Association of smoking, lung function and COPD in COVID-19 risk: a two-step Mendelian randomization study.

BACKGROUND AND AIMS: Smoking increases the risk of severe COVID-19, but whether lung function or chronic obstructive pulmonary disease (COPD) mediate the underlying associations is unclear. We conducted the largest Mendelian randomization study to date, to our knowledge, to address these questions.
DESIGN: Mendelian randomization study using summary statistics from genome-wide association studies (GWAS), FinnGen and UK Biobank. The main analysis was the inverse variance weighted method, and we included a range of sensitivity analyses to assess the robustness of the findings.
SETTING: GWAS which included international consortia, FinnGen and UK Biobank. PARTICIPANTS: The sample size ranged from 193 638 to 2 586 691. MEASUREMENTS: Genetic determinants of life-time smoking index, lung function [e.g. forced expiratory volume in 1 sec (FEV1 )], chronic obstructive pulmonary disease (COPD) and different severities of COID-19.
RESULTS: Smoking increased the risk of COVID-19 compared with population controls for overall COVID-19 [odds ratio (OR) = 1.19 per standard deviation (SD) of life-time smoking index, 95% confidence interval (CI) = 1.11-1.27], hospitalized COVID-19 (OR = 1.67, 95% CI = 1.42-1.97) or severe COVID-19 (OR = 1.48, 95% CI = 1.10-1.98), with directionally consistent effects from sensitivity analyses. Lung function and COPD liability did not appear to mediate these associations.
CONCLUSION: There is genetic evidence that smoking probably increases the risk of severe COVID-19 and possibly also milder forms of COVID-19. Decreased lung function and increased risk of chronic obstructive pulmonary disease do not seem to mediate the effect of smoking on COVID-19 risk.

INTRODUCTION

The coronavirus disease 19 (COVID‐19) pandemic has had a substantial impact on global morbidity and mortality. As of June 2021 there were 176 million confirmed cases, of which 3.8 million (2.16%) had died [1]. Previous observational studies, such as the OpenSAFELY study, showed that people with comorbidities had a higher risk of a COVID‐19‐related death [2]. However, the association of smoking with COVID‐19‐related deaths was inconsistent among analytical models, while other studies suggested a possible role of smoking in the development of symptomatic COVID‐19 [3]. These discrepancies may imply issues in observational studies, such as confounding, selection bias and problems with interpretation of mutually adjusted statistical models [4, 5]. Mendelian randomization studies, a design more robust to confounding than other observational studies due to the use of genetic variants randomly allocated at conception, have suggested that smoking and obesity are probably causes of increased susceptibility to COVID‐19 and more severe illness [6]. Although the effect of smoking on increasing risk of severe COVID‐19 is probably causal, the underlying mechanisms remain unclear. Given that smoking is strongly linked to poorer lung function and higher risk of chronic obstructive pulmonary disease (COPD), these could be potential mediators underlying the detrimental effect of smoking on COVID‐19 and hence be targets of intervention. A recent study suggested that idiopathic pulmonary fibrosis may have a role in severe COVID‐19, although the results were sensitive to the choice of analysis [7]. More importantly, whether smoking increases vulnerability to milder COVID‐19, which accounts for the majority of COVID‐19 cases, is still unascertained but is important to clarify, given the issue of long COVID‐19 [8]. Moreover, smoking is a modifiable life‐style factor and smoking cessation services are readily available. To clarify these interlocking questions, we conducted a Mendelian randomization study to assess the role of smoking in risk of COVID‐19 by severity, as well as assessing the mediating role of lung function and COPD in the association of smoking with risk of COVID‐19 using the most up‐to‐date genome‐wide association study (GWAS) from the COVID‐19 Host Genetic Initiatives (release 6, 15 June 2021) [9].

METHODS

We used a two‐step Mendelian randomization study to assess mediation in a Mendelian randomization framework [10], where we explored separately the association of the exposure with the outcome, exposure with mediators, and mediators with the outcome, using univariable Mendelian randomization [10]. We only considered multivariable Mendelian randomization (MVMR) when there was evidence of mediation from the two‐step Mendelian randomization, given the possibility of weak instrument bias and the direction of bias being unpredictable [11]. Similar to a conventional Mendelian randomization study, the instruments for exposures and potential mediators should fulfil three main assumptions [12]. First, the instruments should be related to the corresponding phenotypes (e.g. exposures/mediators). Secondly, the instruments should not be confounded. Thirdly, the instruments’ association with mediators and outcomes should occur only via their effects on exposure and mediators. Supporting information, Fig. S1 shows the schematic diagram of this study.

EXPOSURE: LIFE‐TIME SMOKING INDEX

Genetic determinants of a life‐time smoking index [standard deviation (SD)] was obtained from a GWAS in UK Biobank (n = 462 690 of European ancestry) by Wootton et al [13]. Smoking heaviness, duration and smoking initiation from a self‐reported questionnaire were used to derive a life‐time smoking index using the method proposed by Leffondre et al., which could more clearly capture smoking behaviours [14]. The study gave 124 instruments for the life‐time smoking index reaching genome‐wide significance (P‐value < 5 × 10−8) and not in high linkage disequilibrium (LD) (r 2 < 0.001). The GWAS was adjusted for sex and genotype array, and accounted for population stratification and relatedness using a linear mixed model (BOLT‐LMM) by the original GWAS investigators.

MEDIATORS: LUNG FUNCTION AND LIABILITY TO DEVELOP COPD

Forced expiratory volume in 1 sec (FEV1) and forced vital capacity (FVC)

Genetic determinants of FEV1 (SD) and FVC (SD) were obtained from publicly available UK Biobank GWAS summary statistics (https://gwas.mrcieu.ac.uk/) by Medical Research Council Integrative Epidemiology Unit (MRC IEU), University of Bristol [15, 16], as per our previous study (n = 421 986 of European ancestry) [17]. Pre‐bronchodilation lung function testing was performed by trained health‐care staff using a Vitalograph Pneumotrac 6800 spirometer (Maids Moreton, UK). The GWAS provided 260 instruments for FEV1 and 320 for FVC which reached genome‐wide significance, and not in high LD, although many of these instruments predicted both FEV1 and FVC. The GWAS was adjusted for sex and genotype array, and accounted for population stratification and relatedness using BOLT‐LMM by MRC IEU investigators. We also used the same GWAS as the outcome when assessing the exposure–mediator association.

FEV1/FVC

Genetic determinants of FEV1/FVC (SD) were obtained from a GWAS of lung function, which was comprised of the UK Biobank and SpiroMeta (n = 400 102 of European ancestry) [18]. Measurement of FEV1/FVC in the UK Biobank was described previously and measurement for SpiroMeta varied throughout studies. FEV1/FVC was inverse normal‐transformed [18]. According to the original GWAS investigators, the UK Biobank GWAS was adjusted for age, age2, sex, height, smoking status and genotyping array, and accounted for population stratification and relatedness using BOLT‐LMM. The studies in SpiroMeta were adjusted for age, age2, sex, height and principal components using linear regression, and smoking status was controlled for via stratification. A different analytical model was used for studies with related participants. The study gave 97 instruments for FEV1/FVC reaching genome‐wide significance and not in high LD. We also used the same GWAS (n = 321 047) as the outcome when assessing the exposure–mediator association.

Genetic liability to COPD

Genetic determinants of liability to COPD were obtained from a GWAS comprised of the UK Biobank and International COPD Genetics Consortium (35 735 cases and 222 076 controls, mixed ancestry) [19]. COPD was defined using pre‐bronchodilator spirometry according to modified GOLD criteria in both studies. The GWAS were adjusted for age, sex, genotyping array (UK Biobank only), smoking pack‐years, smoking status and principal components using logistic regression by the original GWAS investigators. The study gave 71 instruments for the liability of COPD reaching genome‐wide significance and not in high LD. When assessing the exposure–mediator association, GWAS summary statistics for COPD were required, which were not readily available for the GWAS from which the genetic instruments for COPD were extracted [19]. As such, we used summary statistics for COPD (R5_J10_COPD cases: 6915; controls: 186 723) in FinnGen, a consortium of Biobanks in Finland (Freeze 5), when assessing the exposure–mediator association. COPD was defined based on International Classification of Diseases (ICD) codes retrieved from nation‐wide registries in Finland. The two data sources (GWAS and FinnGen) were largely comparable in terms of genetic associations, given that they were both of people of European ancestry. The analyses in FinnGen were adjusted for age, sex, 10 principal components and genotype batch using mixed‐model logistic regression [Scalable and Accurate Implementation of GEneralized mixed model (SAIGE)] by the FinnGen investigators.

OUTCOMES: COVID‐19 PHENOTYPES

Genetic associations with COVID‐19 phenotypes were obtained from the COVID‐19 Host Genetics Initiative (https://www.covid19hg.org/, release 6, 15 June 2021), excluding results from 23andMe [9]. COVID‐19 Host Genetics Initiative is an ongoing international collaborative initiative focusing upon the genetic determinants of SARS‐CoV‐2 infection and COVID‐19 phenotypes (e.g. severity). The majority of the studies were from people of European descent. COVID‐19 phenotypes included all COVID‐19 (cases: 112 612; population controls: 2 474 079, 89% Europeans), hospitalized COVID‐19 (cases: 24 274; population controls: 2 061 529, 88% Europeans) and very severe respiratory confirmed COVID‐19 (cases: 8779; population controls: 1 001 875, 95% Europeans). (https://www.covid19hg.org/). In brief, COVID‐19 status was identified differently in each study, including laboratory‐confirmed SARS‐CoV‐2 infection (based on RNA and/or serology tests) or hospital/physician‐confirmed COVID‐19 via self‐reports. Hospitalized COVID‐19 was defined as hospitalized laboratory‐confirmed SARS‐CoV‐2 infection where hospitalization was due to corona‐related symptoms. Severe COVID‐19 was defined as the condition of hospitalized COVID‐19 and death/respiratory support (e.g. intubation, continuous airway pressure). The GWAS was adjusted for age, age2, sex, age × sex, principal components and study‐specific covariates by the original GWAS investigators, and SAIGE was recommended as the analysis approach.

A more detailed description of the above‐mentioned GWAS, such as mean age (if reported), quality controls and exclusion criteria, can be found in the Supporting information.

Exposure

The primary exposure was genetically predicted life‐time smoking index (SD).

Mediators

The mediators were genetically predicted FEV1, FVC, FEV1/FVC and genetic liability to COPD.

Outcomes

The primary outcome was overall COVID‐19. The secondary outcomes were hospitalized COVID‐19 and severe COVID‐19.

Confounders

To assess whether the instruments were related to potential confounders of the exposure‐outcome relation and hence constituted possible horizontal pleiotropic effects [20], we assessed the relation of life‐time smoking instruments (aligned to risk alleles) with possible confounders using relevant GWAS, i.e. body mass index [n = 681 275 from Genetic Investigation of ANthropometic Traits (GIANT) consortium] [21], educational attainment [n = 328 917 from Social Science Genetic Association Consortium (SSAGC), measured as years of education] [22] and alcohol use [n = 537 349 from GWAS and Sequencing Consortium of Alcohol and Nicotine use (GSCAN), measured as drinks per week] [23] using standard meta‐analysis with additive random effects, as per our previous study [17]. Afterwards, we conducted bi‐directional Mendelian randomization to verify whether the pleiotropy was vertical (consequences of smoking and hence unbiased pathways) or horizontal (causes of smoking and hence biased pathways).

Statistical analyses

To verify whether the instruments were of adequate strength for valid analyses, we assessed the instruments’ strength by calculating the total variance explained (R 2) by the instruments and derived the overall F‐statistic using summary statistics, as per previous studies [17, 24]. We also approximated the F‐statistics for each instrument for genetic liability to COPD using the instrument association with exposure and its standard error [25]. We aligned all genetic associations on the same effect allele, and also used effect allele frequencies for palindromic instruments.

We used inverse variance weighted (IVW) with multiplicative random effects to assess the association of the life‐time smoking index (SD) with COVID‐19 risk (exposure to outcome) in the main analyses, where we corrected for multiple comparisons using a Bonferroni correction (i.e. 0.05/3 = 0.016). The IVW method regressed instrument–outcome associations on instrument–exposure associations for the instruments, weighted by the inverse of the variance of the instrument–outcome association, with the intercept constrained to 0. We also used IVW to assess the association of the association of life‐time smoking index (SD) with lung function and COPD (exposure to mediators), as well as the association of lung function (FEV1, FVC, FEV1/FVC in SD) and liability to COPD (log‐odds) with COVID‐19 risk (mediators to outcome). IVW assumes balanced pleiotropy [26]. We assessed the heterogeneity of instruments using Cochran's Q‐statistics, where high heterogeneity indicates the presence of invalid instruments, and Mendelian randomization Egger (MR‐Egger) intercept and I 2 GX to assess evidence of overall horizontal pleiotropy and possible regression dilution in MR‐Egger, respectively.25 27 We also included other sensitivity analyses which rely upon different assumptions (e.g. majority valid; plurality valid; balanced pleiotropy) for valid inferences (as detailed in the Supporting information), such as MR‐Egger, weighted median, weighted mode and MR‐robust adjusted profile scores (MR‐RAPS) [27, 28, 29, 30]. Consistency in estimate direction across analyses strengthens the certainty of any observed associations [31].

As our previous study showed that height might bias the Mendelian randomization estimates for FEV1 and FVC [17], we additionally adjusted for height using multivariable MR (MVMR) when FEV1, FVC and FEV1/FVC were used as exposures. This is preferable to older approaches, such as removal of instruments associated with height, which would reduce statistical power. We also assessed the robustness of findings by using different GWAS for the height [Genetic Investigation of ANthropometic Traits (GIANT) and UK Biobank] [32]. We calculated the conditional F‐statistic as an indicator of weak instrument bias and repeated the analyses using MVMR‐Egger, which is more robust than MVMR [17, 33, 34].

All analyses were performed using R version 4.0.4 (R Core Team (2021). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria; https://www.R-project.org/) and the R packages (‘TwoSampleMR’) and (‘MVMR’).11,16 As the primary research question and analysis plan were not pre‐registered, the results should be considered exploratory.

Ethics approval

This study only used publicly available data and hence ethics approval was waived. Details of ethical approval and participant consent for each of the studies that contributed to the GWAS can be found in the original publications.

RESULTS

The instruments for life‐time smoking index had an overall F‐statistic of 49.2 and an R 2 of 1.2%, indicating low evidence for weak instrument bias (Supporting information, Tables S1–S2). There was also no evidence for weak instruments concerning lung function‐related traits or COPD (Supporting information, Table S2), and Supporting information, Tables S3–S6 shows the instruments for lung function‐related traits and COPD. Smoking‐increasing alleles were positively associated with body mass index (BMI) and alcohol use and inversely associated with educational attainment (Supporting information, Table S7). However, the associations were bi‐directional and hence indicated a mixture of both horizontal and vertical pleiotropy, i.e. both biasing and downstream pathways. For the main analyses we included up to 118 instruments for life‐time smoking index, 254 instruments for FEV1, 309 instruments for FVC, 94 instruments for FEV1/FVC and 69 instruments for genetic liability to COPD.

Figure 1a shows that a higher life‐time smoking index was associated with a higher risk of all COVID‐19 phenotypes for all analyses except MR‐Egger. The IVW estimates for all three phenotypes remained after considering multiple comparisons (ORCOVID‐19: 1.19 per SD, 95%CI 1.11 to 1.27, PCOVID‐19: 3.5 × 10−7; ORhospitalized COVID‐19: 1.67 per SD, 95%CI 1.42 to 1.97, Phospitalized COVID‐19: 9.7 × 10−10; ORsevere COVID‐19: 1.48 per SD, 95%CI 1.10 to 1.98, Psevere COVID‐19: 0.009). The most consistent findings were observed for severe COVID‐19 where all analyses suggested a positive association. Although there was evidence for heterogeneity based on Cochran's Q‐test (P‐value: 0.0019), the MR‐Egger intercept test did not indicate the presence of horizontal pleiotropy. Figure 1b shows that a higher life‐time smoking index was associated with lower lung function, with consistent evidence across all analyses. Similarly, a higher life‐time smoking index was associated with a higher risk of COPD, although there was horizontal pleiotropy (Fig. 1c).

FIGURE 1

Association of life‐time smoking index [standard deviation (SD)] in: (a) COVID‐19 risk; (b) lung function [forced expiratory volume in 1 second (FEV1), forced vital capacity (FVC) and FEV1/FVC)]; and (c) chronic obstructive pulmonary disease (COPD) risk using Mendelian randomization

Figure 2a,b shows that FEV1 and FVC were not associated with COVID‐19. While Cochran's Q‐test indicated the presence of heterogeneity (ranging from 1.4 × 10−10 to 0.0007), the MR‐Egger intercept did not indicate the presence of overall horizontal pleiotropy. Additional adjustment for height using multivariable Mendelian randomization (IVW and MR‐Egger) did not change the conclusion, regardless of GWAS used for height, although the conditional F‐statistics were generally low (Supporting information, Table S8).

FIGURE 2

Association of: (a) forced expiratory volume in 1 second [FEV1, standard deviation (SD)] and (b) forced vital capacity [FVC, SD] in COVID‐19 risk using Mendelian randomization

Figure 3a shows that higher FEV1/FVC was associated with higher hospitalized and severe COVID‐19 risk. Although there was evidence for substantial heterogeneity among instruments in IVW (Cochran's Q P‐value ranged from 3.74 × 10−17 to 0.00598), the MR‐Egger intercept did not suggest directional horizontal pleiotropy. These associations were not present in sensitivity analyses, including the multivariable Mendelian randomization analyses (Supporting information, Table S8). Similar trends were observed for liability to COPD (Fig. 3b), which was associated with lower risk of hospitalized and severe COVID‐19, but these associations were not evident in other sensitivity analyses.

FIGURE 3

Association of: (a) forced expiratory volume in 1 second (FEV1)/ forced vital capacity (FVC); and (b) liability to chronic obstructive pulmonary disease (COPD, log‐odds) in COVID‐19 risk using Mendelian randomization

Most I 2 GX were close to 100%, indicating that there is little evidence of regression dilution (Supporting information, Table S9). However, I 2 GX indicated possible regression dilution for analyses of COPD liability in COVID‐19 risk (83%) (Fig. 3b).

DISCUSSION

In this large Mendelian randomization study, we confirmed previous studies showing the detrimental role of smoking in severe COVID‐19 risk and have added to these studies by showing that smoking probably increases the risk of hospitalized or overall COVID‐19 [6]. Whether or not smoking cessation can reduce the burden of COVID‐19 among smokers should be considered. As expected, smoking decreased lung function and increased COPD risk. However, our study did not provide strong evidence that these mediate the effect of smoking on COVID‐19 risk.

In this study we focused primarily upon the possible mediating role of traits related to lung function. Although previous studies suggest that COPD is a risk factor for poorer COVID‐19 prognosis [35], this is not consistent with our findings. We could not rule out the possibility that our study only considered liability to COPD rather than the effect of COPD diagnosis, which is one issue with instruments for diseases. As the COPD GWAS was adjusted for smoking, it is also possible that some genetic variants identified from the GWAS were due to collider bias [36]. Inconsistencies could also be due to confounding by smoking and obesity in previous observational studies, or to selection bias due to inevitably only selecting survivors of COPD and competing risk of COVID‐19 to which Mendelian randomization studies are vulnerable [37]. Similarly, we were not able to show strong evidence for an effect of lung function on COVID‐19 risk, but Mendelian randomization studies of lung function could be challenged, given the strong pleiotropic effects of height. However, we accounted for this issue using multivariable Mendelian randomization.

As smoking impacts upon a wide spectrum of health outcomes, it is possible that these could be unexplored mediating pathways [38]. Inflammation, in particular, has received much attention regarding COVID‐19 because of its relevance to the immune system, and may explain why certain people have a poorer COVID‐19 prognosis [39]. Genetic evidence for associations of inflammatory markers with COVID‐19 is limited [40]. Nevertheless, randomized controlled trials of interleukin (IL)‐6R antagonists, such as tocilizumab, suggest that it may be an effective treatment [41]. Apart from the physiological role of smoking in COVID‐19, the higher risk in observational studies could be due to other factors associated with smoking, such as lower socio‐economic position, limited living space and access to facilities hindering adherence to social distancing policies or how cigarettes were being used (repeated removal of masks), which exacerbates the higher risk of COVID‐19 among smokers independently of its biological mechanisms [42, 43, 44].

Our study suggests that smoking probably increases susceptibility of COVID‐19 risk, with the most convincing evidence regarding COVID‐19 severity. This study substantiates the importance of tobacco control in mitigating the disease burden due to COVID‐19, which could narrow COVID‐19‐induced inequalities [43]. The use of more innovative means, such as emphasizing vulnerability to COVID‐19 as another hazard of smoking, to improve messaging of smoking cessation could be further explored [45].

Although we used a design which is more robust to residual confounding than observational studies, there are some limitations. First, Mendelian randomization requires three main assumptions. We chose strong instruments which were derived from a large GWAS to reduce the risk of weak instrument bias. A low R 2 of the instruments can lead to false negatives, although our main analyses indicated detrimental effects of smoking on COVID‐19 risk and severity (Fig. 1) [46]. We could not rule out the possibility of biases due to violation of the instrumental variable assumptions, such as association of instruments with confounders, where smoking‐increasing alleles were associated with confounders which could increase the risk of COVID‐19, such as lower educational attainment and higher BMI. These may exaggerate the overall detrimental effect of smoking. The majority of the sensitivity analyses are consistent with the main analyses regarding the detrimental effect of smoking in COVID‐19 except MR‐Egger. While the underlying reasons are not clear, the MR‐Egger method is also particularly sensitive to outliers and has lower statistical power. Furthermore, MR‐Egger requires instrument strength independent of direct effect assumption which could be violated, as our instruments were related to confounders. These may explain the different results compared to results from other sensitivity analyses, such as the weighted median and MR‐RAPS which rely upon other assumptions. Secondly, we used a two‐step Mendelian randomization design to explore possible mediation, and as there was no strong evidence for an effect of lung function or COPD in COVID‐19 risk we did not assess mediation using other approaches as verification, such as multivariable Mendelian randomization, which were more prone to weak instrument bias, as indicated by the conditional F‐statistics. As we used two‐sample Mendelian randomization we were not able to explore other phenotypes, such as smoking heaviness, which requires stratification by smoking status and hence access to individual level data. Nevertheless, a recent Mendelian randomization study in the UK Biobank suggested that smoking heaviness among smokers was also related to a higher risk of COVID‐19. Taking into account existing evidence from previous Mendelian randomization studies, it is likely that smoking increases COVID‐19 risk [47]. Thirdly, our study was based on European populations. However, many previous observational studies of smoking have shown consistent effects globally, so it is unlikely that the harmful effect of smoking on COVID‐19 differs by ethnicity [48], and similar associations were seen in observational studies among Chinese people [49]. Nevertheless, additional studies in non‐European populations would be valuable to further confirm the harm of smoking in COVID‐19. Fourthly, we could not exclude the possibility that some of the controls were cases who were asymptomatic or who only had mild symptoms, which would bias our estimates towards null due to non‐differential misclassification. The current definition of severe COVID‐19 excluded people who died of COVID‐19 without being hospitalized (e.g. frail older people living in elderly facilities), which may introduce bias. Although smoking probably increased the risk of severe COVID‐19, it would be worthwhile exploring its role in overall COVID‐19 mortality in future studies. Lastly, smoking is difficult to measure precisely, although we demonstrated the known effects of smoking on reduced lung function and increased COPD risk.

Our Mendelian randomization study provides genetic evidence that smoking probably increases the risk of severe COVID‐19 and possibly milder forms of COVID‐19. Our study provides credible evidence on the harmful effect of smoking in COVID‐19 risk and supports the importance of tobacco control in reducing the associated disease burden. Future studies should be conducted to explore the mechanistic pathways in multiple dimensions (i.e. social/biological/behavioural) underpinning these associations to reduce the burden of diseases arising from COVID‐19 in smokers.

DECLARATION OF INTERESTS

None.

AUTHOR CONTRIBUTIONS

Shiu Lun Au Yeung: Conceptualization; data curation; formal analysis; funding acquisition; investigation; methodology; project administration; software; supervision; visualization. Albert Martin Li: Supervision. Baoting He: Methodology. Kin On Kwok: Conceptualization; formal analysis; funding acquisition; investigation; methodology; visualization. C Mary Schooling: Investigation; methodology; supervision.

Figure S1: Schematic diagram of this study

Click here for additional data file.

Table S1: Instruments for lifetime smoking index

Table S2: Strength of the instruments

Table S3: Instruments for forced expiratory volume in 1 second (FEV1)

Table S4: Instruments for forced vital capacity (FVC)

Table S5: Instruments for FEV1/FVC

Table S6: Instruments for liability to chronic obstructive pulmonary disease (COPD)

Table S7: Assessment of (A) lifestime smoking index instruments' relation with confounders and (B) bi‐directional Mendelian randomization analyses

Table S8: The assocaition of FEV1 and FVC in COVID‐19 risk using Mendelian randoimization, adjusted for height

Table S9: I2 GX of the analyses in this study

Click here for additional data file.