SNP identification and data sources
Genetic association datasets for COVID-19
These summarized data were derived from genome-wide association studies (GWAS) analysis of the latest version about the COVID-19 host genetics program from the UK Biobank individuals, which was aimed to determine the genetic determinants of COVID-19 susceptibility and the level of severity (https://www.covid19hg.org/results/) [21]. Including 3,523 patients and 36,634 control participants, 105 studies about COVID-19 have joined this program. These studies are carried out all over the world, including countries in Europe, America, Asia, and Africa.
Genetic association datasets for heart failure
We obtained the association between the specific SNPs and HF from the current largest GWAS meta-analysis of HF in people of European descent [22]. The GWAS meta-analysis, which was conducted by the Heart Failure Molecular Epidemiology for Therapeutic Targets Consortium included 26 studies (17 population cohorts studies, 38,780 HF cases, 893,657 controls and 9 case-control studies, 8,529 cases, 36,357 controls), 47,309 patients with heart failure, and 930,014 control group [22]. This GWAS meta-analysis was adjusted according to gender, age, and main components. In all cohort studies, heart failure was assessed using at least one of the following methods: discharge registration, cause of death registration, or physician decision/diagnosis. Due to insufficient power, GWAS was not stratified according to etiological subtypes. All participants obtained informed consent, and the related ethics committees all approved these studies.
MR analysis
In the main analysis of MR analysis, we used the standard inverse-variance weighted (IVW) method to estimate the overall causal relationship between COVID-19 susceptibility to HF [23]. By using this method, the causal effect of exposure on the outcome is estimated as the ratio of the SNP outcomes associated with the exposures (Wald estimate). According to Mendel's law of inheritance, MR assumes that SNPs are randomly distributed in the general population (separation, independent classification), simulating the process of randomization, SNP always appears before the disease, so reverse causality can be effectively eliminated. To ensure a valid inverse weighted variance method within the MR analysis process, three significant assumptions need to be proved: (1) the SNPs have a robustly relation with COVID-19 (the exposure), (2) the SNPs are in complete independence from any potential confounding factors that influence both COVID-19 and HF, and (3) the SNPs exert influence on HF (the outcome) only by COVID-19 (the exposure) and not via any alternative causal pathways (Figure 1) [24, 25].
The standardized utilization of MR is a single-sample MR, conducted in a group of people, including intact data on the SNPs, exposure, and results of all participants [19]. According to the rare statistics in single-sample MR, the two-sample MR was developed to enable analysis in two independent samples, one for focusing on exposures and the other for outcomes [24]. In our study, we performed the IVW, the Mendelian randomization-Egger (MR-Egger) regression, the simple mode (SM), weighted median and weighted mode. These five two-sample MR methods were performed by the "Two-sample MR" package in R (version 4.0.3) [23, 26]. The related analysis was all one-sided, and the evidence of causal relationship was announced when a pre-specified p-value is lower than 0.05.
According to the superiorities of each MR, these five means can make up a deficiency of each other and provide a more credible causal relationship for our research. In a two-sample MR analysis, we applied the IVW method to analyze the associations between genetically predicted COVID-19 infection and HF. The MR-Egger method was used for the estimation of accidental effects and evaluation of directed pleiotropy under weaker assumptions. When 50% or more of the genetic variations are valid instrumental variables, the median-based method can give a reliable effect estimate, which may be more suitable than the MR-Egger method. The weighting method provides a more accurate causal estimation based on more weight analysis. The simple mode help to avoid the effect of pleiotropic effects in causal reasoning and omit part of the genetic variation from the analysis. Compared with the traditional MR analysis, the robust method estimates the causal effects consistently under weaker assumptions.
Traditional IVW methods are the appropriate method to use aggregated data from GWAS. We used it to initially estimate the impact of COVID-19 on HF [23]. Firstly, we performed the IVW average of SNP-specific associations with fixed effects in HF. If some tools in the causal reasoning hypothesis based on MR analysis are invalid, the analysis gives a biased estimate [27]. Secondly, we resolved the first hypothesis (the true relation between SNPs and COVID-19) by choosing SNPs that robustly predicts COVID-19, and the gene variants that act as COVID-19 agents are likely to satisfy the second hypothesis (there are no confounding factors). We employed the MR-Egger regression to investigate directional pleiotropy to evaluate the potential violation of the third assumption. Subsequently, we created a scatter plot to visually detect the potential pleiotropy by showing the association between each SNP and HF risk against associated with COVID-19. Sensitivity analysis were conducted by using the MR-Egger regression, and weighted median, to explain the potential violation of the effective tool variable hypothesis.