Thread by @jennydove

🧵 Lots of great conversation about the latest study on the effectiveness of #ExposureNotification out of the UK. What is it actually saying and what are the implications? /1🧵

@lmuscato has a high-level overview via @techreview:
technologyreview.com/2021/02/11/101…

Top line result of this study: "We estimated that for every 1% increase in app users, the number of infections can be reduced by 0.8% (from modelling) or 2.3% (from statistical analysis)." via @turinginst

🥄Let's dig in further. /2

turing.ac.uk/blog/demonstra…

What are the goals of #ExposureNotification apps? They alert you if you've been nearby someone who tested positive for #COVID19, prompting you to get tested and self-quarantine. That breaks chains of transmission. /3

Why is this data only showing up now when #ExposureNotification apps were first launched last summer? We designed a privacy-protecting system. This also limited analytics collection. So researchers have had to be creative about data collection. /4

github.com/BDI-pathogens/…

These results are a product of #OpenSource collaboration - the UK team's approach has been informed by what others have learned, and others can take some of the tooling the UK built and drop it into their own apps. /5

While the @NHSuk spent 💰💰💰 on this app, they were also trailblazers in developing the technology. Since then it's become far cheaper to deploy one of these solutions. It's hard to beat this on cost per lives saved, and I hope this study encourages others to deploy EN apps. /6

Could we see these sorts of results elsewhere? The UK has some of the highest EN app adoption in the world at 49% of the eligible population. ➡️they send more alerts per index case - 3.2. But nothing in the study indicates lower adoption would mean no impact, just less. /7

Of those who get notified, they're slightly less likely to test positive than those notified by manual contact tracing (MCT) - 6.1% vs 7.3% for close contacts and 13.5% for direct contacts. 🤓 This data exists because the NHS invested in their technology systems. /8

One feature that the UK built into their app: consent to app-based contact tracing after testing positive. 72% of users consented to this and notified more potential contacts, raising the number of notifications from 3.2 to 4.4. Some people prefer phones to case investigators. /9

"In summary, contact tracing with the app identifies contacts who are approximately equally likely to be infected than with manual contact tracing." The paper suggests that these might be non-overlapping groups, so MORE people are getting notified and testing positive. 💃💃💃/10

The app is sending notifications to the right people, and the right amount of notifications. Similar to what @marcelsalathe and his group have shown, as well as research from @pabloryr et al (linked) /11

nature.com/articles/s4146…

⛓️ Breaking chains of transmission takes more than a notification - notifications need to be fast and people need to take action after the notification (aka test and quarantine). But without the notification, people don't take action. /12

When modeling the number of cases prevented, one method looked not just at the immediate chains broken, but went down the line to how many people those contacts might have infected themselves. They get a range of 176-382k prevented cases from this method. /13

The other method was "matched neighbors" where they took similar demographic regions near each other with different app adoption and compared outcomes. That led to an estimated 594k prevented cases. /14

These "averted cases" were just from November and December - thousands of lives saved in two months from just downloading an app. NHS says CFR is 2%, so that's somewhere between 3,520 and 11,880 lives saved. /15

Finally, they explored if improving the app might improve the outcomes. The largest improvements all come from increasing adoption, whether getting regions with low adoption higher or improving usage across the board. There are also gains from app enhancements. /16

"The app is best understood as part of a system of non-pharmaceutical interventions, not in isolation. The specific role of app-based contact tracing is to speed up contact tracing, and to reach more people per index case." Aka reduce transmission, save lives. /17

Other interesting bits of data: as expected, we see more notifications sent when case numbers increase. Notice the shape of the curves matching nicely. The one labeled "c" is app notifications, the blue chart is overall cases in the UK. /18

@christophraser has his own thread summarizing his research /19

https://twitter.com/ChristoPhraser/status/1359086506653208577

Massive credit goes to the team behind this study - @ChrisWymant, @LucaFerrettiEvo, @OxfordViromics, @mishkendall, @markbriers, @cholmesuk, @christophraser, and a few others not on Twitter. /20