Read on Twitter
A thread about COVID in schools. This recent nature piece summarizes the situation from a research perspective well: we still do not know enough about the biology of the virus in children. Data is scarce & ridden with bias due to asymptomatic cases.
https://www.nature.com/articles/d41586-021-00139-3?utm_source=Nature+Briefing&utm_campaign=ff2b4a1cd3-briefing-dy-20210121&utm_medium=email&utm_term=0_c9dfd39373-ff2b4a1cd3-43675261
1/25
https://www.nature.com/articles/d41586-021-00139-3?utm_source=Nature+Briefing&utm_campaign=ff2b4a1cd3-briefing-dy-20210121&utm_medium=email&utm_term=0_c9dfd39373-ff2b4a1cd3-43675261
1/25
In @CSHVienna's recent policy brief we investigate the impact of different preventive measures in schools. We look at
- wearing masks
- ventilating the rooms
- testing students & teachers regularly
- halving class sizes
2/25
- wearing masks
- ventilating the rooms
- testing students & teachers regularly
- halving class sizes
2/25
In the model we use to simulate dynamics of infection spread in schools and come to the conclusions presented in the policy brief, the age of children is also a factor, but not a dominating one. Let me go a bit into the details here:
3/25
3/25
Differences in how an infection plays out enter the model in a variety of ways:
(1) the probability to have a symptomatic course
(2) the probability to get infected
(3) the probability to infect somebody else
4/25
(1) the probability to have a symptomatic course
(2) the probability to get infected
(3) the probability to infect somebody else
4/25
(1) determines how far an infection can spread in a school before it gets detected and testing/quarantine kicks in, which usually happens after the first symptomatic case shows up: the longer there are only asymptomatic cases, the longer the infection spreads undetected.
5/25
5/25
(1) also influences the infectiousness of a person. The viral load of symptomatic/asymptomatic people seems to be the same ( https://doi.org/10.1016/j.jinf.2020.06.067) but the probability to infect somebody else drops by about 40% ( https://doi.org/10.1101/2020.05.10.20097543), probably because they cough less.
6/25
6/25
To set the probability to have an asymptomatic course depending on the age of the child, we use data from @agesnews that was collected in weeks 35-46 2020 for outbreaks in schools in Austria (% asymptomatic cases over age bracket).
7/25
7/25
In this period of time, Austria (among very few countries in the world) had a quite rigorous policy of testing K2 contact persons (background screening). Therefore the chances are high that they detected many asymptomatic cases connected to an outbreak.
8/25
8/25
Of course, the chances of detecting a symptomatic case over an asymptomatic case are still higher: asymptomatic cases from the beginning of an outbreak might already test negative by the time the background screening starts.
9/25
9/25
So we probably underestimate the probability to be asymptomatic for all age groups but there should not be a systematic bias towards underestimating the probability more in certain age groups if testing of K2 persons was age-agnostic (BIG "if").
10/25
10/25
(2) The probability to get infected: there is a range of theories why children are less likely to get infected (the nature article goes into details) but it is very hard to quantify this effect on the biology side of things.
11/25
11/25
For our study, we again use data from @agesnews on the number of infected students and teachers per school outbreak. Number of infected for teachers (blue) and students (red), school types increase in age of the children from left to right.
12/25
12/25
Again we rely on the fact that there was extensive testing of K2 persons in the period time from which this data comes (weeks 35-46 2020). If the background screening was age-agnostic, we probably undercount the total number of infected, but we do not have an age-bias.
13/25
13/25
(3) The probability to infect somebody else. There are theories around that children are less infectious but this not talked about in the literature that often. One factor very likely is the lower probability to have a symptomatic course (less coughing).
14/25
14/25
Also, children have a smaller lung volume and exhale less air. Both comes down to creating less aerosol. I think this effect is very hard to quantify, because it gets conflated both with the probability to be symtpomatic, and the probability of getting infected:
15/25
15/25
Most children have most of their contacts to other children. Therefore if children are less infectious AND less susceptible, it's hard to pin down how much the contribution of each of these effects is to the overall outcome.
16/25
16/25
In our model we deal with this uncertainty by assuming that both effects contribute equally to lowering the probability of transmission between children and calibrating the effect using observational data.
17/25
17/25
So to summarize: we rely on the assumption that background screening around positive cases was age agnostic and we can (to some extent) rely on the observational data we have to calibrate the age-dependence of susceptibility and infectiousness of children.
18/25
18/25
For the calibration, we match both the distribution of cases to teachers and/or students, as well as the overall outbreak sizes in school types with different age structures: primary (red), lower secondary (purple), upper secondary (blue), secondary (green).
19/25
19/25
The result of that calibration is that children are about 2% less susceptible and infectious per year they are younger than 18. That's not that much: a child in the first year of primary school will be 24% less susceptible than a young adult in their final year of school.
20/25
20/25
This effect is dwarfed by other things: For example, airing the classroom once per hour reduces the chances of transmission by over 60% ( https://www.mpic.de/4747361/risk-calculator) and wearing a surgical mask reduces the risk of transmission by about 50% ( https://doi.org/10.1101/2020.11.18.20233353).
21/25
21/25
Even if we get the age dependence completely wrong, the effect is small compared to other effects in the model. We still see quite different outbreak dynamics in primary schools and in our policy brief we also advise for different measures for school types. So why is that?
22/25
22/25
I think that one of the main factors that is currently not discussed at all has to do with how teaching is organized in different school types: Primary schools have smaller classes (on average 18 students/class) than for example secondary schools (24 students/class)
23/25
23/25
The number of classes is lower (8/school for primary, 28/school for secondary). In addition, teachers in primary school mostly teach a single class whereas in other school types, teachers teach multiple different classes every day. All of this reduces possible contacts.
24/25
24/25
Therefore I think our assumptions regarding age are justified and our outcomes are robust against uncertainty in the age dependence of transmission/reception risk.
BTW: the model code is Open Source and available here: https://github.com/JanaLasser/agent_based_COVID_SEIRX
BTW: the model code is Open Source and available here: https://github.com/JanaLasser/agent_based_COVID_SEIRX
Obviously, I forgot to link the policy brief. Here it is: https://www.csh.ac.at/wp-content/uploads/2021/01/Policy-Brief-Schulen_Final-20210120.pdf
