COVID-19: An opportunity to help by modelling testing and tracing to inform the UK government

post by JustinShovelain · 2020-04-17T17:21:48.656Z · LW · GW · 2 comments

Contents

  If you want to help with modelling, comment in the document above. One comment should include your contact details so that Tim can contact you and so that you can find each other. I've started a comment thread in the document for this purpose.
  Some possibly useful resources to prompt improved models: 
  My guess at the best type of software for it:
None
2 comments

Tim Colbourn, a leading UK epidemiologist, is looking for help with modelling the effectiveness of testing and tracing for COVID-19, along with comparing its effect and cost to other interventions. He has composed a document in order to suggest that a form of testing and tracing be implemented by the UK government: "PTTI: A Population-wide Testing, contact Tracing and Isolation strategy for the UK", [Tim Colbourn et al., 2020] https://docs.google.com/document/d/1vTr4IhVdfr5e-CcDfqgv3zjP-Da1Qh5bas09UHsGODU/edit

The economic case for PTTI is clear: at a cost of £33.6bn over 2yrs, it averts an estimated 271,000 deaths (£124,000 per death averted), 541,000 intensive care unit (ICU) cases, 812,000 hospitalizations, and 34,600,000 Covid-19 cases. These represent £18.0 billion in saved costs for the NHS. There are an additional £8.1 billion in prevented productivity losses due to illness and lockdown. Then there are all the other costs of lockdown not included (Table 4). For the next version we hope to have more detailed models, and include more alternative scenarios e.g. PTTI without the testing i.e. population-wide clinical case identification, contact tracing and isolation (PTTI), face masks, and even a managed herd immunity strategy for comparison. We welcome input from those interested!

Currently, at least in the UK, government appointed modellers are not modelling any alternatives to intermittent lockdown and herd immunity. No other groups seem to be publishing such models either (with the partial exception of the paper https://arxiv.org/pdf/2003.09861.pdf, which is going to be published in updated form soon). There are also some other modelling directions that can be explored, which haven’t yet been explored in the models governments are using:

• Including the spatial dimension to account for the clustered nature of transmission and local hotspots
• Doing economic cost benefit like decision analysis so we can know how beneficial various policies are in comparison to each other
• Modelling other things like the public use of facemasks, or highly effective drugs becoming available

This is an opportunity to improve how the UK government controls the spread of COVID-19 after lockdown has ended by helping model the implementation and effects of testing and tracing. This may also help other governments if they follow the UK’s lead or adapt the strategy suggested.

While we don't want to forget our comparative advantages or optimizing for the long run I agree in part with the sentiment of this post [LW · GW]. Lesswrong has already done good work in predicting, inventing, sharing opportunities, and organizing to help with COVID-19. Here is another opportunity that we collectively have the experience and expertise required to help with.

If you want to help with modelling, comment in the document above. One comment should include your contact details so that Tim can contact you and so that you can find each other. I've started a comment thread in the document for this purpose.

Some possibly useful resources to prompt improved models: 

• https://arxiv.org/pdf/2003.09861.pdf
• https://www.imperial.ac.uk/media/imperial-college/medicine/sph/ide/gida-fellowships/Imperial-College-COVID19-NPI-modelling-16-03-2020.pdf?fbclid=IwAR394EBtQCyLpox-Kjv_PeH6PdrRFYk6dGsSjENptN1dKTirvu8fwupQ1_c
• https://fivethirtyeight.com/features/why-its-so-freaking-hard-to-make-a-good-covid-19-model/
• https://www.lesswrong.com/posts/JiwHPEALELLnco5rA/testing-and-contact-tracing-impact-assessment-model [LW · GW]
• https://science.sciencemag.org/content/early/2020/04/09/science.abb6936?fbclid=IwAR2LD8mze7PstApx9I6f4knIvkDmVKZtHivpdbC9-M_A1JRHfx8OI_3ai2M

My guess at the best type of software for it:

• https://en.wikipedia.org/wiki/Stan_(software)

2 comments

Comments sorted by top scores.