μg/m 3 and ≥10 μg/m 3 ). This was the case whether the comparisons were between those testing positive for COVID-19 and not testing positive (i.e., testers and non-testers in the Ontario population; N = 14,695,579) or between those testing positive for COVID-19 and those testing negative (i.e., testers only; N = 758,791).
The single-pollutant model that accounted for socioeconomic factors in Veronesi et al. (2022) found COVID-19 infections increased 3.6 % (95 % CI: 1.009-1.075) for every 1 μg/m 3 in PM 2.5 . The single-pollutant model that omitted socioeconomic factors was similar but with a slightly bigger effect (RR: 1.051; 95 % CI: 1.027-1.075), which increased substantially when other air pollutants were added to the model (NO 2 : 1.347, NO: 1.105, O 3 : 1.107). This suggests multi-collinearity between air pollutants, which may bias the association between PM 2.5 and COVID-19 infections. Otherwise, associations remained significantly positive in single and multipollutant models.

PM 2.5 exposure and COVID-19 severity
Nine studies examined PM 2.5 and COVID-19 severity, of which five were rated "good", two "fair", and two "poor". These are summarised in Supplementary Table 2. Mendy et al. (2021a) was the only study not to find a significant association, though in a later study the authors found a significant association when the same participants were included in a substantially larger cohort (n = 1128 versus n = 14,783) and PM 2.5 estimates were updated by a year (Mendy et al., 2021b). Aside from Kogevinas et al. (2021), all cohorts were restricted to those diagnosed or hospitalised with COVID-19. Severity was indicated in numerous ways including hospitalisation (N = 5), ICU admission (N = 3), requiring respiratory support (N = 3), clinical symptomatology (N = 1), oxygen saturation (N = 1), and multiple indicators (N = 1).
Pooled results from N = 6 studies indicate the odds of a severe outcome was 127 % higher (95 % CI: 1.41 to 3.66) for every 10 μm/g 3 increase in PM 2.5 (see Fig. 2). Nearly all the variance in effects was due to heterogeneity (I 2 : 97 %; p < 0.001). There was no detectable publication bias (p = 0.132). Trim-and-fill points slightly attenuated the results (OR: 2.04; 95 % CI: 1.29 to 3.21) and the association remained significant in all leave-one-out analyses (see Supplementary Fig. 4).
Across single and multi-pollutant models, Chen et al. (2022a) found a consistent association between PM 2.5 and COVID-19 hospitalisation when using PM 2.5 measured over the previous year. However, these associations were weaker and inconsistently significant when using PM 2.5 from the previous month. Similarly, Li et al. (2022) found a positive association between PM 2.5 and clinically-defined severe COVID-19 across four different lag periods (0-7 days to 0-28 days), which attenuated but remained mostly significant when adjusting for other air pollutants.

PM 2.5 exposure and COVID-19 mortality
Six studies examined PM 2.5 and COVID-19 mortality, which are summarised in Supplementary Table 3. Four of six studies found a significant positive association. The remainder were null. One study was rated "poor"; the rest were "good".
Pooled results from n = 5 studies were positive but non-significant (OR: 1.40; 95 % CI: 0.94 to 2.10) (see Fig. 2), with heterogeneity explaining 75 % of the variance (p = 0.010). There was no evidence of publication bias (p = 0.100). Trim-and-fill points could not be applied, though leave-oneout sensitivity analysis indicated the association remained positive but only became significant with the exclusion of Chen et al. (2022b) Elliott et al. (2021) was the only study that did not restrict its sample to those diagnosed or hospitalised with COVID-19, avoiding associated issues of collider bias. It also had a null finding with a negative point estimate (OR: 0.94, 95 % CI: 0.75-1.18). However, the model included multiple predictors rather than being built around a single exposure-outcome relationship, meaning associations were less reliable.
Of the remaining studies, all but Chen et al. (2022b) found a significant positive association between PM 2.5 and COVID-19 mortality. Chen et al. (2022a) found that PM 2.5 was consistently associated with higher mortality rates across multiple models, regardless of whether PM 2.5 was measured in the previous month or year and whether the model adjusted for NO 2 .

Discussion
Studies of COVID-19 infection were generally of good quality and consistently demonstrated a significant association with PM 2.5 across methodologies and populations. Most importantly, the effect was observed even when adjusting for individual-level socioeconomic indicators, probably the most important confounder (Blakely et al., 2004). While there was evidence of publication bias in COVID-19 literature, trim-and-fill and leaveone-out sensitivity analyses were robust to exclusions.
The association with COVID-19 infection was observed in both single and multi-pollutant models across all studies, suggesting PM 2.5 is not just an indicator of a generalised effect of poor air quality on risk of COVID-19 infection, but an independent, causal predictor. However, multicollinearity may be an issue for multi-pollutant models, as indicated by increase in effect size when other air pollutants were added to PM 2.5 models in Veronesi et al. (2022).
Studies on COVID-19 severity and mortality also indicated a positive association, though the quality of the research was weaker and pooled mortality results were non-significant. Nearly every study was limited to people already diagnosed or hospitalised with COVID-19, introducing potential collider bias, or more specifically endogenous selection bias (Elwert and Winship, 2014). As the above results suggest that PM 2.5 influences who gets COVID-19, it could also mean that the infected cohorts differ substantially from the uninfected based on their PM 2.5 exposure. For instance, PM 2.5 may expand infections into less-vulnerable populations, reducing baseline risk of severe infection and biasing the association between PM 2.5 and severity towards null. Kogevinas et al. (2021) was the lone severity study to include participants who were not already infected. It also designed statistical models to examine the effect of PM 2.5 rather than including multiple predictors in a single model, and used high-resolution measures at 100m 2 , finding a positive association with COVID-19 severity. Elliott et al. (2021) was the only mortality study to include participants not diagnosed or hospitalised with COVID-19. While it found no association with PM 2.5 exposure, all predictors were included in a single model, making the results less reliable (Westreich and Greenland, 2013). However, its exclusion in leave-one-out sensitivity analysis did not meaningfully affect pooled results.
Despite the weakness of evidence for effects of PM 2.5 on COVID-19 severity and mortality, there are still reasons to treat it as real. There is accumulating evidence of a mechanism, including effects of PM 2.5 on receptor expression, cell defence, and cardiovascular and pulmonary health (Bourdrel et al., 2021;Rebuli et al., 2021;Landrigan et al., 2018) which may make infected persons more vulnerable to worse COVID-19 outcomes. Combined with the positive (if not always significant) associations identified in this review, PM 2.5 air pollution should be treated as a risk factor for severe COVID-19 disease and death.

Short versus long-term PM 2.5 exposure
While the reviewed studies used a variety of time periods to measure PM 2.5 , they provide little insight as to whether chronic (or cumulative or long-term) or acute (or short-term) exposures matter most for COVID-19 risk. Simply comparing coefficients across models is insufficient since measurements over different time periods are likely capturing the same underlying variance in exposure rather than distinguishing between short or long-term exposures. For instance, in two of the included studies PM 2.5 measurements from the previous year and previous month were correlated (Sidell et al., 2022;Chen et al., 2022a). López-Feldman et al. (2021) was the only study to include PM 2.5 measures over separate time points (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018)2019, and the two weeks preceding symptom onset) within a single model. While all had a positive association with COVID-19 mortality, only 2000-2018 PM 2.5 was significantly predictive of COVID-19 mortality at the standard significance threshold (p ≤ 0.05). This suggests long-term PM 2.5 is important for mortality outcomes, though the mechanisms for effects on infection and severity/mortality risk may differ; for instance, the increase in ACE2 expression may be more important for infection risk, while the negative effects on general health may be more important for severity/mortality. It is yet unclear whether the ACE2 expression is a response to acute or chronic PM 2.5 exposure, though worsening of general health seems to be a consequence of long-term/chronic exposure. These mechanisms will be important to tease out to better understand how PM 2.5 affects COVID-19 vulnerabilities.

Evidence gaps
We identified several major evidence gaps. The first is a lack of cohort/ case-control studies of COVID-19 severity and mortality that were not limited to those with COVID-19 and that built models specifically around PM 2.5 exposure. The second gap is a lack of cohort/case-control studies on discrete, large-scale PM 2.5 exposures such as smoke from wildfires. It remains unknown whether intensive PM 2.5 exposure increases short and long-term risks of respiratory illnesses like COVID-19. There is some ecological evidence of an association, though this mainly focuses on concurrent PM 2.5 exposures. In the months following the 2019-2020 Black Summer fires in New South Wales, Australia, areas with more burn coverage had higher rates of COVID-19. However, there was no detectable association with larger particulate matter, PM 10 , and the study did not investigate PM 2.5 (Cortes-Ramirez et al., 2022). We therefore have little idea whether and how long people may be at elevated risk of COVID-19 following major smoke exposures.

Strengths and limitations
Among this systematic review's strengths are an inclusion criterion that limited evidence to studies using individual-level data, a quality assessment that indicated most were of good quality, and synthesis of data with a metaanalysis. This review covers studies published in the first 2.5 years of the pandemic, building on previous reviews with more up-to-date evidence.
There are some limitations. Operationalisation of PM 2.5 exposure varied across studies, including when it was measured, precision, and time periods covered. No studies captured variations in individual exposure due to time spent outdoors or regular movement into areas like the workplace. The review was restricted to outdoor air pollution exposures, which, combined with most studies originating in high-income countries, may not be applicable to lower-income countries where indoor air pollution from 'dirty' heating and cooking fuels are a greater threat. While all studies used individual-level data, many used aggregated indicators for important confounders like socioeconomic status. COVID-19 is still a relatively new illness, so this review can only be considered an early snapshot of the evidence.

Conclusion
While the studies included in this systematic review had several methodological limitations, they provide compelling evidence that exposure to PM 2.5 increases the likelihood of COVID-19 infection and severity of disease. This conclusion becomes stronger when considered alongside evidence for a mechanism, namely that PM 2.5 increases expression of the ACE2 receptor, to which COVID-19 binds, and worsens cardiopulmonary health generally. While the pooled mortality result was non-significant, it nevertheless indicated that PM 2.5 increases the likelihood of death due to COVID-19; it would also make little sense for PM 2.5 to increase the severity of COVID-19 disease without increasing likelihood of death. We see good reason to treat the association as real, if not yet fully established. A major gap was the lack cohort of case-control studies focused on discrete, large-scale PM 2.5 exposures such as smoke from wildfires. Such evidence will become increasingly important as climate change increases both the frequency and intensity of wildfires.

Funding
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Registration
This study was registered on PROSPERO on 8 July 2022

Data availability
Available on a public repository, which is cited in the manuscript: https://doi.org/10.26180/21529158.v3

Declaration of competing interest
The authors declare no competing interests. ("air pollut*" OR "traffic pollution" OR "pm2.5" OR "black carbon" OR "particulate matter" OR dust OR "vehicle emissions" OR "particle pollution" OR smoke OR smog OR soot OR "coal ash") AND ("COVID-19" OR "SARS-CoV-2" OR "coronavirus disease 2019") AND (virulence OR "patient acuity" OR morbidity OR mortality OR "patient admission" OR hospitalisation OR "illness severity" OR incidence OR prevalence OR "length of stay" OR "critical illness")