Digital contact tracing may contribute to epidemic suppression of COVID. What are the trade-offs in choosing centralised or decentralised systems? . 1/n
github.com/BDI-pathogens/…
github.com/BDI-pathogens/…
There are three broad aims to be optimised: prevention of infection and disease, minimisation of disruptive requests to isolate, and maximisation of privacy. 2/n
A clear assessment requires acknowledging that we don't know as much as we’d like about the details of how this virus spreads. And nor do we know enough about the context of how this intervention fits in broader public health measures that will get us safely out of lockdown. 3/n
A tool that allows us to adapt and learn is likely to be more effective than a tool that doesn't. Decentralised systems can be adapted, but as yet, there is no good approach to learning. 4/n
Second, there is a growing literature confirming that COVID-19 is often transmitted before symptoms. To be useful, contact tracing needs to be very fast; this was the motivation for developing this in the first place 5/n science.sciencemag.org/content/368/64…
Centralised systems can do this speed more safely. Efforts will be needed to minimise ‘boy who cries wolf’: ‘learning' will help us more quickly land on a good trade-off. The alternative with the decentralised system is very rapid testing, but is anyone ready? 6/n
Third, any system will benefit from very high rates of adoption and use, which requires trust in the system, and high quality engineering. Here decentralised systems seem to be preferred, with support from Google/Apple. 7/n
Centralised systems can gain trust with active governance and transparency, and by meeting privacy and security guarantees. 8/n
There are difficult trade-offs. My plea: let's not pretend this debate is easy. We just don’t know enough about COVID transmission to be sure that we can set up the perfect contact tracing system now. If this was an easy task, we wouldn't need these tools at all. 9/n
I very much welcome debate, and in that spirit, respectfully disagree with the statement by @marcelsalathe and Prof. Cattuto that epidemiology and function are separable. 10/n github.com/DP-3T/document…
We have shared simulations github.com/BDI-pathogens/… and are currently working on the relative benefits of immediate trace-then-test versus test-then-trace, of introducing better risk scoring, and of incorporating other interventions. 11/n
Those first simulations show that the effect of digital tracing in reducing R can be substantial, but also there are drawbacks, mostly too many people getting notifications. Improved self-diagnosis, risk scoring, rapid testing, and controlling the epidemicall improve this 12/n
There are also proposals to differentiate between different types of notifications. 13/n
Please do contribute corrections, references we missed, and suggestions for improvements. 045.medsci.ox.ac.uk/contact-us and github.com/BDI-pathogens/… 14/n
In this noisy debate, there is a real risk of adopting a system that is ‘perfect engineering' but contributes less to epidemic control than it could. We needs tools for the virus we face, not the virus we wish we faced. 15/n
