Privacy researchers, govts, and tech companies work on contact tracing apps, hoping to curb down the pandemic. We often hear that 56, 60, or 70% of the population would need to install these apps. These are likely incorrect reads of a paper published by Ferretti et al. Thread https://abs.twimg.com/emoji/v2/... draggable="false" alt="⤵️" title="Arrow pointing rightwards then curving downwards" aria-label="Emoji: Arrow pointing rightwards then curving downwards"> https://twitter.com/Telegraph/status/1254387257982758913">https://twitter.com/Telegraph...

A team lead by Prof. Fraser at Oxford published a detailed and insightful study on 31/4/20: https://science.sciencemag.org/content/early/2020/04/09/science.abb6936

They">https://science.sciencemag.org/content/e... propose an empirical model with various parameters, to measure the impact of isolating symptomatic cases and quarantining their contacts. No 56/60/70 inside.

Fig. 3 is often discussed online. It shows that immediate contact tracing can curb the pandemic by itself, if we isolate 80% of symptomatic cases and quarantine 40% of contacts *immediately*, or 50/60%, or 20/70%, etc. https://abs.twimg.com/emoji/v2/... draggable="false" alt="⚠️" title="Warning sign" aria-label="Emoji: Warning sign"> BUT…

It needs to be nuanced:
— there are confidence intervals in the panels of Fig 3
— it& #39;s based on a very specific empirical dataset to train the model (“early stages of the epidemic in China”)
— the authors note that this might not be generalised to Europe https://abs.twimg.com/emoji/v2/... draggable="false" alt="😬" title="Grimacing face" aria-label="Emoji: Grimacing face">

So why do we hear that 60% of the pop need to install the app vs. the “near-universal App usage and near-perfect compliance” in the Discussion section?

The model assumes a reproduction number R0 of 2.0: infected, I& #39;ll contaminate 2 people. With confidence intervals of 1.7-2.5!

A recent study from France (not yet peer reviewed) shows that this reproduction number could have been as high as 3.3 before the French lockdown: https://hal-pasteur.archives-ouvertes.fr/pasteur-02548181

Ferretti">https://hal-pasteur.archives-ouvertes.fr/pasteur-0... et al. illustrate this high R0 in supplementary materials (Figs S18 and S19).

Oxford researchers also suggested FT to use a high R0 of 3.5 for their modelling in https://www.ft.com/content/f9fbc64c-4473-4109-b6d3-737936d6805d">https://www.ft.com/content/f...

If you go back to Fig S19, that would mean that with immediate contact tracing, 100% success in isolating cases and 85% in quarantining contacts might be needed in Europe. With a delay of 24h, it seems impossible.

Is that possible given our smartphone penetration rate?

My takeaway as a privacy researcher is that (A), as much as we focus on developing sensible tracking technologies, we should remember that these are unlikely to curb the pandemic alone (because “requiring near-universal App usage and near-perfect compliance”) and

(B) that we should be careful when using numbers such as “60%, 70%, 80% of the population will have to install these apps” because that might not reflect the findings of epidemiologists. Findings that evolve as days pass, new data emerge, and public policies change.