Coronavirus infection rate + outcomes -versus bloodtypes

[jim mcnamara]

TL;DR

Blood types A, and O influence the rate of infection and death rates

https://www.scmp.com/news/china/soc...ay-be-more-vulnerable-coronavirus-china-study
This is a news report on a Chinese study, not a journal article

Type O population of confirmed patients were disproportionately less like to die and less likely to become infected in the first place - compared to the type A. No reason noted, as far as I can tell. There were cases of all blood types and deaths in all bloodtypes in case you are confused.

Interesting. If anyone knows of a technical article, please post a link.

 

[.Scott]

I believe this is the study:
Medrxiv - Relationship between the ABO Blood Group and the COVID-19 Susceptibility

 

[.Scott]

I can't say that I am convinced that the blood group itself is a causal factor.
Since blood type is a hereditary - perhaps they simply had slightly more families in the A group that contracted COVID-19.

 

[BillTre]

Blood types are based on surface molecules on blood cells.
It is conceivable that cell surface molecules could affect how easily a virus can enter a cell in some way.

 

[Vanadium 50]

I didn't find the article convincing.

1. The hypothesis is very complicated when expressed in terms of antigens. Is A good or bad? By itself, it's bad. But with B it's neutral. What about B? By itself it's bad, but with A it's good.
2. The statistics reject the hypothesis that Wuhan and Shenzhen have the same ABO populations at an even higher confidence level than the Covid statistics. Given that diseases cluster in the population, it's far from clear that they are seeing anything but that.

 

[Vanadium 50]

I read it again, and if I were the editor, I would reject it. The statistics is muddled. Apart from the uncontrolled confounding factors described above, they perform one test and then draw conclusions as if they performed another. Then they assume causality where they see correlation.

They do an independence test on blood type and disease incidence and progress. Fair enough. But then they draw conclusions like A is bad, O is good. First, this is an a posteriori test. Second, there is no mention of the need for a trials factor when in fact they are performing multiple tests: (A is good, O is bad; B is bad, AB is bad, A is good...)

The authors don't understand statistics. While it's possible that someone else could go through the data and do it right, that's not what we have in front of us.

 

[.Scott]

The point that I was trying to make before is that the cases are not independent measurements of each other. It's as though you flipped 3000 coins and 1000 came up heads and 2000 came up tails. It may at first seem as though the chances of that happening were extraordinarily remote. But if I then told you that the coins were welded together into three clumps and it was simply a case where one clump (and all the coins it in) landed heads and the other two clumps landed tails, it would be far less impressive.

That is what I see as the most important failing of the statistics they are using.

Many small groups within their patient population are connected by both heredity and proximity. Hence the "welding".

 

[Vanadium 50]

I think we're saying the same thing.

A better-designed study would take the random population, divide it into groups of 1775, and see if one can pick out the one group that's sick. That still won't capture all the correlations, but I suspect that it won't matter: the groups will differ among each other by about the same amount.