Read on Twitter
1. Thread: Model-based mitigation strategies for school reopening.
I will start with a disclosure. The work described was done in collaboration with @Color Health and I was paid as a consultant for my efforts. I have no financial stake in COVID tests, treatments, or vaccines.
I will start with a disclosure. The work described was done in collaboration with @Color Health and I was paid as a consultant for my efforts. I have no financial stake in COVID tests, treatments, or vaccines.
2. Last fall, @RS_McGee and I worked with Color's @ay_zhou, @jrhomburger, and @hewillia34 to develop a stochastic network-based SEIR model of workplace testing protocols. Here's a thread about that study: https://twitter.com/CT_Bergstrom/status/1291466354751467520
3. In his first 100 days, President Biden is aiming to vaccinate at least 100M people, massively ramp up testing, and *reopen US schools*.
Last November we recognized that we could adapt our workplace model to school environments, and use it to study school reopening plans.
Last November we recognized that we could adapt our workplace model to school environments, and use it to study school reopening plans.
4. We've spent the past two months developing the model and exploring the effects of various mitigation strategies that schools may consider. Our non-peer reviewed preprint is here (and submitted to @medrxivpreprint): https://bit.ly/k12model
5 We have also developed an interactive webapp that provides a convenient way for you to explore a range of papers, mitigation measures, and so forth. (Loading may be a bit slow — be patient.)
https://www.color.com/return-to-school-model
https://www.color.com/return-to-school-model
6. In this paper, we use @RS_McGee's stochastic network-based SEIRS+ modeling framework to examine both primary and second schools. The mitigations we explore include cohorting students, proactive testing, quarantining classrooms, and vaccinating teachers. https://github.com/ryansmcgee/seirsplus
7. Like all epidemiological models, ours is a simplification of a complex world. To the degree that those assumptions do not accurately reflect the real world, the model will be ineffective at predicting even the range of possible outcomes.
8. We embrace heterogeneity and stochasticity in our model and address uncertainty by exploring a range of parameters.
Still, in a novel pandemic where many parameters remain uncertain and social and behavioral factors are fluid, some mismatch with the real world is inevitable.
Still, in a novel pandemic where many parameters remain uncertain and social and behavioral factors are fluid, some mismatch with the real world is inevitable.
9. We developed synthetic contact networks to reflect the circumstances of elementary school students and high school students. The former are clustered tightly into their classrooms, whereas the latter mix more broadly as they move from one class to another throughout the day.
10. Most transmissions occur along this contact network, but some occur through happenstance casual interactions e.g in the hallways.
We estimate transmission parameters, latent periods, pre-symptomatic and post-symptomatic infectious periods, etc. from the literature.
We estimate transmission parameters, latent periods, pre-symptomatic and post-symptomatic infectious periods, etc. from the literature.
11. Outbreaks in a school-sized community are highly stochastic events. Some introductions may explode, while others fizzle immediately. We capture this by simulating 1000 runs for each scenario, and we show the results as *jitter plots* in which one dot represents each run.
12. As expected, community prevalence has a huge effect on outbreak size in schools. Below, jitter plots for the fraction of students (blue) and teachers (green) infected in an elementary school over 150 days. Under each bar, we list the prob that >5% of population is infected.
13. High schools (below) are at greater risk, because adolescents are more susceptible than children and because of the different social structure.
In both primary and secondary schools, outbreak risk increases substantially more cases come into the school from the community.
In both primary and secondary schools, outbreak risk increases substantially more cases come into the school from the community.
14. We find that frequent proactive testing of teachers or, better yet, teachers *and* students, can be an effective measure for reducing the risk and size of outbreaks in a school.
15. We also find that cohorting students—separating them into two groups which alternate time on campus—can substantially reduce risk, particularly among students. Benefits are smaller to teachers, who remain on campus the whole time and teach each group in succession.
16. Testing and cohorting synergize. Each is effective, and each works better when used in concert with the other. I won't go into details here, but we summarize these interactions and with diagrams such as the one for primary schools below.
17. In primary schools, we can also consider isolating entire classrooms any time a case is detected therein. This is more effective than isolating just the infected individual, though one needs to consider the cost of missed days of schooling that comes with it.
18. Finally, and this is a big one, vaccinating teachers helps enormously. It protects teachers first of all, but it can also reduce the risk and size of outbreaks among students if the vaccine prevents transmission as well as symptoms.
19. At present, B117 and other highly transmissible variants remain relatively uncommon in the US. This may not last. If these become common, schools face additional challenges. We model these as well, by increasing the R0 values that index transmission rate. It's not so pretty.
20. Still, our model suggests that aggressive mitigation efforts can keep risk relatively low even when more transmissible strains become common—particularly if community prevalence can be kept at or driven down to low levels.
21. Throughout, transmissibility (quantified as R0 in our model) is a critically important parameter. The further it increases above one, the more difficult control becomes. Thus basic precautions—masks, social distancing, handwashing, etc.—are vital to any school reopening plan.
22. A few notes on limitations.
R0 is critical as mentioned above, and we don't know enough about what it is in schools with or without basic barrier precautions and social distancing. The webapp allows you to explore a range of R0 values, and qualitative trends change little.
R0 is critical as mentioned above, and we don't know enough about what it is in schools with or without basic barrier precautions and social distancing. The webapp allows you to explore a range of R0 values, and qualitative trends change little.
23. We make assumptions about contact network structure and distancing compliance when outside of school. Should these be violated, worse outcomes could ensue. We also assume student-teacher transmission is as likely as student-student or teacher-teacher. That may be untrue.
24. We also treat the rate at which cases are introduced as exogeneous. When schools aren't major drivers of community prevalence, this is probably reasonable. If schools are drivers, we may be ignoring important feedbacks between school mitigations and community prevalence.
25. To summarize, one of the best things we can do to reopen schools is to keep community prevalence low. Mitigations help, testing and cohorting particularly. Vaccinating teachers is effective at protect them and their students alike.
26. If more transmissible strains take hold in the US, opening schools and keeping them open will become harder. We need to make every effort to prevent this, and we need active surveillance to allow rapid, flexible responses if such strains start to take off in schools.
27. Reopening schools is one of the highest priorities for the US pandemic response. We are hopeful that with committed effort, those schools that are currently closed can open in the very near future, and that all can stay open for the duration of the school year and beyond.
28. @ay_zhou, CSO at @color, has a thread about the project here: https://twitter.com/ay_zhou/status/1354098854941970437
29. Update: our paper is now up on @medrxivpreprint. https://www.medrxiv.org/content/10.1101/2021.01.22.21250282v1
30. Because of important differences in high schools and primary schools, note our webapp has separate tabs and uses distinct models for these settings.
Primary school tab: https://www.color.com/return-to-school-model
Secondary school tab: https://www.color.com/return-to-secondary-school-model
Primary school tab: https://www.color.com/return-to-school-model
Secondary school tab: https://www.color.com/return-to-secondary-school-model
