Covid app data reveals how outbreaks spread

During the height of the recent coronavirus pandemic, we were urged to install the government’s covid-19 app on our mobile phones, which would ‘ping’ to notify us that we’d been in close contact with someone who had the virus. Many were concerned about the app on a number of levels, including encroachment on personal liberties, but statistical analysis of the data is beginning to reveal fascinating insights into the dynamics of how the disease spread during that period, and the role that digital contact tracing might play in any future pandemic. Michelle Kendall is from the department of statistics at the University of Warwick…

Michelle - The NHS COVID-19 app was brought in in September 2020 in order to try to alert people when they were at high risk of being infected with Covid and allow them to be able to take appropriate precautions. And we've been able to show that it was effective in reducing the number of cases, reducing the burden on the NHS, and in fact saving lives. We estimate that in its first year it saved 10,000 lives. What we realised while we were looking at the data was that it was also actually revealing some really important, useful information about the epidemic that we wouldn't otherwise have been able to measure.

Chris - How did it actually work, that app?

Michelle - What it did was it worked anonymously in the background in a privacy preserving way, and all it did was just measure little low energy Bluetooth signals between phones which had the app. And if two phones are coming close contact shortly before somebody reported testing positive for covid, then it could anonymously alert the other person that they'd been near to someone with Covid recently enough to be at high risk.

Chris - But how many people actually installed that at peak? Because with all kinds of stats and that kind of thing, it all comes down to the power of a study, which is driven by the numbers.

Michelle - Yes. So we're really grateful. We actually at the peak had 18 million users, which was a phenomenal engagement with a new technology. And that meant that the app was able to be super accurate. In particular, in spring/summer 2021, we had lots of users, unfortunately lots of Covid as well. So people being pinged by the app, being told that they might be infected, were at peak times up to a hundred times as likely to be infected as the general population.

Chris - What did this enable you to probe that we couldn't tell about epidemics and the pattern of spread of diseases and outbreaks and so on before this came along.

Michelle - So it gave us a really good window into how much people were meeting up, how many contacts people were having, and then, when they were having those contacts, how likely they were to pass on an infection. The new things that we were able to do with this were that it was available straight away. We could find out the very next day how much people were meeting up. And in particular, it was specific to the very people who were infectious with Covid at the moment. So it doesn't really matter how much people are meeting up if they've not got Covid, it's the ones who are currently infectious that matter for the spread of the disease. And the app data was uniquely reporting on that. Coupled with that, you could see how likely the infections were to be passed on. We saw that likelihood decrease as the vaccination rollout came out, and then we saw that probability increase again when the omicron variant arrived.

Chris - Could you also see the effect of the different interventions? Because one of the big questions that kept rearing its head across the pandemic was, governments would implement some kind of control policy, but it wasn't evidence-based at that time because we just didn't know. We had to assume that face masks would help or not going into a shop more than three people at a time would help. Can you map any changes in the patterns you were seeing in the app onto policy decisions in England and Wales where this was running to try to work out the relative effectiveness at the time of those sorts of interventions?

Michelle - Yes, we can. Early on in the pandemic, you may remember a lot of interventions came piling in on top of each other. We had rules of six and pub closures and all sorts all happening in the same week. Of course it's hard to disentangle that, but a particular window where we do have good data was in early 2021. You may remember in the UK we had a roadmap out of lockdown with various steps of reopening, which each came at least five weeks apart. Because they were nicely spaced out, we were able to, to see what was happening. In particular, we could see that it didn't have much effect on how much people were mixing through steps one and two of reopening, and then step three was the really big one when people mixed a lot more. In fact, step four, which was billed as the final reopening, didn't make all that much difference to how much people were meeting up. So yes, we could get that sort of data and I think that the point to highlight here is that, for the future, this could be really powerful. If we roll this out in a future epidemic, then this could be integrated in the policy making so that you're getting real time feedback on your interventions and much more power to carefully balance those interventions because obviously the interventions have harms, the disease has harms, and you want as much quantifiable data as you can to be able to balance those harms.

Chris - What are the implications, then? Does this mean that if we do see Covid mark two, heaven forbid, turn up next year, or we get the next flu pandemic, we know one will be coming at some point? That basically the government will be in a much stronger position because they'll say we already have a sort of platform, we know it can work and we know we can trust the data. Are these the sorts of messages that you are distilling out of what you've found?

Michelle - Absolutely, yes. That's the reason that we've taken our time to do this analysis really carefully and get it peer reviewed so that we can prove just how effective this is and hopefully motivate the development of digital technologies like this because it could be so powerful, not only for reducing the actual burden of the disease, but for getting a window into, as you say, how the interventions are working and how effective they can be and targeting those interventions. The other category of data we got was about the days of the week. We could see the variation: that the virus spread much more on Saturdays than other days of the week. The settings, how long people were in contact for. Differences between regions, even, and the effects of particular moments. So, we looked at Christmas Day, New Year, and the Euro 2020 tournament - which was held in 2021 - and we're getting really specific information about exactly the effects of all those different things that come together to affect the spread of disease.

Chris - It's been Euros time, again. Do these sorts of mass gatherings make a big difference? And is Taylor Swift off the hook because people have been pointing the finger at her for her billion dollar worldwide tour which is causing a bit of a spike in covid numbers in the UK. I'm sure it's just an association, not causation. Do these sorts of mass gatherings really provoke big outbreaks?

Michelle - Great question. Well, the app has been decommissioned now, so we can't learn anything from the data about Taylor Swift unfortunately. The signal we got about the Euros was that, yes, match days were associated with really steep spikes in cases; transmissions were up to seven times higher than on non match days, even in the middle of the week. The final itself saw a huge spike in cases. And so, yes, stadium gigs may be something to look at, but also when there is coordinated gathering, I think that that's a real driver of spread. Certainly it was for Covid in the context that we were looking at.