COVID-19 first emerged in December 2019 and has become a worldwide pandemic currently (1). At the time of April 18, 2021, there have been 140,821,384 confirmed cases and 3,013,042 deaths globally (2). In this condition, it is essential to identify high-risk groups that need special attention (3). Particularly, with the available but limited supply of COVID-19 vaccines, a crucial challenge is prioritizing groups to receive vaccines (4–6). There have been discussions about vaccine prioritization for racial minorities (4) and diabetes groups (7), since prior studies report these groups are more vulnerable to COVID-19 disease. In contrast, more evidence for other potential high-risk groups such as the population with low educational attainment (EA) is needed.

Educational attainment is a well-established social determinant of health (8) and correlates with many diseases (9–11). Prior observational studies indicated that the same might have happened during the current COVID-19 pandemic and a population with lower EA was found at a higher risk of susceptibility, hospitalization, and mortality of COVID-19 (12–14). However, conventional observational studies lacking randomization designs are generally prone to confounding factors (15). Randomized controlled trials, on the other hand, cannot be conducted.

Mendelian randomization (MR) is a method that uses genetic variants correlated with an exposure (such as EA) to evaluate whether it has a causal effect on the disease outcome (such as COVID-19 susceptibility) (16), which is less likely to be influenced by unmeasured confounding than observational studies (17). MR is especially useful for exploring causal pathways when the risk factors are difficult to randomize (18). There have been several MR studies exploring the risk factors for COVID-19 (18–20).

An extension of the single-variable MR (SVMR) is the multivariable MR (MVMR), which can incorporate genetic variants associated with several exposures into the same model (21). Since EA was identified to correlate with intelligence, income, body mass index (BMI), vigorous physical activity, sedentary behavior, smoking, and alcohol consumption in a prior study (22), we would also perform the paired MVMR to investigate the direct effects of EA when controlling these exposures separately and explore whether the effects of EA on COVID-19 were independent of them.

Therefore, in this study, we aimed to evaluate the total and direct effects of EA on the susceptibility, hospitalization, and severity of COVID-19 using the SVMR and the MVMR separately, trying to provide evidence for public health resources allocation and targeting prevention planning.

In this study, using both the SVMR and MVMR, we found genetically higher EA was related to a lower risk of susceptibility, hospitalization, and severity of COVID-19.

Using data from UK Biobank, a prior study reported that lower EA was correlated with a higher COVID-19 infection risk [relative risk (RR) 2.00, 95% CI 1.66–2.42] (13), indicating EA might be a possible predictor for COVID-19 infection. However, the sample size of this study was relatively small, with only 948 cases enrolled, making their results less reliable. While in this study, using the largest GWAS of EA and latest summary statistics for COVID-19, we have sufficient statistical power and found that EA is correlated with COVID-19 susceptibility. The underlying mechanism between EA and COVID-19 susceptibility could be partly explained by frontline jobs usually without a requirement of an advanced degree (37), thus individuals with lower EA being more likely to be infected. Second, people with lower EA tend to have a lower socioeconomic status. As a result, they are more likely to live with multiple close generations and, therefore, at greater risk of contracting COVID-19 (4).

Hospitalization is another important COVID-19 outcome and the population with lower EA was reported at a higher rate of hospitalization in a prior observational study (38). However, a major disadvantage of this study is that they only used data from 5 New York City boroughs and their sample might not be representative enough. While in this study, this shortcoming was overcome and we found that EA was a predictor for COVID-19 hospitalization in both the SVMR and MVMR. In addition to COVID-19, EA has been identified as a risk factor for hospitalization among many infectious diseases such as pneumonia and bacteremia (39, 40). Data from a Danish population-based case–control study indicated that the population with a short duration of education had a substantially higher risk of bacteremia than those with long duration (39). The underlying mechanism between lower EA and higher risk of hospitalization might be mediated by overcrowding, poor housing conditions, and hygienic practices.

In addition, this study also found that EA was a predictor for COVID-19 severity, which was in accordance with one prior MR study from Japan (41). However, comparing with the study from Japan, we applied the latest 6th round of COVID-19 summary statistics in this study with a larger sample. Several hypotheses are elucidating the mechanism between EA and COVID-19 severity. First, individuals with lower EA and socioeconomic positions are more likely to be affected by job stress such as unemployment, which might lead to a higher risk of immune system disruption and comorbidities (42). So far, both the weak immunity and the presence of comorbidities are recognized risk factors of COVID-19 severity (43, 44). Besides, population with lower EA is more likely to have unhealthy behaviors such as smoking and an unbalanced diet (45), while these unhealthy behaviors have been recognized as risk factors of COVID-19 mortality or severity (46, 47). Additionally, lower EA generally correlates with lower income and socioeconomic status, which means that they have limited access to healthcare (48) and once they are infected, they might not get treatment in time and turn to be critically ill or even dead.

As for the MVMR, we found that EA still maintained its effects on COVID-19 outcomes under almost all the conditions. This showed that the effects of EA were generally independent of these exposures. One limitation of our MVMR was that part of the conditional F-statistic was relatively low. However, since there was partial overlap, the conditional F-statistic might not be accurate enough and should be interpreted cautiously (21).

This study has several strengths. First, using the largest GWAS of EA with 1.1 million participants and the latest summary statistics for COVID-19 from the 6th round of the COVID-19 HGI, we have enough statistical power. Second, although MR study is still prone to potential sources of bias, it is generally less vulnerable to confounding factors than observational studies. Third, in addition to the SVMR, we also conducted the MVMR analyses and results showed that the effects of EA on COVID-19 outcomes were generally independent of some other exposures. However, this study could not be devoid of limitations. First, since individuals with no or mild symptoms are less likely to test for COVID-19 and there might be potential selection bias in this study. Second, COVID-19 outcome ascertainment methods were not exactly the same among all the cohorts in the COVID-19 HGI, leading to the potential heterogeneity. However, we were unable to estimate how these differences might influence our findings since we only had access to the overall summary statistics. Third, in order to increase the sample size and statistical power, we used the overall COVID-19 HGI GWAS in our primary analyses, which was conducted on mixed ancestry and might lead to some potential population stratification biases. However, we have conducted sensitivity analyses using the summary statistics for European only. Fourth, one prior study reported that using GWAS of education-related traits might be biased from population stratification (49). Another limitation of our MR is the partial overlap of participants (e.g., UKB) in the exposure and outcome datasets, which might lead to possible bias (50). However, bias caused by sample overlap would likely to be minimal for both the continuous and binary outcomes (50) and it has also been shown that 2-sample MR methods may be safely used in single sample provided the data are derived from large biobanks (51), as is the case in our analysis. Last, EA is a complex phenotype and might correlate with some cofounders. Although we observed no directional pleiotropy and the MVMR analyses also showed that the effects of EA on COVID-19 outcomes were generally independent of some other exposures, it is still possible that confounding and pleiotropy may be present.

Our MR analyses indicated that EA is a predictor for susceptibility, hospitalization, and severity of COVID-19 disease. Population with lower EA should be provided with a higher prioritization of public health resources to decrease the morbidity and mortality of COVID-19.

Publicly available datasets were analyzed in this study and can be accessed via the references we used.

Ethical review and approval was not required for the study on human participants in accordance with the local legislation and institutional requirements. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.

ZJ, MW, XJ, and XW: designing the study. ZJ and MW: carrying out the study, analyzing the data, and writing the article. XJ and XW: revising the article. All authors read and approved the final manuscript.

This study was funded by Post-Doctor Research Project, West China Hospital, Sichuan University (2020HXBH016).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.