COVID-19 and blood groups – there is an elephant in the room, but who cares? Do we need additional rules for preprints?

In general, associations between blood groups and infectious diseases are well known. They may be either related to the probability of becoming infected or to a more or less severe course of the disease. Recently, associations of ABO blood groups and Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-Cov-2) have attracted a lot of scientific interest, due to a number of studies where such associations have been investigated. One such study has generated enormous public interest.

As to SARS-COV-2 and blood types, initial interest in the field was stimulated by a preprint by Zhao et al., which was initially published on March 16, 2020¹. Based on their results, individuals with blood group A had a significantly higher risk for acquiring COVID-19 compared with non-A blood groups, whereas individuals with blood group O were found to be at a significantly lower risk for the infection. Less than one month later, another preprint appeared also addressing a possible link between COVID-19 and blood groups, but did not find any significant associations between blood groups and intubation or death². Again one month later, the first report on a genome-wide association study (GWAS), carried out by a European consortium, appeared as a preprint³, which was followed by a press release. This was then published as a New England Journal of Medicine (NEJM) paper⁴, again accompanied by a corresponding press release, and apparently all based on the same group of COVID-19 patients receiving oxygen and a control group. The critical considerations outlined herein mainly relate to the preprint by the European consortium, the paper published in NEJM, and the two press releases. The message of all four was that people with blood group A have a much higher risk for respiratory failure during the course of COVID-19 than those with B or 0. This caused a huge public reaction (Table 1), resulting in a lot of anxiousness due to headlines being published in the media, such as “People with blood type A more likely to suffer severe coronavirus symptoms, research finds" (Telegraph, UK, 4^th June 2020^a) and TV reports (e.g. News9 India of June 5, 2020^b). Both the high public interest, as well as possible clinical applications of the findings, warrant a critical analysis of the methods and data applied.

The preprint and its weakness

The European consortium (with K. Ellinghaus and F. Degenhardt from Kiel, Germany, as shared first authors and A. Franke, Kiel, and T. H. Karlsen, Oslo, Norway, as corresponding authors) published the results of a GWAS with COVID-19 patients from Spain and Italy who all received oxygen during the course of the disease³. One of two significant association signals was located at chromosomal band 9q34, corresponding to the ABO blood group locus. It was claimed that the ABO blood groups thus are associated with COVID-19 induced respiratory failure "with higher risk for A-positive individuals and a protective effect for blood group O". Shortly after the preprint’s appearance the headlines and news went viral (Table 1).

In a few media articles, the message was overreached, for example by Medscape, Germany, who stated that an association between COVID-19 mortality and blood groups has been found (12^th June 2020^c). In particular in German-speaking countries, interest in the results of the study was also stimulated by a tweet from Karl Lauterbach^d, professor of Health Economics and Clinical Epidemiology at the University of Cologne and member of the Deutscher Bundestag, which was frequently cited in public media (Figure 1).

Figure 1. Tweet by German health professional and politician Karl Lauterbach addressing the preprint.

[A very remarkable result shown in a robust study. Blood group A about 50% higher risk for severe course of COVID-19. Risk twice as high as with blood group 0. Blood group B in between. Because immune response depends on the blood group it makes sense]. (https://twitter.com/Karl_Lauterbach/status/1268522737573728257, translated from German). Permission to reproduce this tweet was obtained from Karl Lauterbach’s office.

Despite this wide-spread interest, the value of the study suffers from its control group, which comprises a total of 2,381 individuals. According to the preprint, the control group, among others, included “998 randomly selected blood donors at Fondazione IRCCS Cá Granda Ospedale Maggiore Policlinico, Milan with no evidence of Covid-19 who were genotyped for the purpose of the present study.“³. Two further control panels with genotype data derived from previous studies using the same genotyping array were included as well (from Italy n=396 and from Spain n=987). Due to the selection of the control group, the findings do not allow for the conclusion that ABO blood types influence the severity of symptoms, since no comparison has been made between people receiving oxygen versus those mildly affected (e.g. non-hospitalized) as a control group. Accordingly, similar results have to be expected, if blood groups affect one’s chance getting infected at all. This problem was pinpointed early on by some in the public media, for example by The Scientist Magazine (“Ideally, a GWAS analysis would analyze the genomes of people with COVID-19 and compare those who didn’t get very sick to those who experienced severe symptoms, instead of using population-based controls whose exposure to the virus is unknown, says Priya Duggal, a genetic epidemiologist at the Johns Hopkins Bloomberg School of Public Health who did not participate in the study.“; 8^th June 2020^e) and the German Sueddeutsche Zeitung (15^th June 2020^f).

Already at this time, at least one of the authors of this study apparently was well aware of this major shortcoming: “The study was not perfect. Ideally there would be another full group of patients for comparison who were infected but who did not develop the severe disease. This was one reason, said Professor Franke, that he could not be confident of the exact risk increase linked to blood type: “It may be 20 per cent, it may be 30 per cent, it may be 50 per cent.” (as quoted in The Australian, 12^th June 2020^g). Nevertheless, a press release^h by the lead author’s own institution (Universitätsklinikum Schleswig-Holstein) on 10^th June 2020 revealed that according to their results “people with blood group A have a roughly 50% elevated risk for a severe course of COVID-19 than people with other blood groups. Vice versa, people with blood group O were protected against severe disease by nearly 50%” (translated from German).

The study received accolade, but doubt remains

On 17^th June 2020, the paper was published in the New England Journal of Medicine⁴, and the study received recognition. Unfortunately, the problem with the control group became even more complicated because the description was changed from the preprint: “We recruited 998 randomly selected blood donors at Fondazione IRCCS Cá Granda Ospedale Maggiore Policlinico, Milan, who underwent genotyping for the purpose of the present study. A total of 40 of these participants had evidence of the development of anti–SARS-CoV-2 antibodies, all of whom had mild or no Covid-19 symptoms.“⁴

In contrast, according to the preprint³, all 998 blood donors did not show evidence of COVID-19. Also, it is unclear if all had been tested for antibodies and by which test. What was the control group then? A pre-COVID-19 population? Apparently not, but maybe in part. Patients infected, but with mild or absent symptoms of COVID-19 constituting part of the controls, make interpretation of the data even more difficult if not impossible. Despite this, a second press releaseⁱ by the lead author’s institution was published on 18^th June 2020, which unequivocally concludes: “The study had shown that people with blood group A have an approximately 50 percent higher risk of severe Covid-19 progression than people with other blood groups. In contrast, people with type O blood groups were almost 50 percent better protected against serious covid-19 disease.”. The appearance of the now peer-reviewed paper caused a second wave of headlines (see Table 1). Nevertheless, most notably a large part of the media coverage was gained in the two-week period when only the preprint was available (Figure 2).

Figure 2. Timeline showing dates of preprint, press releases, and publication of the paper in New England Journal of Medicine.

NEJM: New England Journal of Medicine. UKSH: Universitätsklinikum Schleswig-Holstein, Germany.

The enormous attention is also reflected by the preprint and article metrics, e.g. compared to a paper by Latz et al.⁵ also addressing a possible association between blood groups and the severity of COVID-19, but which came to an opposite conclusion (Figure 3). Simultaneously, the publication caused an enormous level of public interest and anxiety as witnessed by the coverage gained in public media (Table 1).

Figure 3. Article metrics for the preprint by Ellinghaus et al.³ (number of abstract and preprint views), the Annals of Hematology paper by Latz et al.⁵ (number of accesses), and the New England Journal of Medicine paper by Ellinghaus et al.⁴ (number of views).

Metrics for 5 and 14 days after publication are given.

Are additional rules for preprints needed?

How could it happen that, to say the least, a misleading message spread so rapidly and at a preprint stage? What makes the story so appealing for public recognition? It’s not the association with an anonymous gene. Instead, people know about blood groups and many even know about their own. Besides genetic association there is obviously a direct interest in personal risk. Can we thus blame the hunger of the public for intriguing messages on COVID-19 that might affect people personally? At least this is not the only factor explaining the high public interest in this case. It seems reasonable to speculate that a scientifically more correct interpretation of the limits of the study at the preprint stage and in the press releases might have led to a drastically decreased level of interest by public media, but the message of the study has been made more fascinating by the authors themselves – with their press releases and the preprint. Should thus the preprint system of publishing a piece of work without peer review be blamed?

One of the main arguments in favour of sharing work in its preliminary preprint form is that articles are improved by feedback from a wider group of readers instead of only formal peer review by few experts, and that science works faster if results are made available sooner after being completed⁶. Basically, both principles have worked here. As for the speed, the preprint was published some two weeks earlier than the paper in NEJM. However, it could be said that there was no such need for speed, and more important was the opportunity to gather feedback from a larger community, which has worked nearly perfectly. Currently, there are 39 comments for the preprint by Ellinghaus and coworkers³. Of these, some comment on the issues faced with the controls^j, and this problem was also pinpointed very early on by experts in the public media and, curiously enough, even at least by one of the authors (see above). Moreover, there were also contradictory findings, which were neither appropriately mentioned in the preprint nor in the NEJM paper, e.g. the preprint by Zietz and Tatonetti² that appeared earlier on the same server (medRxiv).

One of the main advantages of a preprint, i.e. the chance to have it critically considered by a larger group of peers, has evaporated here due to the short time between the preprint and appearance of the paper, giving the authors no chance to consider comments and criticism. Of note, even now the preprint has not been corrected, though the control group is apparently not correctly described given that the description in the NEJM is correct.

There is a low level of control until preprints go public, which clearly increases the author’s responsibilities. Aimed at the benefit for the scientific community, preprints have to balance a possible need for speed with “suitable safeguards to protect the public”⁶ as well. The case presented here illustrates the adverse effects associated with a loss of such balance and clearly advocates additional rules for preprints. To put it simply, authors should not leave the room while the elephant is still in it. Whenever submitting a manuscript which corresponds to a preprint, a time of some weeks should be given to the extended group of peers to comment on and vice versa, authors should communicate possible comments along with the manuscript and they should declare that they have read and considered all these comments carefully. During that time, there should not be any press releases from the authors. In a comment on their study⁵ investigating possible associations between the severity of COVID-19 and ABO blood groups, Latz et al. made the rigorous statement that they are confident their principle finding “will help debunk the kinds of clinically unfounded rumours and misinformation that can readily gain traction in the midst of a pandemic, and in some cases become part of accepted medical practice”^k. Certainly, we as scientists have a direct responsibility to avoid this type of misinformation.

And what on the actual news on COVID-19 severity and blood groups? “Risk does not depend (much) on blood type” stated The New York Times on 15^th July 2020^l. The Latz et al.⁵ study, based on 1,289 COV+ individuals, concluded that blood type was not associated with intubation or mortality in COVID-19 patients. Similarly, the recently published second version of the preprint by Zietz and Tatonetti⁷ even revealed a significantly lesser risk for intubation for individuals with blood group A.

Data availability

No data is associated with this article.