Read on Twitter
As we struggle with the difficult decisions about reopening K-12 schools, one thing I& #39;ve been unsure about is the degree of transmission by children.
A new study from Korea provides high-quality data about this question based on a large sample size.
https://wwwnc.cdc.gov/eid/article/26/10/20-1315_article">https://wwwnc.cdc.gov/eid/artic...
A new study from Korea provides high-quality data about this question based on a large sample size.
https://wwwnc.cdc.gov/eid/article/26/10/20-1315_article">https://wwwnc.cdc.gov/eid/artic...
It is not good news, by and large. Older children (10-19) appear to spread disease at rates comparable to adults, though younger children may be less likely to transmit.
The @Nytimes provides context. https://www.nytimes.com/2020/07/18/health/coronavirus-children-schools.html">https://www.nytimes.com/2020/07/1...
The @Nytimes provides context. https://www.nytimes.com/2020/07/18/health/coronavirus-children-schools.html">https://www.nytimes.com/2020/07/1...
This is not wholly unexpected—I was never persuaded by the small studies suggesting reduced transmission in this age group–but it& #39;s not going to help one bit as we try to keep the pandemic under control while returning to some sense of normalcy.
Covid-denialists are out in force in response to this study, highlighting one clause of the methods (below) and claiming that it invalidates the conclusions.
This is absolutely false, and if one reads the rest of the paragraph the paper explains right there.
This is absolutely false, and if one reads the rest of the paragraph the paper explains right there.
Ideally one would know the direction of transmission. If so, one could simply count up transmission events, calculate the average number of transmissions by members of each age group, and be done with it.
As they acknowledge, they can& #39;t do that. This where *statistics* comes in.
As they acknowledge, they can& #39;t do that. This where *statistics* comes in.
What they show is that there is a statistically significant difference between the fraction of household contacts who exhibit disease when the index case is 10-19 years old, and the fraction who exhibit disease when the index case is drawn from the population at large.
Now I do have two concerns. 1) I would like to understand more about if / how they corrected for multiple comparisons. 2) I& #39;d like see a careful treatment of whether there is any other causal path that could explain the pattern. I can& #39;t think of one, but that doesn& #39;t mean much.
Even with these concerns, I find the study compelling.
And the key point here is that the objection "the study is because they can& #39;t infer the direction of causality" is a silly claim that ignores the possibility of doing statistical inference.
And the key point here is that the objection "the study is because they can& #39;t infer the direction of causality" is a silly claim that ignores the possibility of doing statistical inference.
Here& #39;s an analogy. Suppose we want to know if smoking causes cancer. It would be nice if we could directly *watch* particles from cigarette smoke cause lung cancer. That would be a sort of gold standard, akin to being able to determine directly that 10-19-year-olds infect others.
But we can& #39;t watch that happen.
So what we do is we look at a population that is exposed to the risk (smokers) and compare it to a population that is not exposed to the risk (non-smokers).
And we find higher rates of lung cancer in smokers.
So what we do is we look at a population that is exposed to the risk (smokers) and compare it to a population that is not exposed to the risk (non-smokers).
And we find higher rates of lung cancer in smokers.
This provides us with statistical evidence that smoking is causing lung cancer, even though with any individual lung cancer patient we cannot say definitively that he or she got cancer *from* smoking.
It& #39;s the same with having a 10-19 year old index case in the house and being infected.
We can& #39;t say precisely which subsequent infections came from the index case, but we can conclude that overall 10-19 year olds pose a greater danger to their contacts.
We can& #39;t say precisely which subsequent infections came from the index case, but we can conclude that overall 10-19 year olds pose a greater danger to their contacts.
Update: my original interpretation of this EID paper was that the 10-19 age class transmits at comparable rates to adults; I didn& #39;t make much of the significantly higher rate of infections among contacts of index cases in this group.
But....
But....
I also noted that the observation was a statistical association, not a causal relationship, between having an index case in the 10-19 age class and having a higher rate of infected contacts. One obvious causal path is direct infection. I couldn& #39;t think of obvious alternatives.
In a separate thread, @apsmunro has suggested an alternative path: the happenstance event that the index cases age 10-19 share a common exposure with the secondary cases in that cluster. https://twitter.com/apsmunro/status/1284794506265722881">https://twitter.com/apsmunro/...
If so, this could explain the association found in the study *without* requiring higher transmission from the 10-19 age class (or, in principle, even adult levels of transmission). I sketched a diagram of the association and two alternative causal mechanisms.
If this is indeed the case, I find it very odd that the authors did not report this given its direct relevance to this discussion in the paper. I would very much appreciate from them some clarity on this point.
In the meantime we& #39;re left trying to adjudicate between the authors who have the unpublished raw data and another researcher who claims to have seen it.
This is not something anyone can hope to without seeing the data themselves or least knowing a lot more about the situation.
This is not something anyone can hope to without seeing the data themselves or least knowing a lot more about the situation.
