The role of smoking in COVID-19 progression: a comprehensive meta-analysis

Discussion

In this comprehensive systematic review on the association between tobacco smoking and COVID-19 outcomes among SARS-CoV-2-positive individuals or COVID-19 patients we found that, compared to never-smokers, ever-smokers had a 16% higher risk of being hospitalised, a 44% higher risk of having a more severe form of the disease, a 39% higher risk of COVID-19 mortality and a 45% higher risk for progression.

These findings update and confirm results from previous meta-analyses systematically showing an excess risk of progression of diseases for current and former smokers combined compared to never-smokers [8, 10–12, 25, 26].

Current smokers had a 34% excess risk for COVID-19 severity (based on 124 studies) and 32% excess risk for mortality (based on 119 studies). For current versus never-smokers, Simons et al. [13] found an inconclusive increased relative risk of 1.26 (95% credible interval (CrI) 0.92–1.73; based on 11 studies) for severity and 1.12 (95% CrI 0.84–1.47; based on 19 studies) for mortality. Discrepancies between our results and those by Simons et al. [13] are likely due to the differences in the selection criteria of the two meta-analyses. In their analysis, Simons et al. [13] considered both peer-reviewed and pre-print publications and selected studies based upon defined quality indexes while the present systematic review considered only studies published in peer-reviewed journals. To maximise the comprehensiveness of the systematic review, it was decided not to select studies on the basis of defined quality indexes but consider aspects such as sample size and the use of adjustment to conduct stratified analyses in our meta-analysis. This decision likely results in an increased heterogeneity – all our pooled estimates showed substantial to considerable heterogeneity – that may be compensated by the increased statistical power. Another point of difference is the statistical methodologies employed, Simons et al. [8] used statistical methods based on a Bayesian approach while this study used classical meta-analytical methods.

Whereas Simons et al. [8] found a direct association with severity and mortality for former smokers, using classic analytical methods we did not observe major differences in the risk of disease progression between current and former smokers. The excess risk found by their analysis for former smokers is likely due to the residual excess risk in terms of morbidity and mortality due to previous smoking history, which can persist for several years [27, 28]. This may also be due to the fact that former smokers may have quit because of health problems, thus the health status of former smokers may be generally worse as compared to that of current smokers [29].

Only four reports analysed longitudinal studies based on the general population to quantify the effect of smoking status on the combination of COVID-19 incidence and mortality [30–33]. Excluding the report by Elliott et al. [30] that analyses the same cohort used in the report by Clift et al. [31], the pooled estimates showed a statistically significant 45% excess risk of COVID-19 mortality in ever-smokers compared to never-smokers. The point estimate for current smokers versus never-smokers was even higher, although the relatively limited number of current smokers did not allow us to reach a statistically significant excess risk. We strongly encourage the dissemination of data from these cohorts based on the general population (and not on COVID-19 patients). Such cohorts are unlikely to suffer from selection or sampling biases. Using fully adjusted models, including in particular smoking-related diseases (such as chronic respiratory conditions), the risk estimates did not substantially decrease. As well discussed by Williamson et al. [32], adjusting for these comorbidities does not capture the causal effect of smoking, since these comorbidities are likely to mediate the effect of smoking on COVID-19-related death [32]. Actually, various comorbidities in the fully adjusted model act as intermediate variables on a causal pathway between exposure (tobacco smoking) and outcome (COVID-19 death). This represents a well-established over-adjustment bias [34].

Studies show that age is by far the main characteristic influencing COVID-19 severity and, to an even greater extent, mortality [35]. Accordingly, the COVID-19 case fatality rate approximately triples for each decade of age. Among COVID-19 patients, current smokers are on average younger than former smokers. For example, in an Italian multicentric study based on 1406 COVID-19 patients, the mean age of current smokers was 57.7 years, that of never-smokers was 63.8 years and that of former smokers was 66.9 years, respectively [36, 37]. Considering crude risk estimates – not adjusted, among other covariates, by age – could provide highly misleading results on the association between smoking status and COVID-19 mortality. Overall, 18 studies provided odds ratios from multivariable analysis, adjusted at least for age. To evaluate the effect of covariates in odds ratio estimates, we computed the crude odds ratios for those studies. Indeed, compared to never-smokers, the adjusted odds ratio of current smokers was consistently lower and that of former smokers consistently higher than their corresponding crude estimates. This means that the crude odds ratio estimates for current smokers are likely underestimated and those for former smokers overestimated. Given that most included studies (60% to 70% of all studies) do not provide adjusted estimates, current evidence is likely biased with a strong underestimate of odds ratios for current and an overestimate of odds ratios for former smokers. Our findings are therefore compatible with a higher risk for current versus never-smokers than for former versus never-smokers.

It is worth noting the large limitations of the studies conducted to date, including: in all the available cohort studies, smoking data are secondary, i.e. no study has been designed with the primary aim to quantify the association between smoking and COVID-19 outcomes; none of the studies report biochemically verified smoking status; time since quitting has never being taken into account, thus no smoking in the last few days might be misreported as being a former smoker; missing data on smoking status; odds ratio is time-point dependent and changes in follow-up across studies will introduce heterogeneity and problems in interpretation of the summary result; and, finally, the large majority of available studies did not provide adjusted risk estimates.

Among the limitations of the present systematic review, we provided quality assessment only for studies focusing on current versus never-smokers. However, no substantial differences have been found between good-quality studies compared to others, leading us not to perform such sensitivity analyses on other classes. Excluding publications based on languages might be considered as a limitation of the present systematic review. However, out of the 39 non-English publications excluded, only nine would have been eligible for full-text screening and most of them were of limited quality. We are therefore confident that our results are not affected by this exclusion criterion.

The original methodology used to identify eligible articles in the scientific literature is a major strength of the present systematic review. This strategy, first described by the authors of this review [14], involves the further conduction of an umbrella review, which allowed us to include 187 articles (51% of all eligible articles) that would not have been otherwise included if we would have done a traditional review only. Overall, we found 367 eligible articles, approximately tripling the number of papers found by the most comprehensive systematic reviews available so far [7, 12, 13].

For future research, we strongly recommend using large longitudinal cohort studies to quantify the association between time since stopping smoking and COVID-19 outcomes. Currently, scant, if any, information is available on the association between intensity and duration of smoking and COVID-19 progression; future studies should also provide these estimates in order to evaluate dose–risk relationships. All the studies should provide results necessarily from multivariate analyses, with relative risks estimated, at least, after adjustment for age and sex. For a correct interpretation of results, further adjustment for smoking-related comorbidities should be considered as a sensitivity analysis

In conclusion, although some degree of uncertainty should not be ruled out due to the heterogeneity of results, our comprehensive review is compatible with a 30–50% excess risk of progression of COVID-19 (severity and mortality) for current and former smokers versus never-smokers.

In this scenario, preventing serious COVID-19 complications, including mortality and disease progression, appears to be the newest good reason to avoid smoking. These results suggest that a non-negligible proportion of the 6.2 million deaths worldwide due to COVID-19 are attributable to tobacco smoking. These deaths should be added to the estimated 20 million deaths attributable to smoking that occurred since the beginning of the COVID-19 pandemic.