Background
Coronavirus disease 2019 (COVID-19) first occurred in Wuhan, China in December 2019 and has spread rapidly around the world. As of April 2021, the virus had infected over 150 million people and caused more than 3 million deaths [
1]. Old people and those with pre-existing medical conditions including respiratory disease, hypertension, diabetes, cardiovascular disease, and cancer are more vulnerable to becoming critically ill when infected [
2].
Smoking may enhance the risk of COVID-19 by its biological effects and behaviors of smokers. Smoking impairs lung function and pulmonary immune function, compromising the body’s defense mechanisms against infections [
3]. Smoking is also a well-established risk factor for chronic diseases that are linked to more severe COVID-19. The World Health Organization (WHO) has advised the public that smoking could increase the risk of contracting COVID-19 because the behavior of smokers involves contact of fingers with the lips and removal of the protective face masks to smoke [
4].
Our recent meta-analysis of the 19 peer-reviewed papers found that smokers have double the odds of COVID-19 progression risk [
5]. Some people argue that the association between underlying health conditions and risk factors such as smoking to the severity of COVID-19 is still unclear due to inadequate adjustment of confounding factors [
6]. In addition, it is unclear whether the association between smoking and severity of COVID-19 varies by age. This paper updates and extends our previous meta-analysis [
5] of 19 studies to add 27 additional studies, including 6 that provided adjusted odds ratios and compared the association between smoking and COVID-19 disease progression between unadjusted analyses with adjusted analyses to examine whether smoking is an independent risk factor. We also assessed the effect of age of patients and conducted a sub-analysis for the risk of smoking on the mortality of COVID-19.
Methods
This study followed the Preferred Reporting in Systematic Reviews and Meta-Analyses (PRISMA) guidelines and is registered with PROSPERO (CRD42020186864).
Data source and search strategy
We conducted a systematic search using PubMed and Embase on May 25, 2020, with the search term: “((smoking) OR (characteristics) OR (risk factors) OR (retrospective*) OR (outcomes) OR (smoker*)) AND ((COVID-19) OR (COVID) OR (coronavirus) OR (sars cov-2) OR (sars cov 2))” for studies published between January 1, 2020 and May 25, 2020. A total of 2600 studies were retrieved through PubMed and 1962 studies through Embase.
Eligibility criteria
Eligible studies included published peer-reviewed observational studies, retrospective cohort studies, prospective cohort studies, cross-sectional studies, case series, and case reports that reported demographic characteristics, comorbidities specifically smoking status, clinical manifestations, and clinical or disease outcomes of COVID-19 patients on disease progression of COVID-19 to more severe or critical conditions or death. We included both inpatient and outpatient settings. We excluded studies that did not report smoking status and outcomes, studies of children, studies that included other coronavirus infection and not specifically to COVID-19, studies that the number of smokers was zero or omitted, and studies in which all patients had the same outcome. There were no language restrictions.
Study selection and data extraction
One author (RP) extracted information for each study, screened the abstract or the full text, with questions resolved through discussion among both authors (Fig.
A1).
The exposure group for our analysis were those who had a history of smoking (current smokers and former smokers) and unexposed group was never smokers, non-smokers, or not having a smoking history. Outcomes were progression of COVID-19 to more severe or critical conditions or death. Definitions of smoking status and disease progression for each study are shown in Table
A1.
Quality assessment
We evaluated the quality of studies using a modification of the ACROBAT-NRSI [
7] tool on 5 domains: study population, exposure measurement, outcome assessment, measurement of confounders, and adequate follow-up. Each one of these domains was scored from 0 (low risk of bias) to 2 (high risk of bias) and the average score of each study was computed and discussed among both authors (Additional file and Table
A2). Studies with the average score higher than 1 were considered high risk and excluded in a sensitivity analysis.
Statistical analyses
Our meta-analyses were based on unadjusted odds ratios (OR) that were either reported in the studies or computed unadjusted OR and 95% confidence interval (CI) using the number of smokers (current and former) and never smokers with and without disease progression. We also did a sensitivity analysis to determine the results changed when the 5 studies with high risk of bias were excluded.
We performed subgroup analyses of (1) the studies that reported association of smoking on COVID-19 mortality and (2) the association of COVID-19 disease progression between current smokers and never smokers (i.e., excluding former smokers), and former smokers and never smokers using the studies that reported whether the patient was a current, former, or never smoker (as separate categories).
We also computed the pooled adjusted OR using the studies that reported adjusted OR and 95% CI and compared it with the pooled unadjusted OR.
The results of the included studies were pooled with random-effect models using the Stata version 14.0 metan command and metabias command with Egger’s test for the presence of publication bias. We used metareg command (with dummy variables to account for the pairing of adjusted and unadjusted ORs) to determine whether the adjustment of OR affected the results. We used locally weighted regression and smoothing because both visual inspection and analysis of residuals using a preliminary linear regression indicated a nonlinear relationship between odds of disease progression and mean or median age reported in each study. The lowess command was used to generate a nonparametric fit estimate. We also tested for a trend using metareg command with mean age of each study as a continuous variable.
Discussion
With more than twice as many studies available compared to our earlier meta-analysis [
5], smoking remains a risk factor for COVID-19 disease progression, with smokers having 1.59 times the odds of progression in COVID-19 severity than non-smokers (Fig.
1). The risk of smoking on COVID-19 disease progression was not changed significantly by adjusting for confounders, which suggests that smoking is an independent risk of COVID-19 progression. We also find that smokers are at increased risk of death from COVID-19 (Fig.
A3). These findings are not surprising because the well-established evidence that smoking is associated with a higher risk of viral infection [
3]. In the past pandemics such as influenza [
3] and Cov-MERS [
57] smoking is also among leading risk factors for worse outcomes.
Although the studies included in our paper (published as of May 25, 2020) reflect the first wave of COVID-19 pandemic, our results (OR 1.59, 95% CI 1.33–1.89,
p = 0.001) are consistent with a meta-analyses published after our paper was submitted in September 2020 based on 109 studies from when the pandemic started [
58] to February 2021 (1.55, 95% CI 1.41–1.71) [
59]. Our finding of a higher OR for former than current smokers is also consistent with other analyses published after our paper was submitted [
58,
59].
Younger smokers appear to have a higher risk of COVID-19 disease progression than older smokers (Fig.
2). A recent study also found that younger adults are more medically vulnerable to severe COVID-19 illness if they are smokers [
60]. The greater effect of smoking among young people is particularly important because in the U.S., almost 40% of COVID-19 patients are aged 18–44 years [
61], and in China, 44% of COVID-19 patients are adults aged 20–49 years [
62]. Even so, younger adults tend to perceiving lesser infection-fatality risks of COVID-19 [
63] so that they are less likely to protect themselves from the infection. Our finding is consistent with a recent meta-analysis study [
64] which concluded that age was negatively significantly associated with the effect of smoking on COVID-19 disease severity.
While there is not yet direct peer reviewed evidence of the effect of e-cigarette use on COVID-19 risk, the fact that e-cigarettes have similar adverse effects on pulmonary immune function [
65] combined with the fact that e-cigarette use is concentrated among younger people, raises concerns and points to the need to collect data on e-cigarette use and COVID-19 risk.
Some have argued that smoking has a protective effect against COVID-19 because of the low smoking prevalence of reported among COVID patients [
66‐
68]. This is not new. There were also rumors that smoking protected patients from developing Cov-SARS during the 2003 pandemic [
69]. However, a case-control study of 447 patients showed that smoking did not protect patients from contracting Cov-SARS after adjusting for confounding by age, gender, contact history, and occupation [
69].
Reported smoking prevalence in the 33 studies in China ranged from 1.4 to 29.8% (median = 7.3%), which was substantially lower than 27.7% (52.1% for men and 2.7% for women) smoking prevalence in 2015 [
70]. Four studies [
10,
15,
17,
35] in the U.S. that reported the smoking prevalence among current smokers ranged from 1.3 to 33.3% (median = 5.2%), which was also lower than 13.7% (15.6% for men and 12.0% for women) smoking prevalence in 2018 [
71]. The other 4 studies [
9,
23,
26,
34] in the U.S. reported the ever-smoking prevalence ranged from 13.3–33.5%, which was also lower than 41.9% (47.2% for men and 37.3% for women) in 2017 [
72]. One study [
37] in Italy reported the smoking prevalence among current smokers of 3.2%, which was also substantial lower than 21.1% (26% for men and 17.2% for women) in 2016 [
73]. The remaining studies that reported the ever-smoking prevalence (2 studies [
16,
33] in Italy, 1 study [
8] in UK and 1 study [
25] in South Korea) were also lower than the countries’ rates (Italy: 16.7–30% vs.43.9% (50% for men and 38.3% for women) in 2010 [
74]; UK: 16.7% vs. 40.2 (44.3% for men and 36.5% for women) in 2018 [
75]; South Korea: 18.5% vs.39.1% (81.6% for men and 6.9% for women) in 2015 [
76]). These low levels of reported smoking among COVID-19 patients may reflect the difficulty of obtaining accurate smoking histories among seriously ill patients, especially when most medical facilities are operating at or above normal capacity. Despite the fact that the reported levels of smoking have been below population prevalences; however, the reported smoking prevalence among people with worse outcomes was significantly higher than those with less severe outcomes (16.1% vs. 11.6%,
p = 0.023).
Limitations
The studies used a variety of clinical definitions of disease progression and smoking status (Table
A1). (This is a common practical problem when conducting meta-analyses.) The varying definitions of disease progression include severity of disease based on clinical manifestations, development to more severe conditions, increasing oxygen supplements, prolonged viral shredding, organ injuries, ICU admission, and death. These varying definitions likely introduced increased variance in the pooled risk estimates and probably accounts for at least some of the heterogeneity between studies that was observed.
However, smoking was significantly associated with death – a clearly defined endpoint – in the 7 studies [
12,
16,
26,
33,
34,
45,
54] that used this endpoint.
Most studies reported smoking status as having a smoking history, often without clearly stating how they categorized former smokers. Of the 46 studies we reviewed, only 7 [
10,
12,
18,
27,
35,
37,
50] reported all three smoking categories (current, former, and never smokers). A meta-analysis of these studies found that current smoking was associated with a similar increase in the point estimate for the odds of disease progression (OR 1.35, 95% CI 0.83–2.22; Fig.
A4) as the other studies (OR 1.54, 95% CI 1.24–1.91), but the odds for current smoking did not reach conventional statistical significance (
p = 0.202).
Studies that only describe patient smoking history as “smoking history” or “history of smoking” do not provide enough information to analyze smoking as a risk factor given the fact that time since quitting could have significant influence on the patient’s outcomes.
When estimating adjusted ORs, it is important to have an appropriately specified model. The studies that reported adjusted odds ratios accounted for a variety of covariates (Table
A1), which means that the resulting ORs are not strictly comparable [
77]. (This is another commonly-encountered problem in conducing meta-analyses.)
All these limitations add to misclassification errors, which tend to bias results toward the null, suggest that this analysis underestimates the risk of smoking in terms of increasing COVID-19 severity.
The effects of smoking on COVID-19 disease progression by age reported in our paper is limited to the mean or median age in the studies. Individual level data on smoking, age, e-cigarette use, demographics and other risk factors are needed to perform a more sophisticated analysis. In addition, most of the studies were retrospective cohorts or case series, there might be recall bias, and could not conclude a causal relationship. Most of the meta-analyses in this study had moderate and statistically significant heterogeneity; the reliability of the meta-analyses might be compromised.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.