The last couple of weeks have seen discussion over smartphone contact tracing apps change substantially. At first, protection of privacy and use of location data were the centre of debate. Now increasing attention is being paid to their effectiveness. Over that time, I’ve been taking part in the first public trial of the UK’s proposed solution, adopting the centralised model which has attracted so much criticism. This is my ten day report, in which I look at the problems we’ve encountered, and where this is all going.

The app

The idea behind the current version (1.0.2, build 356) of the NHS COVID-19 app is fairly simple: using Bluetooth LE, it assesses the effective distance and duration of contacts with other smartphones with the app installed. Each user has a unique ID, and periodically each phone uploads details of its contacts to the server.

If a user feels that they have the symptoms of Covid-19 infection, they’re encouraged to report it using the app, which performs basic diagnosis. If the app advises that you’re likely to have Covid-19, then you’re invited to report that to the server. When that happens, all those you came into contact with (close enough and for a sufficient period) during the period you may have been infectious are then sent push notifications advising them of the risk that they may have been infected, and recommending that they go into quarantine.

Those who are given a probable diagnosis, based on symptoms, are offered a test for Covid-19. Testing and results are not currently included in the information held on the server, though. Human contact tracers should also contact those who are accorded a diagnosis of probable Covid-19, to trace others who may not have registered with the smartphone contacts.

Trial subjects

The Isle of Wight is the UK’s most populous ‘small’ island, which is currently quite isolated, with very limited ferry services being the only way to escape in either direction. Its largely semi-rural population of around 140,000 has a high proportion of older people, and is thought to have relatively few smartphone users. During the current Covid-19 pandemic, rates of positive cases have remained low compared to the mainland, currently around 120 positive tests per 100,000, but it has had a relatively high number of deaths, which probably reflects the many older people with significant medical conditions.

Adoption

Adoption of the new app has been relatively high. Estimates of those with suitable smartphones suggest the maximum possible is probably around 80,000. UK government has been annoyingly vague about how many use the app: although the number registered with the server must be well-known, figures given refer to the number of “downloads” of the app, and range from “at least 50,000” to 73,000.

I suspect the real total of those on the Island who have registered through the app is around 50,000, i.e. slightly more than 60% of those with compatible smartphones. Given that this is a well-publicised trial, that isn’t as good a response as might have been expected, and perhaps explains why more specific figures haven’t been given. Follow-up surveys are in progress to establish, among other things, why more users haven’t installed the app yet.

Possible cases

Early in the trial, a government source reported that an average of 25 users per day were being diagnosed as possible cases of Covid-19 by the app. No further information has been provided about this, but if that was accurate, I suspect that in the first ten days of its general use on the Island, between 200-250 people have been diagnosed as having Covid-19 on the basis of their symptoms.

If each of those cases agreed to contact tracing, and returned an average of 3 contacts who are at high risk of infection, that would have resulted in a total of at least 800 people being advised to quarantine (including the index cases). With an average of 5 contacts each, a total of at least 1200 people would have been advised to quarantine, which is approaching 1% of the population.

There have been stories of whole streets being put into quarantive because of single cases. These claim that a care home worker living in one of a long run of terraced houses reported symptoms. Because the wall between them and their neighbour is relatively thin, this then triggered the neighbour as a contact, and so on all the way down the street.

Fortunately, these are completely false. The app doesn’t cascade from index case to contacts and on to those contacts’ contacts, or it would quickly propagate to the entire population. Not only that, but few areas on the Island have long sections of terraced housing, and Bluetooth attenuation is likely to rate through-wall contacts as being well outside the margin of risky exposure. But this type of story makes entertaining reading in the press.

Positive cases

The UK does produce quite finely resolved daily figures for positive cases, and I have looked at latest numbers for the Island, blocked into periods of ten days. During the steep rise in UK infections, the number of those testing positive on the Island averaged between 2-3 per day, and peaked at nearly 5 per day for the next two ten-day periods. Since the app trial started, they have averaged 2.4 per day, around a tenth of those allegedly being diagnosed by the app.

If contact tracing were performed not on the basis of those reporting symptoms, but only those who test positive, numbers put into quarantine would be about a tenth of those when tracing is performed as a result of symptoms alone.

As I have explained before, testing isn’t currently integrated into the app, and the results of testing aren’t recorded on the server. One obvious improvement which could be made is to use test results as a stronger indication of the need for contacts to go into quarantine. But that would require test results to be stored on the server, which in turn raises privacy concerns.

As testing is an integral part of the trial, time taken to obtain a test result is considerably shorter than is currently being reported from other parts of the UK. The target appears to be to have sampling completed within 36 hours of reporting symptoms and requesting a test, with results being returned within another 24 hours. I was a little surprised to see that the trial is sending its test swabs to a laboratory in Northern Ireland, which is almost as far from the Island as possible in the UK. Maybe logistics will simplify in time.

Priorities

The UK’s smartphone app has greatest promise not in tracing contacts, but in the early reporting of cases, to enable public health authorities to take early action on local outbreaks.

During the rapid increase in Covid-19 cases in England, it wasn’t until 20 March that daily cases exceeded 1,000. Lockdown was applied over 21-24 March, but cases didn’t peak until 7 April, nearly three weeks later, by which time there were nearly 60,000 cases in England, and the total number of deaths in the UK had risen from 194 to nearly 7,500.

Future management of Covid-19 depends on reducing the delays in detecting and reporting cases. Currently, time is on the side of the coronavirus, and allows small outbreaks to grow almost invisibly for more than two weeks. Smartphone reporting isn’t the only method of achieving this, but currently looks the most promising. Experience shows that online (browser-based) reporting and call centres are only used by a small minority.

I’m unaware of any rigorous assessments of the reliability of self-diagnosis, even with the assistance of the algorithm built into this app. However, the evidence available so far indicates that 90% of all those diagnosed by the app turn out to have negative test results. Whatever these data are used for, integrating test results would make a substantial difference to reliability of diagnosis, albeit adding a delay of around 3 days to that confirmation.

The current version of the app has two significant omissions too. It provides no means for parents/carers to enter symptoms on behalf of their children, and would result in significant under-reporting of cases among the under-16s, a group which is already not well understood. It also fails to follow up on test results, leaving users with negative test results wondering whether they should remain in quarantine.

Testing and utility

There are three different sets of tests and diagnostic criteria for Covid-19:

• symptoms only, as used by the app to result in what WHO terms a suspect case,
• positive laboratory test for virus, which is the primary criterion for what WHO terms a confirmed case,
• positive test for antibodies, which doesn’t indicate active infection but some degree of immunity which may have resulted from prior Covid-19.

As I suggest above, diagnosis on symptoms alone has a very high false positive rate; it also has a significant false negative rate, for those who do have Covid-19 but whose symptoms are so mild or different that they don’t self report. It can be valuable for providing early warning to public health authorities, and for provisional contact tracing, but needs to be linked with testing.

Laboratory testing for the virus is currently the only means of confirming infection. It too has a high false negative rate, estimated at between 20-50%, so may need to be repeated when symptoms or suspicion persist. It does appear to have a low false positive rate, though, and then mainly in those who had Covid-19 some weeks earlier, but whose swabs still contain the (nonviable) remains of virus.

Antibody tests are still in their infancy, and their track record so far isn’t encouraging. A recent comparison of claims by manufacturers indicates that sensitivity can be as low as 45%, although specificity tends to be higher than 95%. When any will be suitable for use on millions is anyone’s guess, despite periodic reports that they will be available for routine use very soon.

Immunisation

Several governments are starting to make rash promises of the availability this autumn of large quantities of proven immunisations against Covid-19. However promising current trials might prove, achieving a sufficient number of immune people isn’t going to be quick by any means. In the UK, for example, this might require something of the order of 30 million successful immunisations. Even at an unprecedented rate of one million a day, that would take at least a month to complete. All that has to performed with effective distancing measures too.

Conclusions

Nothing about the current pandemic is easy. While we all hope for rapid gains, we have to rely on making steady inroads to reduce transmission, and to keep it under control. The UK app still has some way to go before it could have significant impact on monitoring and control of infection, but none of its current shortcomings are unsurmountable.

We need to be more realistic in what we expect from it, though: as a timely reporting tool it has obvious value which isn’t matched by anything else. UK government is going to have to be more realistic about what it expects from the app, and more open about its true potential. Confused aims and poor adoption could prove fatal, and dash hopes of ever using smartphone technology to our benefit in tackling the pandemic.