Journal of the American Heart Association ORIGINAL RESEARCH Leveraging Genetic Data to Elucidate the Relationship Between COVID-­19 and Ischemic Stroke Verena Zuber , PhD*; Alan Cameron , MBChB, PhD*; Evangelos P. Myserlis , MD; Leonardo Bottolo , PhD; Israel Fernandez-­Cadenas , PhD; Stephen Burgess , PhD; Christopher D. Anderson , MD, MMSc; Jesse Dawson, MD, MBChB; Dipender Gill , BMBCh, PhD BACKGROUND: The relationship between COVID-­19 and ischemic stroke is poorly understood due to potential unmeasured confounding and reverse causation. We aimed to leverage genetic data to triangulate reported associations. METHODS AND RESULTS: Analyses primarily focused on critical COVID-­19, defined as hospitalization with COVID-­19 requiring respiratory support or resulting in death. Cross-­trait linkage disequilibrium score regression was used to estimate genetic correlations of critical COVID-­19 with ischemic stroke, other related cardiovascular outcomes, and risk factors common to both COVID-­19 and cardiovascular disease (body mass index, smoking and chronic inflammation, estimated using C-­reactive protein). Mendelian randomization analysis was performed to investigate whether liability to critical COVID-­19 was associ- ated with increased risk of any cardiovascular outcome for which genetic correlation was identified. There was evidence of genetic correlation between critical COVID-­19 and ischemic stroke (rg=0.29, false discovery rate [FDR]=0.012), body mass index (rg=0.21, FDR=0.00002), and C-­reactive protein (rg=0.20, FDR=0.00035), but no other trait investigated. In Mendelian randomization, liability to critical COVID-­19 was associated with increased risk of ischemic stroke (odds ratio [OR] per logOR increase in genetically predicted critical COVID-­19 liability 1.03, 95% CI 1.00–­1.06, P-­value=0.03). Similar estimates were ob- tained for ischemic stroke subtypes. Consistent estimates were also obtained when performing statistical sensitivity analyses more robust to the inclusion of pleiotropic variants, including multivariable Mendelian randomization analyses adjusting for potential genetic confounding through body mass index, smoking, and chronic inflammation. There was no evidence to sug- gest that genetic liability to ischemic stroke increased the risk of critical COVID-­19. CONCLUSIONS: These data support that liability to critical COVID-­19 is associated with an increased risk of ischemic stroke. The host response predisposing to severe COVID-­19 is likely to increase the risk of ischemic stroke, independent of other potentially mitigating risk factors. Key Words: COVID-­19 ■ cross-­trait linkage disequilibrium score regression ■ ischemic stroke ■ Mendelian randomization S ARS-­CoV-­2 infection is the cause of the COVID-­19 greater than in influenza infection,3 with up to 5% of pandemic that has resulted in a health crisis of people with severe COVID-­19 suffering stroke.5 Strokes unprecedented magnitude.1,2 While much of the that occur in individuals with COVID-­19 are more se- disease burden relates to respiratory failure and sep- vere, have poorer outcomes, and higher mortality rates sis, some studies suggest an increased risk of ischemic than in those without COVID-­19, despite similar acute stroke.3–­6 This has been estimated to be seven times management.6,7 Indeed, almost two-­ fifths of people Correspondence to: Verena Zuber, Department of Epidemiology and Biostatistics, School of Public Health, Medical School Building, St Mary’s Hospital, Imperial College London, UK, W2 1PG. E-­mail: v.zuber@imperial.ac.uk *V. Zuber and A. Cameron contributed equally to this article and are co-­first authors. Preprint posted on MedRxiv March 1, 2021. doi: https://doi.org/10.1101/2021.02.25.21252441. Supplementary Material for this article is available at https://www.ahajo​urnals.org/doi/suppl/​10.1161/JAHA.121.022433 For Sources of Funding and Disclosures, see page 8. © 2021 The Authors. Published on behalf of the American Heart Association, Inc., by Wiley. This is an open access article under the terms of the Creat​ive Commo​ns Attri​bution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. JAHA is available at: www.ahajournals.org/journal/jaha J Am Heart Assoc. 2021;10:e022433. DOI: 10.1161/JAHA.121.0224331 Zuber et al COVID-­19 and Ischemic Stroke plausible mechanisms by which COVID-­19 may be in- CLINICAL PERSPECTIVE creasing ischemic stroke risk. COVID-­19 can trigger a cytokine storm with upregulation of pro-­inflammatory What Is New? signaling and endothelial dysfunction that predisposes • This study identified evidence of genetic corre- to a hypercoagulable state and can lead to throm- lation between critical COVID-­19 and ischemic boembolic events.15 Indeed, COVID-­19 also appears stroke, body mass index and C-­reactive protein. to promote the development of other cardiovascular • In Mendelian randomization, liability to critical disorders including myocardial injury, myocardial isch- COVID-­19 was associated with increased risk of emia, arrhythmias, heart failure, and venous thrombo- any ischemic stroke, with similar estimates ob- embolism.15 Furthermore, pre-­existing cardiovascular tained for ischemic stroke subtypes. disease (CVD) is associated with high mortality in peo- • There was no evidence to suggest that genetic ple with COVID-­19, which has raised the possibility of liability to ischemic stroke increased the risk of a bidirectional interaction between COVID-­19 and the critical COVID-­19. cardiovascular system.15 MR analyses also can allow What Are the Clinical Implications? the exploration of such bidirectional relationships. • These data support that liability to critical Elucidating the relationship between COVID-­ 19 COVID-­19 is associated with an increased risk and risk of ischemic stroke could prove important for of ischemic stroke. optimizing prevention and treatment strategies. With • The host response predisposing to severe this in mind, we performed cross-­trait LDSC to inves- COVID-­ 19 is likely to increase the risk of is- tigate whether there is a genetic correlation between chemic stroke, independent of other potentially COVID-­19 and ischemic stroke, and followed this up mitigating risk factors. with MR analyses to investigate whether any such sta- tistically significant correlation might be explained by liability to COVID-­19 being associated with increased risk of ischemic stroke. Nonstandard Abbreviations and Acronyms Fdr false discovery rate IVW inverse-­variance weighted METHODS LDSC linkage disequilibrium score regression All genetic association data used in this work are pub- MR Mendelian randomization licly accessible. Appropriate patient consent and ethi- cal approval had been obtained in the original studies from which they were obtained (Table S1). Statistical with COVID-­19 who develop stroke consequently die.8 code related to the analyses performed in the current However, some studies do not support an increased study is freely available from Github (https://github. risk of stroke in individuals with COVID-­19.9,10 Obtaining com/veren​a-­zuber/​covid​19_and_stroke). unbiased estimates for the risk of stroke in people with COVID-­19 is challenging due to difficulty diagnosing mild Study Overview COVID-­19 and an overall reduction in the rate of admis- First, we performed cross-­trait LDSC to estimate ge- sion to hospital with stroke, and minor stroke in particu- netic correlations for COVID-­19 with ischemic stroke, lar, during the pandemic,9,11 Furthermore, observational other related CVDs, and risk factors common to both studies investigating the association between COVID-­19 COVID-­19 and CVD. Second, for CVD outcomes that and stroke are vulnerable to potential confounding and showed evidence of genetic correlation with COVID-­19, reverse causation.3–­6 For example, there are common MR analysis was performed to investigate whether li- risk factors for severe COVID-­19 and stroke, such as ability to COVID-­19 was also associated with these obesity and smoking.12 Similarly, patients with acute outcomes. Finally, bidirectional MR was carried out to stroke have a dampened immune response and may investigate potential reverse associations, ie whether be more susceptible to severe COVID-­19.13 genetic liability to the CVD outcome was also associ- Leverage of genetic data can help overcome some ated with increased risk of COVID-­19. A graphical over- of these issues. Cross-­ trait linkage disequilibrium view of the analysis plan is presented in Figure S1. score regression (LDSC) can be used to estimate the genetic correlation between traits. Mendelian randomization (MR) can be employed to investigate Exposure Definitions and Genetic whether genetic variants predicting an exposure (such Association Estimates for COVID-­19 as COVID-­19) also associate with risk of an outcome Genetic association estimates for COVID-­19 were ob- (such as ischemic stroke).14 There are numerous tained from release 5 of the COVID-­19 host genetics J Am Heart Assoc. 2021;10:e022433. DOI: 10.1161/JAHA.121.0224332 Zuber et al COVID-­19 and Ischemic Stroke consortium.16,17 In our main analysis we focused on the artery disease were measured on 60 801 cases and most severe definition of COVID-­19 available (referred 123 504 controls and taken from the Coronary ARtery to critical COVID-­19 from here), where a critical case is DIsease Genome wide Replication and Meta-­analysis defined as an individual who was hospitalized with lab- (CARDIOGRAM) plus The Coronary Artery Disease oratory confirmed SARS-­CoV-­2 infection and required (C4D) consortium (CARDIoGRAMplusC4D),19 for heart respiratory support or died. Genetic associations were failure were measured on 47 309 cases and 930 014 derived from 5101 cases and 1 383 241 controls from controls and taken from the HEart failuRe Molecular the general population. Hospital admission and requir- Epidemiology for therapeutic targetS (HERMES) con- ing respiratory support or death is a proxy for disease sortium20 and for atrial fibrillation were measured on severity and is preferred here over other case defini- 65 446 cases and 522 744 controls and taken from a tions which are solely based on a positive COVID-­19 transethnic meta-­analysis.21 test result. Previous studies have shown that bias may impact analyses identifying cases based on likelihood Risk Factors Related to Both COVID-­19 of testing for SARS-­ CoV-­2 infection, because par- and CVD ticipants being tested for SARS-­CoV-­2 infection are To investigate whether any genetic correlation be- selected for a wide range of genetic, behavioral, and demographic traits.9 tween critical COVID-­19 and the CVD outcomes was related to confounding factors, we further considered Results based on other COVID-­19 definitions from common risk factors to both, including obesity, smok- the COVID-­ 19 host genetics consortium were per- ing, and chronic inflammation.22–­25 Genetic associa- formed as further sensitivity analysis. As the first sensi- tion estimates to proxy these traits were taken from a tivity analysis definition, we compared individuals with genome-­wide association study (GWAS) on body mass laboratory confirmed SARS-­CoV-­2 infection who had index (BMI) measured on 694 649 subjects,26 lifetime been hospitalized (cases) versus individuals with lab- smoking index measured on 462 690 subjects,27 and oratory confirmed SARS-­CoV-­2 infection who did not CRP (C-­reactive protein) measured on 361 194 indi- require hospitalization (4829 cases and 11 816 con- viduals in UK Biobank.28 trols). As a second analysis sensitivity definition, we compared individuals with laboratory confirmed SARS-­ CoV-­2 infection who had been hospitalized (cases) ver- Statistical Analysis sus the general population (9986 cases and 1 877 672 Cross-­Trait Linkage Disequilibrium Score controls). The third sensitivity analysis definition was Regression based on individuals with reported COVID-­19 (labo- We performed LDSC to estimate the genetic corre- ratory confirmed, physician-­reported or self-­reported; lation (rg) of critical COVID-­19 with the primary out- cases) versus controls from the general population come ischemic stroke, and secondary outcomes (38 984 cases and 1 644 784 controls). An overview of coronary artery disease, heart failure, and atrial fi- the COVID-­19 definitions is given in Table S1. brillation, using GWAS summary statistics data.29 We also estimated correlation with possible genetic Outcomes confounders, including BMI, lifetime smoking index, Ischemic Stroke and CRP. We restricted our analyses to HapMap 3 The primary outcome was any ischemic stroke (34 217 single-­nucleotide polymorphisms (SNPs), which are cases). In secondary hypothesis-­generating analyses, known to be well-­imputed across most studies and stroke subtypes were further explored as large ar- utilized the pre-­computed European LD-­scores es- tery stroke (LAS, 4373 cases), cardioembolic stroke timated using the 1000G reference panel, provided (CES, 7193 cases), and small vessel stroke (SVS, 5386 by the LDSC creators. For each set of summary sta- cases).18 The common control pool included 406 111 tistics, the SNP-­specific sample size information was individuals. Genetic association data were derived used. If not available, we assumed that all SNPs had from the MEGASTROKE consortium.18 the same sample size for that trait, defined as the total sample size for continuous phenotypes or as the sum of cases and controls for case/control phenotypes. Related CVD Outcomes By default, LDSC also removed variants that were We considered other CVD outcomes related to ischemic duplicate, strand-­ambiguous, not SNPs (eg indels), stroke in their pathophysiology. These were coronary with P-­values not between 0 and 1, with alleles that artery disease (including myocardial infarction, acute did not match with the 1000G reference panel, and coronary syndrome, chronic stable angina, or >50% with low effective sample size or not included in all coronary artery stenosis), heart failure, and atrial fi- studies of a GWAS meta-­analysis (if such information brillation. Genetic associations with risk for coronary was available) for traits with no effective sample size J Am Heart Assoc. 2021;10:e022433. DOI: 10.1161/JAHA.121.0224333 Zuber et al COVID-­19 and Ischemic Stroke information. After estimation of the genetic correla- CRP). Multivariable MR includes all the respective ge- tion across all phenotypes, we corrected for multiple netic associations in a joint model to account for ge- hypothesis testing using the Benjamini and Yekutieli netic confounding.36 While univariable MR measures false discovery rate (FDR).30 FDR-­corrected P-­values the total estimate of an exposure, multivariable MR <0.05 were considered statistically significant. measures the direct estimate of the exposure inde- pendent of other risk factors (ie, pleiotropy or genetic confounders) in the model.37 In the multivariable MR Mendelian Randomization Analyses model, we selected instruments based on the primary Genetic Variants Used as Instrumental Variables exposure of critical COVID-­19. We compared the mul- Genetic variants were selected based on associations tivariable MR model with the univariable MR model with critical COVID-­19. In our main analysis, we selected using likelihood ratio test to evaluate if accounting for uncorrelated genetic variants (clumped at correlation the pleiotropic pathway provides a better model fit than threshold r2<0.01) at P-­value <5×10−6. In sensitivity the univariable MR model. analyses, we applied a more stringent threshold and considered only genome-­wide significant genetic vari- Bidirectional MR ants (P-­value <5×10−8). For CVD outcomes that showed evidence of genetic correlation with critical COVID-­ 19 in LDSC, bidirec- tional MR was also performed to investigate for any Main Analysis association of genetic liability to that CVD outcome For CVD outcomes that showed evidence of genetic with risk of critical COVID-­19. Uncorrelated genetic correlation with critical COVID-­19 in LDSC, MR anal- variants (r2<0.01) associated with the CVD outcome at ysis was performed to estimate the association of ge- a P-­value <5×10−6 were selected as instruments. netically predicted liability to critical COVID-­19 with that outcome using the random effects 2-­s ample Power Calculation inverse-­variance weighted (IVW) method.31 The IVW Power to detect an association of genetically proxied estimate can be biased by pleiotropy when a genetic liability to critical COVID-­ 19 with all-­ cause ischemic variant associates the outcome (eg, ischemic stroke) stroke and its subtypes was assessed using power via a pathway other than through the exposure (ie, calculations for MR.38 Sample sizes were set ac- liability to critical COVID-­19). Pleiotropy can cause cording to the number of cases and controls in the heterogeneity in the MR estimates obtained by dif- MEGASTROKE consortium as given in Table S1. ferent variants employed as instruments, which was Heritability of the exposure was calculated from the assessed using the Q-­statistic and the respective F-­statistic of the genetic variants selected as instru- heterogeneity P-­value.32 A Mendelian randomization mental variables, which was approximated using the estimate with P-­value <0.05 for the main IVW analy- squared regression coefficient divided by its squared sis was deemed to represent supportive evidence, standard error. The F-­statistic was then transformed to given that MR was only performed to follow up posi- the proportion of variance in the phenotype explained tive LDSC findings. by the genetic variants R2 using the Cragg–­Donald transformation.39 Sensitivity Analyses—­Robust Methods MR estimates are expressed as odds ratios (OR) per We performed sensitivity analyses with pleiotropy-­ unit increase in the logOR of the exposure for binary robust two-­ sample summary-­ level MR approaches, traits. All analyses were performed using the ieugwasr including the weighted median MR33 and MR-­Egger34 (version 0.1.5) and MendelianRandomization (version to compare the MR estimates between different MR 0.5.0) R packages.40 models. Each of these methods provides a statistically consistent estimator of the true causal estimate under different assumptions. The intercept of the MR-­Egger RESULTS model represents a test for directional pleiotropy and we included this in sensitivity analyses.35 LD Score Regression Performing LDSC, we found evidence of genetic cor- Pleiotropic Pathways—­Inflammation and Cardiometabolic Risk relation between critical COVID-­ 19 and ischemic Factors stroke (rg=0.29, FDR-­P-­value=0.012) (Figure 1). Critical We further performed multivariable MR to adjust for COVID-­ 19 was also genetically correlated with BMI potential pleiotropic pathways via cardiometabolic risk (rg=0.21, FDR-­P-­value=0.00002) and CRP (rg=0.20, factors that are known to affect risk of both COVID-­19 FDR-­P-­value=0.00035). We did not observe evidence and CVD12, including obesity (BMI),26 lifetime smok- for genetic correlation between critical COVID-­19 and ing index,27 and chronic inflammation (estimated using the CVD outcomes (Table S2), and therefore focused J Am Heart Assoc. 2021;10:e022433. DOI: 10.1161/JAHA.121.0224334 Zuber et al COVID-­19 and Ischemic Stroke the consequent MR analysis only on ischemic stroke analyses based on pleiotropy-­robust approaches as in and its subtypes. the main analysis and none of the intercept estimates of MR-­Egger suggested directional pleiotropy (Table S4). Mendelian Randomization In multivariable MR to investigate potential pleiotropy We selected 31 uncorrelated genetic variants as instru- through risk factors common to both COVID-­19 and mental variables for liability to critical COVID-­19. These CVD, there was little evidence for attenuation of the are detailed in Table S3, along with their associations size of the estimate in any of these analyses (Figure S4), with ischemic stroke and its subtypes. MR estimates which was confirmed by likelihood ratio test (Table S5). are presented in Figure 2. In a univariable MR analysis, genetically proxied liability to critical COVID-­19 was as- Sensitivity Analysis Based on Genome-­ sociated with all-­cause ischemic stroke (OR 1.03, 95% Wide Significant Genetic Variants CI 1.00 to 1.06, P-­value=0.03). Restricting to ischemic As an additional sensitivity analysis, we used a more stroke subtypes, there were similar MR estimates for stringent P-­value threshold based on genome-­ wide cardioembolic stroke (OR 1.06, 95% CI 1.01 to 1.12, P-­ significance to select genetic variants as instrumental value=0.03), large artery stroke (OR 1.07, 95% CI 1.00 variables. We identified 9 uncorrelated genetic variants to 1.14, P-­value=0.06) and small-­ vessel stroke (OR that associated with critical COVID-­19 at genome-­wide 1.05, 95% CI 1.00 to 1.11, P-­value=0.06). Power cal- significance (P-­value <5×10−8). This MR analysis based culations shown in Figure S2 supported lower power on fewer variants generated consistent estimates to for ischemic stroke subtypes compared to all-­cause the main analysis, but with wider CIs that crossed the ischemic stroke. For the MR estimates generated by null, reflective of lower statistical power. Results are different variants, we observed heterogeneity greater displayed in Figure S5. than would be expected by chance only for cardioem- bolic stroke (heterogeneity P-­value=0.049), but none of Comparison With Other COVID-­19 the other considered ischemic stroke categories. Definitions We further considered other COVID-­ 19 definitions Mendelian Randomization Sensitivity (Figure S6 and Table S6). Genetically predicted Analyses COVID-­19 requiring hospitalization as compared to not Diagnostic scatterplots for ischemic stroke outcomes requiring hospitalization was associated with increased are presented in Figure S3. We observed consistent risk of any ischemic stroke (OR 1.05, 95% CI 1.01 to MR estimates for ischemic stroke risk in sensitivity 1.10, P-­value=0.01) and small-­vessel stroke (OR 1.22, Figure 1. Genetic correlation coefficient (x-­a xis) with 95% CI for critical COVID-­19 and ischemic stroke, related cardiovascular disease outcomes, and risk factors for both COVID-­ 19 and cardiovascular disease (y-­a xis), estimated by cross-­t rait linkage disequilibrium score regression. Multiple testing adjustment using the Benjamini and Yekutieli false discovery rate (FDR) are given on the right. J Am Heart Assoc. 2021;10:e022433. DOI: 10.1161/JAHA.121.0224335 Zuber et al COVID-­19 and Ischemic Stroke Figure 2. Forest plot illustrating the Mendelian randomization estimates of liability to critical COVID-­19 with stroke outcomes based on inverse-­variance weighted Mendelian randomization using genetic variants that were associated with critical COVID-­19 at a P-­value level of 5×10 -­6 or smaller. Mendelian randomization estimates represent the odds ratio of ischemic stroke outcomes per unit increase in the log-­odds ratio of liability to critical COVID-­19. Additional columns include the Mendelian randomization estimate, its 95% CI, the P-­value of the inverse-­variance weighted Mendelian randomization estimate to be different from 1 (P-­val), and heterogeneity measured by the Q-­statistic and the respective heterogeneity P-­value (Het-­P). Outcomes included any ischemic stroke, cardioembolic stroke, large artery stroke, and small vessel stroke. 95% CI 1.11 to 1.34, P-­value=0.000055). Considering to shared risk factors, such as obesity, smoking, and reported COVID-­19 (laboratory confirmed, physician-­ chronic inflammation. Furthermore, there was no MR reported or self-­reported) versus controls from the evidence that genetic liability to ischemic stroke in- general population, this was associated with increased creases risk of critical COVID-­19. risk of any ischemic stroke (OR 1.13, 95% CI 1.01 to When we considered critical COVID-­19 and isch- 1.26, P-­value=0.04) and large artery stroke (OR 1.46, emic stroke subtypes, only cardioembolic stroke 95% CI 1.18 to 1.81, P-­value=0.00042). remained statistically significant. There are 2 main possible explanations for this observation. The first is that there are pathophysiological differences across Bidirectional MR stroke subtypes. Indeed, it is conceivable that the host There was no strong evidence to support that genetic response to critical COVID-­19 may be more likely to liability to any of the considered ischemic stroke out- culminate in cardioembolic stroke through an acute, comes was associated with increased risk of critical pro-­inflammatory, hypercoagulable state that results in COVID-­19, as illustrated in Figure 3. cardiac thromboembolism; as opposed to occlusion of small penetrating cerebral arteries and small vessel strokes which are generally a result of more longstand- DISCUSSION ing conditions such as hypertension or diabetes.41 The In this study, we used cross-­trait LDSC to explore the second possible explanation is that smaller sample genetic correlation of critical COVID-­19 with ischemic sizes for each stroke subtype compared to any isch- stroke, other CVD outcomes, and risk factors com- emic stroke may mean that we were unable to detect mon to both. We identified a genetic correlation be- a statistically significance difference in stroke subtypes tween critical COVID-­ 19 and ischemic stroke, and due to reduced power as shown in Figure S2. This performed MR analyses that found genetic liability to explanation would be supported by similar point esti- critical COVID-­19 to be associated with increased risk mates to any ischemic stroke across stroke subtypes. of ischemic stroke. Notably, there was no evidence To date, studies assessing the incidence of isch- to support that these associations were attributable emic stroke during the COVID-­ 19 pandemic have J Am Heart Assoc. 2021;10:e022433. DOI: 10.1161/JAHA.121.0224336 Zuber et al COVID-­19 and Ischemic Stroke Figure 3. Forest plot of the bidirectional Mendelian randomization analysis illustrating the inverse-­variance weighted Mendelian randomization estimate of liability to stroke phenotypes with critical COVID-­19. Genetic variants which were associated with the stroke phenotypes were selected as instrumental variables at a P-­value level of 5×10 -­6 or smaller. Mendelian randomization estimates represent the odds ratio of critical COVID-­19 per unit increase in the log odds ratio of stroke phenotype. Additional columns include the Mendelian randomization estimate, its 95% CI, the P-­value of the inverse-­variance weighted Mendelian randomization estimate to be different from 1 (P-­val), and heterogeneity measured by the Q-­statistic and the respective heterogeneity P-­value (Het-­P). Exposures included any ischemic stroke, cardioembolic stroke, large artery stroke, and small vessel stroke. produced contrasting findings. It has also been un- which contributes to increased ischemic stroke risk. certain whether any association between COVID-­19 However, it is also important to note that our study and ischemic stroke is due to a direct effect of SARS-­ design cannot directly inform on whether the SARS-­ CoV-­2 viral infection, the host response in COVID-­19, CoV-­2 virus itself also be increases ischemic stroke or both. On one hand, some studies demonstrate risk, irrespective of the host inflammatory response. that the likelihood of stroke is 7-­fold higher in people Mechanisms that increase risk of ischemic stroke in with COVID-­19 than with influenza,3 that COVID-­19 is patients with COVID-­19 are complex,5,15 and include associated with 21-­fold increased odds of in-­hospital systemic inflammation and endotheliopathy.15,43–­45 stroke compared to patients without COVID-­19,6 and COVID-­19 can trigger a cytokine storm with upregu- that stroke is the most common neurological/neuro- lation of pro-­inflammatory cytokines and chemokines psychiatric complication of COVID-­19.4 On the con- such as tumor necrosis factor-­α (TNF-­α), interleukin-­1 trary, other studies have demonstrated a reduced rate (IL-­1) and IL-­615. Endothelial inflammation can induce a of hospital admissions with stroke during the first wave microvascular and macrovascular endotheliopathy that of the pandemic compared to one year before.9 Two contributes to a pro-­thrombotic state.15,43 main hypotheses have been proposed as explana- While prophylactic low molecular weight hepa- tions for these contrasting findings. The first is that the rin is used to prevent thromboembolism in patients incidence of stroke declined during the first wave of with COVID-­ 19, more targeted approaches to pre- the pandemic and that COVID-­19 is not mechanisti- vent strokes are yet undefined.5,46 Moreover, the cally associated with stroke, and the second is that the REMAP-­ CAP (A Randomised, Embedded, Multi-­ observed reduction in stroke presentations was due factorial, Adaptive Platform Trial for Community-­ to a higher proportion of people with mild strokes not Acquired Pneumonia), ACTIV-­4 (Anti-­thrombotics for reaching stroke services.11,42 Adults Hospitalized With COVID-­ 19), and ATTACC We have leveraged large-­scale genetic data to ad- (Antithrombotic Therapy to Ameliorate Complications dress this and find that liability to critical COVID-­19 of COVID-­19) trials have recently reported that thera- is associated with increased risk of ischemic stroke. peutic doses of anticoagulation do not improve clinical Moreover, our results are consistent with the hy- outcome and may increase bleeding for people with pothesis that it is the host response in COVID-­ 19 COVID-­19 in the critical care setting. Previous work J Am Heart Assoc. 2021;10:e022433. DOI: 10.1161/JAHA.121.0224337 Zuber et al COVID-­19 and Ischemic Stroke using an MR approach anticipated a beneficial effect the exposure. While this can never be completely ex- of IL-­6 receptor inhibition on both risk of ischemic cluded, we employed methods that are robust to ge- stroke and severe COVID-­19.47,48 More recently, clinical netic confounding (pleiotropy) in a series of sensitivity trials have demonstrated that IL-­6 receptor inhibition analyses (including pleiotropy-­robust MR methods and can improve outcomes in patients hospitalized with accounting for measured pleiotropy using multivariable COVID-­19.49 Targeting the deleterious host immune MR) and the estimates were consistent with our main response through similar approaches may also help analyses. We cannot be certain that genetic associa- to reduce the risk of ischemic stroke and should be tions with liability to critical COVID-­19 accurately reflect further evaluated. the pathophysiological process that actually occurs Our findings also support the hypothesis that few during critical COVID-­19. For example, while genetic patients with minor strokes reached stroke services predisposition may place an individual at increased lia- during the first wave of the COVID-­19 pandemic.11 This bility to critical COVID-­19, it is not possible to determine is reinforced by data that demonstrate the reduction from our analyses whether that factor is involved in the in stroke admissions observed in some centers during pathophysiological response to COVID-­19. the first wave of the pandemic was driven mainly by In conclusion, we have found genetic evidence that a reduction in presentations with minor stroke syn- liability to critical COVID-­19 is associated with increased dromes.11 People with minor stroke are at high risk risk of ischemic stroke. Our results are consistent with of early recurrence50 and public health messaging the host response in critical COVID-­19 underlying this should encourage people to attend stroke services if relationship, and support the evaluation of strategies to they have any symptom of stroke during the COVID-­19 mitigate this. pandemic. We did not observe genetic correlation between critical COVID-­19 and other CVD outcomes such as ARTICLE INFORMATION coronary heart disease, heart failure or atrial fibrilla- Received May 11, 2021; accepted September 23, 2021. tion. Other studies have reported acute coronary syn- drome, heart failure, and arrhythmia in people with Affiliations Department of Epidemiology and Biostatistics, School of Public Health, COVID-­19.15 There are a number of possible explana- Imperial College London, London, UK (V.Z., D.G.); Dementia Research Institute tions for these findings, which at first may appear dis- at Imperial College London, London, UK (V.Z.); Institute of Cardiovascular cordant. First, myocardial injury and myocarditis may and Medical Sciences, University of Glasgow, UK (A.C., J.D.); Center for Genomic Medicine (E.P.M.); and McCance Center for Brain Health (E.P.M., be a more common cardiac manifestation of COVID-­19 C.D.A.), Massachusetts General Hospital, Boston, MA ; Program in Medical than coronary artery plaque rupture and thrombosis. and Population Genetics, Broad Institute of MIT and Harvard, Cambridge, Second, the pulmonary oedema that is observed in MA (E.P.M., C.D.A.); Department of Medical Genetics, School of Clinical Medicine, University of Cambridge, UK (L.B.); The Alan Turing Institute, people with critical COVID-­19 is usually accompanied London, UK (L.B.); Medical Research Council Biostatistics Unit, University of by acute respiratory distress syndrome and is mainly Cambridge, UK (L.B., S.B.); Stroke Pharmacogenomics and Genetics Group, regarded as non-­cardiogenic.15 Third, while arrhyth- Biomedical Research Institute, Sant Pau, Spain (I.F.); Department of Public Health and Primary Care, Cardiovascular Epidemiology Unit, University of mias are common manifestations of COVID-­19, these Cambridge, UK (S.B.); Department of Neurology, Brigham and Women’s may be triggered by acute myocardial injury and sys- Hospital, Boston, MA (C.D.A.); Clinical Pharmacology and Therapeutics temic factors (such as fever, sepsis, hypoxia, and elec- Section, Institute of Medical and Biomedical Education and Institute for Infection and Immunity, St George’s, University of London, London, UK trolyte imbalance), rather than atrial fibrillation due to a (D.G.); Clinical Pharmacology Group, Pharmacy and Medicines Directorate, chronic atrial cardiopathy, which is more likely to make St George’s University Hospitals NHS Foundation Trust, London, UK (D.G.); up cases in the pre-­pandemic atrial fibrillation GWAS.20 and Novo Nordisk Research Centre Oxford, , Oxford, UK (D.G.). Our current study has strengths. We have made ef- Acknowledgments ficient use of existing large-­scale data resources to ad- This work is dedicated to the memory of Maria Mion who died from COVID-­19 dress an important clinical issue in the context of the related complications. Leonardo Bottolo is very grateful to the doctors, nurses, and staff of the Ospedale di Oderzo (TV, Italy) for the high standard rapidly evolving global pandemic. A key strength of MR of care for his mother. analysis is the use of randomly allocated genetic variants to help overcome environmental confounding, which is Sources of Funding VZ is supported by UK Dementia Research Institute at Imperial College analogous to randomization of treatment allocation in London, which is funded by the Medical Research Council, Alzheimer’s clinical trials. This has helped to overcome some of the Society and Alzheimer’s Research UK (MC_PC_17114). The research leading limitations of previous observational studies (either ret- to these results has been conducted as part of the COVIRNA project (grant rospective or cross-­sectional) assessing the relationship agreement n°101016072) funded by the European Union’s Horizon 2020 Framework Programme for Research and Innovation. EPM is supported between COVID-­19 and ischemic stroke.3–­6 by the National Institutes of Health of the United States (R01NS103924, Our work also has limitations. A series of modelling U01NS069763). LB acknowledges the MRC grant MR/S02638X/1, The assumptions are made when using MR, in particular, BHF-­Turing Cardiovascular Data Science Awards 2017 and The Alan Turing Institute under the Engineering and Physical Sciences Research Council that the genetic variants do not affect the consid- grant EP/N510129/1. IF-­C is supported by Inmungen-­Cov2 project, Centro ered outcomes through pathways independent of Superior de Investigaciones Científicas (CSIC). SB is supported by a Sir J Am Heart Assoc. 2021;10:e022433. DOI: 10.1161/JAHA.121.0224338 Zuber et al COVID-­19 and Ischemic Stroke Henry Dale Fellowship jointly funded by the Wellcome Trust and the Royal traits, sepsis, and severe COVID-­19: a Mendelian randomization inves- Society (204623/Z/16/Z). This research was supported by the UKRI Medical tigation. Circulation. 2020;142:1791–­ 1793. doi: 10.1161/CIRCU​ L ATIO​ Research Council [MC_UU_00002/7] and the NIHR Cambridge Biomedical NAHA.120.050753 Research Centre (BRC-­1215-­20014). The views expressed are those of the 13. Shi K, Wood K, Shi F-­D, Wang X, Liu Q. Stroke-­induced immunosup- author(s) and not necessarily those of the NIHR or the Department of Health pression and poststroke infection. Stroke Vasc Neurol. 2018;3:34–­41. and Social Care. CDA is supported by the National Institutes of Health of doi: 10.1136/svn-­2017-­0 00123 the United States (R01NS103924, U01NS069763). DG is supported by the 14. Didelez V, Sheehan N. Mendelian randomization as an instrumen- British Heart Foundation Centre of Research Excellence at Imperial College tal variable approach to causal inference. Stat Methods Med Res. London (RE/18/4/34215) and by a National Institute for Health Research 2007;16:309–­330. doi: 10.1177/09622​80206​077743 Clinical Lectureship at St. George’s, University of London (CL-­2020-­16-­0 01). 15. Nishiga M, Wang DW, Han Y, Lewis DB, Wu JC. COVID-­19 and cardio- This research was funded in part by the Wellcome Trust. vascular disease: from basic mechanisms to clinical perspectives. Nat Rev Cardiol. 2020;17:543–­558. doi: 10.1038/s4156​9 -­020-­0413-­9 Disclosures 16. COVID-­19 Host Genetics Initiative. A global initiative to elucidate the role CDA receives sponsored research support from the American Heart of host genetic factors in susceptibility and severity of the SARS-­CoV-­2 Association, Massachusetts General Hospital, and Bayer AG, and has con- virus pandemic. Eur J Hum Genet. 2020;28:715–­ 718. doi: 10.1038/ sulted for ApoPharma, Inc. DG is employed part-­time by Novo Nordisk. The s4143​1-­020-­0636-­6 remaining authors have no disclosures to report. 17. COVID-­19 Host Genetics Initiative. Mapping the human genetic archi- tecture of COVID-­19. Nature. 2021. doi: 10.1038/s4158​6-­021-­03767​-­x Supplementary Material 18. Malik R, Chauhan G, Traylor M, Sargurupremraj M, Okada Y, Mishra Table S1–­S6 A, Rutten-­Jacobs L, Giese A-­K, van der Laan SW, Gretarsdottir S, Figure S1–­S6 et al. Multiancestry genome-­wide association study of 520,000 sub- jects identifies 32 loci associated with stroke and stroke subtypes. Nat Genet. 2018;50:524– ­537. doi: 10.1038/s4158​8-­018-­0 058-­3 19. Nikpay M, Goel A, Won HH, Hall LM, Willenborg C, Kanoni S, Saleheen D, Kyriakou T, Nelson CP, Hopewell JC, et al. A comprehensive 1000 REFERENCES Genomes–­based genome-­wide association meta-­analysis of coronary 1. Carter P, Anderson M, Mossialos E. Health system, public health, and artery disease. Nat Genet. 2015;47:1121–­1130. doi: 10.1038/ng.3396 economic implications of managing COVID-­19 from a cardiovascular 20. Shah S, Henry A, Roselli C, Lin H, Sveinbjörnsson G, Fatemifar G, perspective. Eur Heart J. 2020;41:2516–­2518. doi: 10.1093/eurhe​artj/ Hedman ÅK, Wilk JB, Morley MP, Chaffin MD, et al. Genome-­wide ehaa342. association and Mendelian randomisation analysis provide insights 2. Clerkin KJ, Fried JA, Raikhelkar J, Sayer G, Griffin JM, Masoumi A, Jain into the pathogenesis of heart failure. Nat Commun. 2020;11:163. doi: SS, Burkhoff D, Kumaraiah D, Rabbani L, et al. COVID-­19 and cardio- 10.1038/s4146​7-­019-­13690​-­5 vascular disease. Circulation. 2020;141:1648–­1655. doi: 10.1161/CIRCU​ 21. Roselli C, Chaffin MD, Weng L-­C, Aeschbacher S, Ahlberg G, Albert LATIO​NAHA.120.046941 CM, Almgren P, Alonso A, Anderson CD, Aragam KG, et al. Multi-­ 3. Merkler AE, Parikh NS, Mir S, Gupta A, Kamel H, Lin E, Lantos J, ethnic genome-­wide association study for atrial fibrillation. Nat Genet. Schenck EJ, Goyal P, Bruce SS, et al. Risk of ischemic stroke in patients 2018;50:1225–­1233. doi: 10.1038/s4158​8-­018-­0133-­9 with Coronavirus Disease 2019 (COVID-­19) vs patients with influenza. 22. Lindsberg PJ, Grau AJ. Inflammation and infections as risk factors JAMA Neurol. 2020;77:1–­7. doi: 10.1001/jaman​eurol.2020.2730 for ischemic stroke. Stroke. 2003;34:2518–­ 2532. doi: 10.1161/01. 4. Varatharaj A, Thomas N, Ellul MA, Davies NWS, Pollak TA, Tenorio EL, STR.00000​89015.51603.CC Sultan M, Easton A, Breen G, Zandi M, et al. Neurological and neuro- 23. Toyoda K, Ninomiya T. Stroke and cerebrovascular diseases in patients psychiatric complications of COVID-­19 in 153 patients: a UK-­wide sur- with chronic kidney disease. Lancet Neurol. 2014;13:823–­ 833. doi: veillance study. Lancet Psychiatry. 2020;7:875–­882. doi: 10.1016/S2215​ 10.1016/S1474​-­4422(14)70026​-­2 -­0366(20)30287​-­X 24. Kernan WN, Inzucchi SE, Sawan C, Macko RF, Furie KL. Obesity: a 5. Qureshi AI, Abd-­ Allah F, Al-­ Senani F, Aytac E, Borhani-­ Haghighi A, stubbornly obvious target for stroke prevention. Stroke. 2013;44:278–­ Ciccone A, Gomez CR, Gurkas E, Hsu CY, Jani V, et al. Management of 286. doi: 10.1161/STROK​E AHA.111.639922 acute ischemic stroke in patients with COVID-­19 infection: report of an 25. Wolf PA, D’Agostino RB, Kannel WB, Bonita R, Belanger AJ. Cigarette international panel. Int J Stroke. 2020;15:540–­554. doi: 10.1177/17474​ smoking as a risk factor for stroke. The Framingham Study. JAMA. 93020​923234 1988;259:1025–­1029. doi: 10.1001/jama.1988.03720​07002​5028 6. Katz JM, Libman RB, Wang JJ, Sanelli P, Filippi CG, Gribko M, Pacia 26. Pulit SL, Stoneman C, Morris AP, Wood AR, Glastonbury CA, Tyrrell J, SV, Kuzniecky RI, Najjar S, Azhar S. Cerebrovascular complica- Yengo L, Ferreira T, Marouli E, Ji Y, et al. Meta-­analysis of genome-­wide tions of COVID-­ 19. Stroke. 2020;51:e227–­ e231. doi: 10.1161/STROK​ association studies for body fat distribution in 694 649 individuals of EAHA.120.031265 European ancestry. Hum Mol Genet. 2019;28:166–­174. doi: 10.1093/ 7. Fuentes B, Alonso de Leciñana M, García-­ Madrona S, Díaz-­ Otero hmg/ddy327 F, Aguirre C, Calleja P, Egido JA, Carneado-­ Ruiz J, Ruiz-­ Ares G, 27. Wootton RE, Richmond RC, Stuijfzand BG, Lawn RB, Sallis HM, Rodríguez-­Pardo J, et al. Stroke acute management and outcomes Taylor GMJ, Hemani G, Jones HJ, Zammit S, Davey Smith G, et al. during the COVID-­19 outbreak. Stroke. 2021;52:552–­562. doi: 10.1161/ Evidence for causal effects of lifetime smoking on risk for depression STROK​E AHA.120.031769 and schizophrenia: a Mendelian randomisation study. Psychol Med. 8. Li Y, Li M, Wang M, Zhou Y, Chang J, Xian Y, Wang D, Mao L, Jin H, Hu 2020;50:2435–­2443. doi: 10.1017/S0033​29171​9002678 B. Acute cerebrovascular disease following COVID-­19: a single center, 28. Neale lab. Rapid GWAS of thousands of phenotypes in the UK Biobank retrospective, observational study. Stroke Vasc Neurol. 2020;5:279–­ 2020. http://www.neale​lab.is/uk-­bioba​nk/ 284. doi: 10.1136/svn-­2020-­0 00431 29. Bulik-­Sullivan B, Finucane HK, Anttila V, Gusev A, Day FR, Loh PR, 9. Sacco S, Ricci S, Ornello R, Eusebi P, Petraglia L, Toni D, Rota E, Duncan L, Perry JRB, Patterson N, Robinson EB, et al. An atlas of Bruzzone G, Testa L, Bongioanni R, et al. Reduced admissions for genetic correlations across human diseases and traits. Nat Genet. cerebrovascular events during COVID-­ 19 outbreak in Italy. Stroke. 2015;47:1236–­1241. doi: 10.1038/ng.3406 2020;51:3746– ­3750. doi: 10.1161/STROK​E AHA.120.031293 30. Benjamini Y, Yekutieli D. The control of the false discovery rate in mul- 10. Qureshi AI, Baskett WI, Huang W, Shyu D, Myers D, Raju M, Lobanova tiple testing under dependency. Ann Stat. 2001;29:1165–­ 1188. doi: I, Suri MFK, Naqvi SH, French BR, et al. Acute ischemic stroke and 10.1214/aos/10136​99998 COVID-­19: an analysis of 27 676 patients. Stroke. 2021;52:905–­912. 31. Pierce BL, Burgess S. Efficient design for mendelian randomization doi: 10.1161/STROK​E AHA.120.031786 studies: subsample and 2-­sample instrumental variable estimators. Am 11. Perry R, Banaras A, Werring DJ, Simister R. What has caused the J Epidemiol. 2013;178:1177–­1184. doi: 10.1093/aje/kwt084 fall in stroke admissions during the COVID-­19 pandemic? J Neurol. 32. Del Greco MF, Minelli C, Sheehan NA, Thompson JR. Detecting plei- 2020;267:3457–­3458. doi: 10.1007/s0041​5-­020-­10030​-­2 otropy in Mendelian randomisation studies with summary data and 12. Ponsford MJ, Gkatzionis A, Walker VM, Grant AJ, Wootton RE, Moore a continuous outcome. Stat Med. 2015;34:2926–­2940. doi: 10.1002/ LSP, Fatumo S, Mason AM, Zuber V, Willer C, et al. Cardiometabolic sim.6522 J Am Heart Assoc. 2021;10:e022433. DOI: 10.1161/JAHA.121.0224339 Zuber et al COVID-­19 and Ischemic Stroke 33. Bowden J, Davey Smith G, Haycock PC, Burgess S. Consistent esti- 42. Aguiar de Sousa D, Sandset EC, Elkind MSV. The curious case of the mation in Mendelian randomization with some invalid instruments using missing strokes during the COVID-­19 pandemic. Stroke. 2020;51:1921–­ a weighted median estimator. Genet Epidemiol. 2016;40:304–­314. doi: 1923. doi: 10.1161/STROK​E AHA.120.030792 10.1002/gepi.21965 43. Varga Z, Flammer AJ, Steiger P, Haberecker M, Andermatt R, 34. Bowden J, Davey Smith G, Burgess S. Mendelian randomization with Zinkernagel AS, Mehra MR, Schuepbach RA, Ruschitzka F, Moch invalid instruments: effect estimation and bias detection through Egger H. Endothelial cell infection and endotheliitis in COVID-­ 19. Lancet. regression. Int J Epidemiol. 2015;44:512–­525. doi: 10.1093/ije/dyv080 2020;395:1417–­1418. doi: 10.1016/S0140​-­6736(20)30937​-­5 35. Bowden J, Smith GD, Haycock PC, Burgess S. Consistent estima- 44. Lee SG, Fralick M, Sholzberg M. Coagulopathy associated with tion in mendelian randomization with some invalid instruments using COVID-­19. Can Med Ass J. 2020;192:E583. doi: 10.1503/cmaj.200685 a weighted median estimator. Genet Epidemiol. 2016;40:304–­314. doi: 45. Fauvel C, Weizman O, Trimaille A, Mika D, Pommier T, Pace N, Douair A, 10.1002/gepi.21965 Barbin E, Fraix A, Bouchot O, et al. Pulmonary embolism in COVID-­19 36. Burgess S, Thompson SG. Multivariable Mendelian randomization: patients: a French multicentre cohort study. Eur Heart J. 2020;41:3058–­ the use of pleiotropic genetic variants to estimate causal effects. Am J 3068. doi: 10.1093/eurhe​artj/ehaa500 Epidemiol. 2015;181:251–­260. doi: 10.1093/aje/kwu283 46. Robinson RG, Bolduc PL, Price TR. Two-­year longitudinal study of post- 37. Burgess S, Thompson DJ, Rees JMB, Day FR, Perry JR, Ong KK. stroke mood disorders: diagnosis and outcome at one and two years. Dissecting causal pathways using Mendelian randomization with sum- Stroke. 1987;18:837–­843. doi: 10.1161/01.STR.18.5.837 marized genetic data: application to age at menarche and risk of breast 47. Georgakis MK, Malik R, Gill D, Franceschini N, Sudlow CLM, Dichgans cancer. Genetics. 2017;207:481–­487. doi: 10.1534/genet​ics.117.300191 M. Interleukin-­6 signaling effects on ischemic stroke and other cardio- 38. Burgess S. Sample size and power calculations in Mendelian random- vascular outcomes: a Mendelian randomization study. Circ Genom ization with a single instrumental variable and a binary outcome. Int J Precis Med. 2020;13:e002872. doi: 10.1161/CIRCG​EN.119.002872 Epidemiol. 2014;43:922–­929. doi: 10.1093/ije/dyu005 48. Larsson S, Burgess S, Gill D. Genetically proxied inhibition of interleu- 39. Burgess S, Thompson SG, Collaboration CRPCHDG. Avoiding bias kin-­6 signaling: opposing associations with susceptibility to COVID-­19 from weak instruments in Mendelian randomization studies. Int J and pneumonia. Eur Respir J. 2021;57:2003545. Epidemiol. 2011;40:755–­764. doi: 10.1093/ije/dyr036 49. Salama C, Han J, Yau L, Reiss WG, Kramer B, Neidhart JD, Criner GJ, 40. Yavorska OO, Burgess S. MendelianRandomization: an R package for Kaplan-­Lewis E, Baden R, Pandit L, et al. Tocilizumab in patients hos- performing Mendelian randomization analyses using summarized data. pitalized with COVID-­19 pneumonia. New Engl J Med. 2021;384:20–­30. Int J Epidemiol. 2017;46:1734–­1739. doi: 10.1093/ije/dyx034 doi: 10.1056/NEJMo​a 2030340 41. Adams HP, Bendixen BH, Kappelle LJ, Biller J, Love BB, Gordon DL, 50. Coull AJ, Lovett JK, Rothwell PM. Population based study of early risk Marsh EE. Classification of subtype of acute ischemic stroke. Definitions of stroke after transient ischaemic attack or minor stroke: implications for use in a multicenter clinical trial. TOAST. Trial of Org 10172 in acute for public education and organisation of services. BMJ. 2004;328:326. stroke treatment. Stroke. 1993;24:35–­41. doi: 10.1161/01.STR.24.1.35 doi: 10.1136/bmj.37991.635266.44 J Am Heart Assoc. 2021;10:e022433. DOI: 10.1161/JAHA.121.02243310 Supplemental Material Table S1: Overview of the publicly available summary-level data of genetic associations used for the analysis. Sample Phenotype Description Cases Controls Population Pubmed Author Size Main: Critical Covid vs. population 5,101 1,383,241 The COVID-19 Sensitivity Analysis 1: Hospitalized Covid vs. not hospitalized 4,829 11,816 Host Genetics Covid Covid-19 EUR Initiative (2021) Sensitivity Analysis 2: Hospitalized Covid vs. population 9,986 1,877,672 Sensitivity Analysis 3: Covid vs population 38,984 1,644,784 Any ischemic stroke 34,217 406,111 Large artery stroke 4,373 406,111 Stroke EUR 29531354 Malik et al. (2018) Cardioembolic stroke 7,193 406,111 Small vessel stroke 5,386 406,111 Coronary artery disease 60,801 123,504 EUR, SEA 26343387 Nikpay et al. (2015) Cardiovascular disease Heart failure 47,309 930,014 EUR 31919418 Shah et al. (2020) outcomes Atrial fibrillation 65,446 522,744 TRANS 29892015 Roselli et al. (2018) Obesity Body mass index 694,649 EUR 30239722 Pullit et al. (2019) Wootton et al. Smoking Lifetime smoking index 462,690 EUR 30239722 (2020) Inflammation C-reactive protein 361,194 EUR Neale lab Table S2. Cross-trait linkage disequilibrium score regression (LDSC) analysis results of critical Covid-19 with ischemic stroke, cardiovascular disease (CVD) outcomes, and risk factors related to both Covid-19 and CVD. rg represents the genetic correlation between critical Covid-19 and each phenotype. p-values are corrected for multiple testing with the Benjamini and Hochberg false discovery rate (FDR). rg FDR-p value Ischemic stroke 0.2922 4.65E-03 Coronary artery disease 0.1914 7.60E-02 Heart failure 0.1491 1.12E-01 Atrial fibrillation 0.0198 7.17E-01 Body mass index 0.2088 6.26E-06 Smoking 0.1029 5.91E-02 C-reactive protein 0.2006 1.35E-04 Table S3: Overview of the genetic variants used as instrumental variables for liability to critical Covid-19 based on the Covid-19 host genetics initiative. We selected 31 uncorrelated (clumped at correlation threshold r2 < 0.01) genetic variants as instrumental variables for laibility to critical Covid-19 that were associated at a p-value level of 5 × 10-6 or smaller. The table additionally includes summary-level data (beta coefficients of genetic association, their standard error and corresponding p-value) for liability to critical Covid-19 as exposure and any ischemic stroke (AIS) and its subtypes cardioembolic stroke (CES), large artery stroke (LAS), and small vessel stroke (SVS) as outcomes. Alt: alternative allele; Chr: chromosome; Pos: position; Ref: reference allele; SE: standard error. Critical Covid-19 AIS CES LAS SVS Variant Chr Pos Ref Alt Beta SE P Beta SE P Beta SE P Beta SE P Beta SE P rs10087754 8 121819908 T A -1.3E-01 2.7E-02 7.1E-07 -1.2E-02 1.0E-02 2.4E-01 1.0E-04 1.9E-02 9.9E-01 -4.2E-02 2.5E-02 9.6E-02 -9.6E-03 2.3E-02 6.8E-01 rs11085727 19 10355447 C T 1.7E-01 2.9E-02 3.7E-09 -2.2E-02 1.1E-02 5.2E-02 -5.5E-02 2.2E-02 1.1E-02 -5.0E-04 2.8E-02 9.8E-01 3.5E-02 2.5E-02 1.7E-01 rs111508230 1 155181061 C T -2.1E-01 4.4E-02 2.6E-06 -8.1E-03 1.6E-02 6.1E-01 -7.1E-02 3.1E-02 2.4E-02 9.3E-02 3.9E-02 1.7E-02 2.9E-02 4.1E-02 4.8E-01 rs114969787 5 65770656 C T 3.1E-01 6.6E-02 3.6E-06 4.1E-02 3.1E-02 1.9E-01 -4.7E-02 6.4E-02 4.7E-01 1.2E-01 7.7E-02 1.4E-01 4.4E-02 7.2E-02 5.5E-01 rs11658357 17 36097317 A T -2.0E-01 4.4E-02 4.9E-06 9.7E-03 1.3E-02 4.4E-01 -1.8E-02 2.4E-02 4.5E-01 -1.5E-02 3.1E-02 6.1E-01 4.4E-02 2.9E-02 1.3E-01 rs117232645 13 74553195 G A -3.3E-01 7.0E-02 2.5E-06 -3.1E-02 3.0E-02 3.0E-01 1.0E-02 5.9E-02 8.6E-01 4.4E-02 7.3E-02 5.4E-01 3.8E-03 6.8E-02 9.6E-01 rs13050728 21 33242905 T C -2.0E-01 2.9E-02 2.4E-12 -4.4E-03 1.1E-02 6.8E-01 -2.9E-02 2.0E-02 1.5E-01 -2.7E-02 2.6E-02 3.1E-01 2.7E-02 2.5E-02 2.8E-01 rs13080258 3 69672908 A C -1.4E-01 3.0E-02 3.6E-06 2.8E-03 1.2E-02 8.1E-01 6.8E-03 2.3E-02 7.6E-01 -1.6E-02 2.9E-02 5.9E-01 -1.8E-02 2.7E-02 5.1E-01 rs13274496 8 22583385 G A -2.0E-01 4.2E-02 1.8E-06 7.0E-04 1.3E-02 9.6E-01 1.7E-02 2.4E-02 4.8E-01 -1.9E-02 3.2E-02 5.5E-01 -3.2E-02 2.9E-02 2.6E-01 rs143334143 6 31153649 G A 2.9E-01 4.3E-02 6.0E-12 9.0E-03 1.9E-02 6.4E-01 3.6E-02 3.6E-02 3.1E-01 -1.0E-02 4.8E-02 8.3E-01 3.6E-02 4.6E-02 4.3E-01 rs1974792 19 50353078 A G -1.4E-01 2.7E-02 5.1E-07 -2.0E-02 1.0E-02 4.7E-02 -9.6E-03 2.0E-02 6.3E-01 -6.3E-02 2.5E-02 1.3E-02 3.2E-02 2.4E-02 1.8E-01 rs2109069 19 4719431 G A 2.6E-01 2.8E-02 6.1E-20 -7.1E-03 1.1E-02 5.2E-01 2.0E-03 2.2E-02 9.3E-01 -6.2E-03 2.7E-02 8.2E-01 -1.1E-02 2.6E-02 6.8E-01 rs2237698 7 107967457 C T 2.4E-01 4.0E-02 2.4E-09 -2.7E-02 2.1E-02 1.9E-01 -2.7E-02 3.9E-02 4.8E-01 4.8E-02 5.1E-02 3.5E-01 -2.3E-02 4.7E-02 6.3E-01 rs2597569 11 97922951 T C -1.8E-01 3.4E-02 9.3E-08 -7.0E-03 1.0E-02 4.9E-01 -1.5E-02 1.9E-02 4.5E-01 -3.3E-02 2.5E-02 2.0E-01 -3.5E-02 2.4E-02 1.4E-01 rs2733839 12 10393411 T C 2.7E-01 5.8E-02 3.5E-06 -3.9E-03 2.7E-02 8.9E-01 4.3E-02 5.3E-02 4.2E-01 -4.1E-02 7.2E-02 5.7E-01 3.0E-02 6.4E-02 6.4E-01 rs340850 1 213941523 T G -2.8E-01 6.0E-02 4.3E-06 2.0E-02 2.4E-02 4.0E-01 5.6E-02 4.8E-02 2.5E-01 -9.7E-02 5.7E-02 9.0E-02 -5.3E-02 5.5E-02 3.3E-01 rs35081325 3 45848429 A T 6.3E-01 4.5E-02 5.8E-45 3.0E-02 2.0E-02 1.3E-01 4.3E-02 3.8E-02 2.6E-01 5.4E-02 5.1E-02 2.8E-01 2.7E-02 4.9E-02 5.9E-01 rs36932 7 123877938 G A -1.6E-01 3.4E-02 4.0E-06 -6.2E-03 1.4E-02 6.5E-01 -2.1E-02 2.6E-02 4.3E-01 -3.2E-02 3.3E-02 3.3E-01 -4.8E-02 3.1E-02 1.2E-01 rs4076440 1 9630418 A G 2.1E-01 4.3E-02 9.4E-07 -1.5E-02 2.1E-02 4.7E-01 4.2E-03 4.1E-02 9.2E-01 -5.8E-02 5.3E-02 2.7E-01 5.1E-03 4.8E-02 9.2E-01 rs5767981 22 47769327 A G -1.6E-01 3.4E-02 5.0E-06 -2.0E-02 1.1E-02 5.7E-02 -7.9E-03 2.1E-02 7.0E-01 -3.0E-03 2.7E-02 9.1E-01 -2.6E-02 2.5E-02 2.9E-01 rs622568 7 54580201 A C 2.3E-01 3.7E-02 1.0E-09 3.5E-03 1.4E-02 8.0E-01 5.2E-02 2.6E-02 4.4E-02 -5.7E-02 3.4E-02 9.1E-02 -1.7E-02 3.1E-02 5.8E-01 rs633862 9 133279871 T C -1.7E-01 3.4E-02 1.1E-06 -2.6E-02 1.0E-02 1.2E-02 -5.5E-02 1.9E-02 3.8E-03 -5.1E-02 2.4E-02 3.7E-02 -1.0E-04 2.3E-02 1.0E+00 rs6478109 9 114806486 A G 1.5E-01 2.8E-02 2.4E-07 1.2E-02 1.1E-02 2.5E-01 2.2E-02 2.1E-02 2.8E-01 6.0E-02 2.7E-02 2.7E-02 4.4E-02 2.5E-02 7.4E-02 rs6712600 2 125728622 G A -1.7E-01 3.4E-02 3.3E-07 6.9E-03 1.3E-02 5.8E-01 5.0E-02 2.4E-02 3.7E-02 -9.4E-03 3.1E-02 7.6E-01 -3.1E-02 2.9E-02 2.8E-01 rs7135260 12 112943944 T C 1.9E-01 2.8E-02 6.1E-12 2.6E-02 1.1E-02 1.5E-02 0.0E+00 2.1E-02 1.0E+00 4.0E-02 2.7E-02 1.4E-01 4.1E-02 2.5E-02 1.0E-01 rs77406469 7 150758864 C T -3.4E-01 6.9E-02 6.5E-07 1.8E-02 2.7E-02 5.1E-01 -4.3E-02 6.0E-02 4.7E-01 5.4E-02 7.6E-02 4.8E-01 6.6E-02 6.8E-02 3.3E-01 rs77534576 17 49863303 C T 4.6E-01 7.5E-02 8.5E-10 2.8E-02 3.1E-02 3.7E-01 5.4E-02 6.6E-02 4.1E-01 7.3E-02 8.2E-02 3.7E-01 1.3E-01 7.6E-02 7.6E-02 rs79833209 5 163300447 C T 4.4E-01 9.2E-02 2.2E-06 1.0E-01 3.6E-02 3.9E-03 8.1E-02 7.9E-02 3.0E-01 -8.4E-03 9.9E-02 9.3E-01 -2.0E-02 9.0E-02 8.2E-01 rs9287218 1 237113798 A C -3.2E-01 6.2E-02 3.9E-07 2.8E-03 2.4E-02 9.1E-01 2.3E-02 4.7E-02 6.3E-01 1.2E-02 6.2E-02 8.5E-01 -4.9E-02 5.8E-02 4.0E-01 rs9577175 13 112889041 C T 2.0E-01 4.1E-02 7.9E-07 -4.0E-03 1.2E-02 7.3E-01 3.8E-03 2.3E-02 8.7E-01 -2.5E-02 3.0E-02 4.0E-01 1.2E-02 2.8E-02 6.7E-01 rs9871880 3 197399535 C T -2.5E-01 5.0E-02 4.1E-07 -2.7E-03 1.8E-02 8.8E-01 -1.0E-01 3.6E-02 4.1E-03 5.4E-02 4.4E-02 2.2E-01 1.0E-02 4.2E-02 8.1E-01 Table S4: Sensitivity analysis for the Mendelian randomization analysis of liability to critical Covid-19 on ischemic stroke outcomes including the inverse-variance weighted (IVW) Mendelian randomization and pleiotropy-robust Mendelian randomization approaches (simple, weighted median and MR-Egger). Mendelian randomization estimates represent the odds ratio for ischemic stroke outcomes per unit increase in the log-odds ratio of liability to critical Covid-19. In addition to the Mendelian randomization estimates, we included their 95% confidence interval (CI) and corresponding p-value. The intercept of the MR-Egger method was used to test for directional pleiotropy. Instrument selection was based on genetic variants that were associated with liability to critical Covid-19 with a p-value equal to or smaller than 5 × 10-6. Main outcome was any ischemic stroke, and we further included the subtypes cardioembolic stroke, large artery stroke, and small vessel stroke. 95% CI 95% CI Outcome Method Estimate p-value Lower Upper IVW 1.031 1.004 1.058 0.027 Simple median 1.022 0.986 1.059 0.227 Any ischemic Weighted median 1.031 0.995 1.068 0.088 stroke MR-Egger 1.022 0.947 1.103 0.573 MR-Egger intercept 0.002 -0.014 0.018 0.820 IVW 1.060 1.005 1.119 0.034 Simple median 1.072 1.001 1.147 0.047 Cardio- embolic Weighted median 1.072 1.001 1.147 0.046 stroke MR-Egger 1.092 0.935 1.276 0.266 MR-Egger intercept -0.007 -0.040 0.026 0.689 IVW 1.067 0.997 1.141 0.059 Simple median 1.079 0.988 1.180 0.091 Large artery Weighted median 1.087 0.994 1.188 0.066 stroke MR-Egger 0.919 0.762 1.108 0.374 MR-Egger intercept 0.034 -0.006 0.073 0.095 IVW 1.054 0.999 1.112 0.055 Simple median 1.075 0.990 1.167 0.087 Small vessel Weighted median 1.051 0.968 1.140 0.237 stroke MR-Egger 1.012 0.867 1.182 0.879 MR-Egger intercept 0.009 -0.024 0.042 0.585 Table S5: Likelihood ratio test to compare the model fit of the multivariable Mendelian randomization model considering risk for critical Covid-19 as exposure for ischemic stroke outcomes accounting for potential pleiotropic pathways (including life-time smoking index, body mass index, and c-reactive protein) with the univariable Mendelian randomization model. Model fit is evaluated using residual sum of squares for the univariable Mendelian randomization model (RSS 1) with the residual sum of squares for the multivariable Mendelian randomization model (RSS 2). There was one degree of freedom difference between the multivariable and the univariable Mendelian randomization model because there is one additional parameter to estimate in the multivariable Mendelian randomization model. The F-statistic quantifies the reduction in residual sum of squares by adding the pleiotropic risk factor to the Mendelian randomization model. The respective p-value tests if the multivariable Mendelian randomization model including the pleiotropic pathways provides a significantly better model fit of the genetic association estimates with the stroke outcome than the univariable Mendelian randomization model. Instrument selection was based on genetic variants that were associated with liability to critical Covid-19 with a p-value equal to or smaller than 5 × 10-6. Main outcome was any ischemic stroke, and we further included the subtypes cardioembolic stroke, large artery stroke, and small vessel stroke. Outcome Pleiotropic pathway RSS 1 RSS 2 F-statistic p-value Smoking 40.230 39.436 0.584 0.451 Any ischemic Body mass index 40.230 36.919 2.601 0.118 stroke C-reactive protein 37.457 34.841 2.103 0.158 Smoking 43.899 42.439 0.998 0.326 Cardio- embolic Body mass index 43.899 43.733 0.110 0.743 stroke C-reactive protein 42.765 37.544 3.894 0.058 Smoking 41.290 40.612 0.483 0.492 Large artery Body mass index 41.290 41.290 0.000 0.997 stroke C-reactive protein 40.872 39.648 0.865 0.360 Smoking 26.742 26.150 0.656 0.425 Small vessel Body mass index 26.742 23.993 3.322 0.079 stroke C-reactive protein 26.177 26.176 0.001 0.978 Table S6: Mendelian randomization estimates from the inverse-variance weighted Mendelian randomization analysis considering different Covid-19 phenotypes as exposure for ischemic stroke subtypes. Covid-19 phenotypes were based on the definitions by the Covid-19 host genetics initiative. Mendelian randomization estimates represent the odds ratio for ischemic stroke outcomes per unit increase in the log-odds ratio of liability to the respective Covid-19 definition. In addition to the Mendelian randomization estimates, we included their 95% confidence interval (CI) and corresponding p-value. Instrument selection was based on genetic variants that were associated with the respective Covid-19 definition with a p-value equal to or smaller than 5 × 10-6. Moreover, we displayed heterogeneity measured by the Q-statistic and the corresponding heterogeneity p-value (Het. p-value). Main outcome was any ischemic stroke, and we further included the subtypes cardioembolic stroke, large artery stroke, and small vessel stroke. 95% CI 95% CI Q- Het. Exposure Outcome Estimate p-value Lower Upper statistic p-value Any ischemic stroke 1.054 1.012 1.099 0.011 5.311 0.915 Hospitalized for Cardioembolic Covid-19 versus 1.044 0.963 1.133 0.294 5.062 0.928 stroke controls with laboratory- Large artery stroke 1.061 0.957 1.177 0.258 5.080 0.927 confirmed Covid-19 Small vessel stroke 1.219 1.107 1.342 5.5×10-5 7.964 0.717 Any ischemic stroke 1.026 0.981 1.073 0.268 46.406 0.021 Cardioembolic Hospitalized for 1.090 0.991 1.198 0.078 56.523 0.002 stroke Covid-19 versus population controls Large artery stroke 1.081 0.978 1.194 0.128 37.369 0.137 Small vessel stroke 0.991 0.912 1.078 0.841 30.302 0.399 Any ischemic stroke 1.126 1.005 1.262 0.041 59.811 1.1×10-4 Cardioembolic Reported Covid-19 1.158 0.960 1.396 0.125 44.238 0.010 stroke versus population controls Large artery stroke 1.464 1.184 1.811 4.2×10-4 33.982 0.108 Small vessel stroke 1.043 0.879 1.237 0.629 25.093 0.457 Cross-trait linkage disequilibrium score regression: Primary outcome ischemic stroke, and secondary outcomes coronary artery disease, heart failure and atrial fibrillation, and genetic confounders (BMI, lifetime smoking index and CRP). FDR-corrected p-values <0.05 are taken forward. Main Mendelian randomization analysis: Exposure: Critical Covid-19 vs population Outcome: Ischemic stroke and subtypes Instrumental variables: Uncorrelated (r2<0.01 ) genetic variants associated at p-value <5 x10-6 with the exposure Method: Two-sample inverse-variance weighted Mendelian randomization Sensitivity analysis Robust Accounting Restricting Comparison Bidirectional methods for pleiotropic instrumental with other Mendelian Weighted pathways variables Covid-19 randomization median Multivariable Instrumental definitions Exposure: MR-Egger Mendelian variables: 1: Hospitalized Ischemic stroke randomization Uncorrelated Covid-19 vs and subtypes accounting for (r2<0.01 ) population Outcome: c BMI, lifetime genetic 2: Hospitalized Critical Covid- smoking index, variants Covid-19 vs 19 vs and CRP). associated at confirmed population p-value <5 Covid-19 x10-8 with the 3: Confirmed exposure Covid-19 vs population Figure S1: Illustration of the analytical plan. Figure S2: Power calculation showing power (y-axis) as a function of Mendelian randomization estimates (odds ratio, x-axis). Different line types describe different case and control numbers representing ischemic stroke and its subtype. See Table S 1 for the exact numbers of cases and control for each outcome. Note that the odds ratio is per one standard deviation unit increase in the exposure, which is not comparable to the Mendelian randomization odds ratios presented in the manuscript which are per unit increase in the log odds ratio of the exposure. A B 0.2 Genetic association with CES Genetic association with AIS ● 0.1 0.1 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.0 ● ●●● ● ● ●● ●● ● ● ●● ● ● ● ● ● ● 0.0 ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● −0.1 ● −0.1 −0.50 −0.25 0.00 0.25 0.50 0.75 −0.50 −0.25 0.00 0.25 0.50 0.75 Genetic association with critical Covid−19 Genetic association with critical Covid−19 C D 0.3 0.2 0.2 Genetic association with SVS Genetic association with LAS ● ● 0.1 ● 0.1 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0.0 ● ● ● ● ● ● ● ● ● ● ● ● ● 0.0 ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●●● ● ●● ● ● ● −0.1 ● −0.1 −0.2 −0.2 −0.50 −0.25 0.00 0.25 0.50 0.75 −0.50 −0.25 0.00 0.25 0.50 0.75 Genetic association with critical Covid−19 Genetic association with critical Covid−19 Figure S3: Scatterplots for diagnostics plot the genetic association of the 31 genetic variants used as instrumental variables with the exposure (liability to critical Covid-19) on the x-axis against genetic associations with the outcome (ischemic stroke phenotypes) on the y-axis. Error bars indicate the standard error of the genetic association. The inverse-variance weighted Mendelian randomization estimate is represented by the red dashed line through the origin. Each panel shows main outcome: Panel A) any ischemic stroke (AIS), B) cardioembolic stroke (CES), C) large artery stroke (LAS), and D) small vessel stroke (SVS), respectively. Ischemic stroke Estimate and CI Total estimate 1.031 [1.004, 1.058] Direct (adjusted for genetically predicted smoking) 1.031 [1.004, 1.060] Direct (adjusted for genetically predicted BMI) 1.023 [0.996, 1.052] Direct (adjusted for genetically predicted CRP) 1.025 [0.997, 1.054] Direct (adjusted for all) 1.023 [0.994, 1.054] Cardioembolic stroke Total estimate 1.060 [1.005, 1.119] Direct (adjusted for genetically predicted smoking) 1.062 [1.006, 1.121] Direct (adjusted for genetically predicted BMI) 1.057 [0.997, 1.120] Direct (adjusted for genetically predicted CRP) 1.041 [0.984, 1.101] Direct (adjusted for all) 1.046 [0.987, 1.109] Large artery stroke Total estimate 1.067 [0.997, 1.141] Direct (adjusted for genetically predicted smoking) 1.069 [0.999, 1.144] Direct (adjusted for genetically predicted BMI) 1.067 [0.992, 1.147] Direct (adjusted for genetically predicted CRP) 1.050 [0.975, 1.131] Direct (adjusted for all) 1.055 [0.976, 1.140] Small vessel stroke Total estimate 1.054 [0.999, 1.112] Direct (adjusted for genetically predicted smoking) 1.056 [1.003, 1.111] Direct (adjusted for genetically predicted BMI) 1.038 [0.985, 1.093] Direct (adjusted for genetically predicted CRP) 1.057 [0.998, 1.118] Direct (adjusted for all) 1.047 [0.990, 1.108] 0.951 1 1.051 1.162 Odds ratio Figure S4: Forest plot contrasting the Mendelian randomization estimates and confidence intervals (CI) from univariable Mendelian randomization (total estimate) and multivariable Mendelian randomization accounting for potential pleiotropic pathways (direct estimate). The total estimate of liability to critical Covid-19 on ischemic stroke outcomes was derived from a univariable (unadjusted) Mendelian randomization model; the direct estimate of Covid- 19 on ischemic stroke outcomes was estimated in a multivariable Mendelian randomization model after adjusting for genetically predicted smoking intensity, body mass index (BMI), or C-reactive protein (CRP), and all three potential pleiotropic pathways jointly. Mendelian randomization estimates represent the odds ratio for ischemic stroke outcomes per unit increase in the log-odds ratio of liability to critical Covid-19. Outcome Pvalue Het.Pvalue Estimate [CI] Ischemic stroke 0.4 0.08 1.020 [0.974, 1.067] Cardioembolic stroke 0.298 0.078 1.048 [0.960, 1.144] Large artery stroke 0.256 0.462 1.051 [0.965, 1.145] Small vessel stroke 0.333 0.308 1.044 [0.957, 1.139] 0.95 1 1.05 1.1 1.15 Odds ratio Figure S5: Forest plot illustrating the Mendelian randomization estimates of liability to critical Covid-19 on stroke outcomes based on inverse-variance weighted Mendelian randomization using 9 genetic variants which were associated with liability to critical Covid-19 at genome-wide significance (p-value < 5×10-8). Mendelian randomization estimates represent the odds ratio for ischemic stroke outcomes per unit increase in the log- odds ratio of critical Covid-19 liability. Additional columns include the p-value (p-value) of the Mendelian randomization estimate to be different from the null, represented by a dashed line at an odds ratio of 1, and the heterogeneity of the Mendelian randomization model measured by the heterogeneity p-value (Het. p-value) as well as the Mendelian randomization estimate and its 95% confidence interval (CI). Hospitalized Covid−19 vs. not hospitalized Covid−19 Estimate and CI Ischemic stroke 1.054 [1.012, 1.099] Cardioembolic stroke 1.044 [0.963, 1.133] Large artery stroke 1.061 [0.957, 1.177] Small vessel stroke 1.219 [1.107, 1.342] Hospitalized Covid−19 vs. population Ischemic stroke 1.026 [0.981, 1.073] Cardioembolic stroke 1.090 [0.991, 1.198] Large artery stroke 1.081 [0.978, 1.194] Small vessel stroke 0.991 [0.912, 1.078] Covid−19 vs. population Ischemic stroke 1.126 [1.005, 1.262] Cardioembolic stroke 1.158 [0.960, 1.396] Large artery stroke 1.464 [1.184, 1.811] Small vessel stroke 1.043 [0.879, 1.237] 0.819 1 1.221 1.492 1.822 Odds ratio Figure S6: Forest plot illustrating the inverse-variance weighted Mendelian randomization estimate and 95% confidence interval (CI) considering Covid-19 phenotypes as exposure for ischemic stroke outcomes. Covid-19 phenotypes were based on the definitions by the Covid-19 host genetics initiative. Mendelian randomization estimates represent the odds ratio for ischemic stroke outcomes per unit increase in the log-odds ratio of liability to the respective Covid-19 definition. Genetic variants which were associated with the Covid-19 definition were selected as instrumental variables at a p-value level equal to 5×10-6 or smaller.