"Sustainable containment of COVID-19 using smartphones in China: Scientific and ethical underpinnings for implementation of similar approaches in other settings"
github.com/BDI-pathogens/…
github.com/BDI-pathogens/…
This is work with clinical-scientist @OxfordViromics and ethicist @michaelethox . We'll be following up with the analytics piece soon.
We'll be updating the docs on the github. Some of this is on medarxiv, but I think they are swamped so we can update this version quicker.
Accompanying this policy discussion, we did some data analysis and math modelling to test the feasibility and potential effectiveness of this approach.
"Quantifying dynamics of SARS-CoV-2 transmission suggests that epidemic control and avoidance is feasible through instantaneous digital contact tracing" github.com/BDI-pathogens/…
(There has been a version of this on Medarxiv since last week, but, please use this one on github which will be kept up to date.)
A few points from the paper. 1/ We recap estimates of doubling time, the incubation period. 2/ We estimate the generation time distributions, the distribution of times from being infected to transmitting. From this we also get R0.
3/ We estimate the proportion of transmissions that occur before people get symptoms in two separate ways a/ transmission clusters, and b/comparing generation time and incubation period distributions.
4/ Here's our summary. Imagine a large number of infected individuals lined up by their time since infection (x-axis). There will be lots of variation between people, some mild some serious, etc. Y-axis is mean rate of transmitting to others, broken down according to type. 
5/ As is usual in modelling, we explored uncertainties in all the assumptions and parameters. Across the range, we found that because of pre-symptomatic transmission, isolation of patients alone would not stop the epidemic. Nor would classical contact tracing: too slow.
6/ However, this changes if users use an app-based approach that records past contacts. If a person is verified as a new infection (by test or empirical diagnosis), contacts are immediately notified if they may have been infected, and what measures to take.
This approach has the potential to reduce R<1, for a range of uptake and adherence. Epidemic rate of change is r, solid line separate growth (red) from decline (green).
7/ Since the assumptions are important, as are delays, we created a Shiny app to explore them. bdi-pathogens.shinyapps.io/covid-19-trans…
8/ The maths allowed us to solve the problem quickly, see github.com/BDI-pathogens/…
9/ However, the maths potentially under-estimate the impact of the intervention, which is recursive. Also, they force dependencies between delays. Next step will be simulation.
10/ There are implementation issues that could complicate this approach. Public health is about linking things that work, and getting public trust. Too early to tell if app based-tracing is ingredient of "epidemic control", but the science increases policy options.
11/ This feels like useful moment to recall Ronald Ross's great discovery: you don't need to stop all infections to stop an epidemic, you need to get and keep R<1. A population that has a system of public health that does that has 'herd protection'. Not easy, but not impossible.
12/ There is no one approach to get us to 'herd protection', any public health system that is adaptive can get us there. ('Herd protection' is not 'herd immunity', the latter is used to describe the population effect of vaccination.)
13/ As there has been some interest, and we are assisting with some feasibility studies in different countries, our comms team put together a website to collate resources and assist discussion. 045.medsci.ox.ac.uk
14/ Remember, there is no one approach, or combination of approaches. There isn't much time, this is a scientific feasibility study; many things need to be prioritised, preparing and protecting hospitals and people, achieving rapid social distancing, maintaining social cohesion.
15/ The WHO has been compiling information and comparing approaches from countries that have, for now, managed epidemic containment. Much inspiration there.
(not peer reviewed.)
16/ Important note: the data findings are not peer reviewed. Released early because a/ It's now good practice in outbreak analysis & a requirement b/ It seems appropriate to have a wider debate given general question of app-based approaches and ongoing feasibility studies.
