There are numerous issues with drawing policy implications from this study. The first is that the population being looked at is not representative of the general population. 23andMe customers are likely more affluent than the general population, as they have the discretionary income for such services. Unless 23andMe has hospitals (and no bias in participating hospitals!) to trying to recruit all their patients, the COVID-19 population in the study is not representative of the general COVID-19 population.
In their statistical modeling, 23andMe attempt to correct for age, race, ethnicity, BMI, and whether or not any other major comorbidities are present. Of note, they only have a present/absent figure for other serious comorbidities, rather than dealing with them separately. I have no idea how much of an impact this modeling choice may have; I just want to highlight it as unusual. Also, unlike any other model I’ve seen, they include both age and age-squared, presumably to deal with the sharp increase in risk as age increases past 70, which I think is a clever idea.
The lower rate of infection for people with type O blood was the difference of a 1.3% infection rate and a 1.5% infection rate. In other words, it would take 500 people before a difference of blood type would mean one fewer infection….assuming that blood type is causal, and the study has the true effect size. Is it plausible that the issues I’ve summarized above could account for a difference of this size?
Other papers that have similar findings include “Relationship between the ABO Blood Group and the COVID-19 Susceptibility” by Zhao et al. and “The ABO blood group locus and a chromosome 3 gene cluster associate with SARSCoV-2 respiratory failure in an Italian-Spanish genome-wide association analysis” by Ellinghaus et al. Zhao et al. found an increased risk for blood type A, and a decreased risk for blood type O (68% of the infection rate). Patients were recruited from hospitals, and compare to the general population in the area of the hospital. The authors highlighted the normal issues I criticize these studies for; most notably they didn’t correct for comorbidities. The authors also mentioned limitations in the data, and that further work is needed before drawing any conclusions, and, for instance, devoting serious resources to studying what it is about the proteins that vary with blood type that affects COVID-19 vulnerability. One other flaw that I didn’t pick up on, but found in an online discussion of the paper, is that blood type might not be the right thing to measure. If the gene that mattered was near the gene that codes for blood type, blood type could be highly associated with COVID-19 susceptibility with no causal relationship.
Ellinghaus et al. looked at 1980 patients across several hospitals. COVID-19 patients who were at least sick enough to require Oxygen were compared to retrospective matched cases from blood donors. Several million genetic locations were tested, and several were found to have significant correlations with suffering from COVID-19. Notably, one of the places they found an association is on the gene that codes for blood type, but:
- Does not itself affect ABO blood type
- Is not on of the 11 locations in the gene measured by 23andMe.
Like Zhao et al., Ellinghaus et al. found a similar reduction in infection rate to the Zhao et al. study for people with blood type O (65%), and an elevated risk for people of type A. Looking at retrospective case control studies has historically been problematic - back in the 1930’s this methodology suggested it might be a good idea to infect people with Lung Cancer with Tuberculosis, since Tuberculor patients were less likely to have Lung Cancer! A thorough explanation of this affect is beyond the scope of this post.
So what? I’d like to applaud the authors of Zhao et al. and Ellinghaus et al. for performing the best studies they could with the data they had access to, for being clear and open in their write-ups, and in doing what they can to contribute to the general knowledge-base. Perfect should not be the enemy of good. But what does an association between blood type and COVID-19 risk mean? Should people with type O be extra-cautious or protected because they are more vulnerable? We’re looking at hundreds of people before we get an extra infection, so probably not...especially since, with the issues highlighted above, I’m not convinced this effect is real. Does it give us a hint that we should look at the blood proteins associated with blood type to better understand the pathology of COVID-19? Based on Ellinghaus et al., likely not; it is something else that differs in the proteins that might matter. But this is a nice case study in correlation being different from causation, and that you might not be measuring what you think you are measuring.
No comments:
Post a Comment