Read on Twitter
Quick thread on evidence being presented that schools are main drivers of transmission (and here: primary aged children just as much!).
There really are papers claiming to show some of these things, but I think it is easy to emphasize the wrong kind of data, as I will explain.1/ https://twitter.com/dgurdasani1/status/1362390053029306368
There really are papers claiming to show some of these things, but I think it is easy to emphasize the wrong kind of data, as I will explain.1/ https://twitter.com/dgurdasani1/status/1362390053029306368
There are three lines of evidence in the thread:
1) ONS survey data apparently showing a coincidence involving when schools were closed and what happened with cases in children,
2) Correlation analyses on closures in the Spring
3) A modeling paper on school effects on R0. 2/16
1) ONS survey data apparently showing a coincidence involving when schools were closed and what happened with cases in children,
2) Correlation analyses on closures in the Spring
3) A modeling paper on school effects on R0. 2/16
Let's start with the first. There is an understandable tendency in the UK to understand the impact on school policy primarily using UK data.
The problem is that with relatively uniform school policies, this involves interpretations of coincidences in a single time-series. 3/16
The problem is that with relatively uniform school policies, this involves interpretations of coincidences in a single time-series. 3/16
Recall, for example, the fall case increases across European countries. Some were quick to blame their schools, because the increases occurred in the fall after schools opened. But in other cases (e.g., Spain) it was clear the increase began before schools re-opened.
[...]
4/16
[...]
4/16
Focusing on a single country's data, it would be easy to have the impression that schools were largely responsible for fall COVID increases. Taking a global view (or just looking across European countries or US states) undercut such a view.
5/16
5/16
In particular, the CDC director's view should come as no surprise. Her department oversees a country with a great diversity of school policies which are largely uncorrelated with local epidemic trajectories. This makes it hard to have dogmatic views! 6/16 https://twitter.com/dgurdasani1/status/1362390054451183619
Indeed the @CDCDirector's viewpoint is based in evidence. To dismiss her view, one should engage with the evidence, rather than simply present other evidence that goes the other way.
7/16 https://jamanetwork.com/journals/jama/fullarticle/2775875
7/16 https://jamanetwork.com/journals/jama/fullarticle/2775875
On to evidence type 2. This concerns correlation studies that try to associate the timing of the initial school closures with changes in the epidemic patterns.
Putting aside the many problems in these types of correlation analyses, interpreting these studies as capturing..
Putting aside the many problems in these types of correlation analyses, interpreting these studies as capturing..
transmission risk in schools involves an assumption that when long-term school closures were implemented for the first time in modern history, the only impact on population-level behavior was on interaction rates among children.
But closings schools was associated... 9/16
But closings schools was associated... 9/16
(both causally and coincidentally) with massive shifts in behavior among adults too. To the extent that spring school closures are associated with epidemic effects primarily through incidental effects on adults, we should realize that reopening schools is NOT... 10/16
necessarily the reverse of closing them. And indeed, studies of the impact of reopening schools have not consistently found a corresponding increase in transmission levels. 11/16
https://biocomsc.upc.edu/en/shared/20201002_report_136.pdf
https://biocomsc.upc.edu/en/shared/20201002_report_136.pdf
For this reason it is sensible to look at school impact around the globe when trying to understand the epidemiological role of schools. Looking only at spring closures, or only at experience over time of one country (the UK) gives a limited picture. 12/16
Finally the thread discusses a modeling paper. This paper does not use any actual epidemiological data about COVID cases in schools, etc. Instead,
13/16 https://twitter.com/JDMunday/status/1361358151807889409
13/16 https://twitter.com/JDMunday/status/1361358151807889409
it uses self-reported contact patterns in and out of school, assumes these contacts are all equally epidemiologically relevant despite any differences in duration, distance, etc, and runs an age-structured abstract epidemic model.
This is a nice modeling exercise, 14/16
This is a nice modeling exercise, 14/16
and I also ran these types of models early in the pandemic to explore qualitative phenomena. However, it is a very indirect form of evidence at a time when states and countries around the world have actual on-the-ground experience with schools. 15/16
In summary, the reason people like the @CDCDirector believe schools can operate without children driving transmission is because precisely this seems to be happening around the world. A focus on analyses of last spring or time-series data from one country gives a limited view./16
Let me also link to @apsmunro's thread, which discusses a broad set of data from around the world on schools (and also discusses the comparison of the number outbreaks in primary vs secondary schools in the UK, which I forgot to address). https://twitter.com/apsmunro/status/1362446759545495554
