Is it possible to trace contacts using Bluetooth on smartphones?

One of the great hopes in harnessing technology to tackle the current pandemic has been in contact tracing using smartphones. One of the few options to limit the spread of infection, Apple and Google combined forces to offer facilities on iOS and Android which could enable this. So far, though, smartphone-based contact tracing hasn’t proved successful. Why?

Early debate centred on privacy. Some systems proposed rely on the use of location data, which would be an obvious gift to any government for tracking its citizens. Even those which only record potential contacts centrally, such as that currently being trialled here on the Isle of Wight, have been attacked for putting too much sensitive data in the hands of those who can’t be trusted.

A few more technical assessments expressed concerns over whether an app could remain in the background in iOS, yet be able to wake itself up when another smartphone came within range. Apple’s support for this was released in iOS 13.5 as part of what it now refers to as exposure notification rather than contact tracing. The UK’s app predated that, and uses its own ingenious techniques which are acknowledged to not be perfectly reliable, but in practice seem to work reasonably reliably.

Another well-known problem with trying to establish the distance between Bluetooth LE (BLE) systems is the reliability of radio signal attenuation as an estimate of that distance. So far, almost all debate surrounding smartphone contact tracing has assumed that measured BLE signal strength can be relied on as an accurate measure of distance over the range 0-5 metres.

According to a recent BBC report, it’s this problem which is preventing these novel contact tracing systems from being brought into use. Citing a study by Leith and Farrell of Trinity College Dublin which was published over a month ago, the BBC report claims this is one of the main reasons that the UK’s new app is “running behind schedule”, and is being examined carefully by others developing similar apps, such as Switzerland.

Yet anyone who knows anything about the propagation of radio signals, particularly those used by BLE, has been fully aware of these limitations since the technology was first introduced. Leith and Farrell’s experimental measurements demolish one story about the Isle of Wight trial which has circulated in the UK press, that ‘contacts’ are likely when smartphones are placed on each side of the common wall between different houses in a terrace, for example.

Furthermore, this study leads us to two important conclusions with respect to the Apple-Google decentralised model for contact tracing: first, that the limits it imposes on the frequency of handshaking between phones make it more difficult to overcome these problems, and second that making the decision in the app as to whether a contact has been sufficiently close isn’t a wise choice. On both these counts, the centralised approach adopted by the UK app has a clear advantage.

On the same day, the new English contact tracing service (based on web and call-centre interfaces) released its first figures. Although a promising start, these reveal several causes for concern when you look at its data cascade.

Figures quoted cover the period 28 May to 3 June, and are for England alone. Over that week, the total number of lab-confirmed cases reported for England was 2,890, but the total number of confirmed cases reported to WHO for the whole of the UK was over four times higher at 12,616. The number reported to the contact tracing service for England was 8,000 – nearly three times the figure claimed for England alone, but only 63% of those reported for the whole of the UK. It’s also a suspiciously round number.

Of those 8,000 who tested positive, only two-thirds were prepared to allow their contacts to be traced, that’s about 5,400. From those, 31,794 contacts were identified, but only 26,985 of them could be traced, an average of almost exactly 5 contacts per person (another disturbingly round number).

Although tracing nearly 27,000 contacts may seem impressive, and certainly helps control the spread of Covid-19, it’s perhaps salutory to estimate the contacts who weren’t traced, and could have gone on to become infectious themselves. We know that the false negative rate for PCR virus testing is likely to have been at least 20%, so this series of 8,000 positive cases represents at least 10,000 who almost certainly were fresh infections during that week. If each of those had the same number of contacts as the 5,400 who agreed to have their contacts traced, then there should have been at least 59,000 identifiable contacts in total.

Thus, at best, England’s new contact tracing system managed to trace only 46% of all contacts made by fresh cases during that week.

This is enough to help control spread, but clearly needs to improve significantly. One obvious issue which these figures fail to address is the well-known problem of anonymous contacts: people with whom an index case knows they have been in significant contact, but can’t identify to a contact tracer. “The bloke who sat next to me on the bus” or “in the train” can only be traced by smartphone.

This doesn’t have to be perfect in order to significantly improve the number of contacts traced: just one or two additional to those already identified would have increased the outcome here by 5,000-10,000, raising the overall proportion of those traced from 46% to 54-63%. That could make all the difference. Yet England continues to dither about its smartphone app, which now hasn’t been updated since its initial release more than five weeks ago. It’s as if no one cares.