I have outlined the short and expensive life of the UK’s smartphone app intended to trace contacts of those with Covid-19. While that was briefly flourishing, other nations were completing their own equivalents, several of which used the new frameworks for de-centralised exposure identification introduced by Apple and Google. This article looks at progress in smartphone contact tracing, drawing in part on an excellent report by Rory Cellan-Jones and Leo Kelion of the BBC.

At its heart, what smartphone contact tracing aims to do is fill the gap in conventional contact tracing, which can’t identify and trace anonymous contacts – people you don’t know who you might sit with for some time when using public transport, or visiting a restaurant, for example. The way this can be done using smartphones is for the two people’s phones to recognise one another, estimate their distance apart, and the period they were in close proximity. At some later time, if one of those two contacts develops Covid-19 infection and tests positive, then the other can be notified that they came into contact, and given advice about being tested and quarantine.

A naive approach to doing this would be to send information about contacts to a central server, where exposure matching takes place, and decisions can be made about which contacts should be notified. The major problem with that centralised approach is that anyone with access to that centrally held data would have a great deal of sensitive information which could be used to build a picture of people’s social contacts, activities, even their love lives. In Europe and North America, few people have sufficient trust in their governments to provide them with so much personal information.

Apple and Google proposed a decentralised model, in which contact matching is performed in each phone, and no sensitive information is sent to anyone else, something which has achieved much wider (but still not universal) acceptance. This was implemented as Exposure Notification in iOS 13.5, and has just been released as open source.

A major disadvantage of a wholly decentralised approach is the lack of information collected. For a smartphone app to have wider value in Public Health, it needs to notify positive cases, identify contacts and follow each up to determine whether they too become ill. Numbers and locations are of fundamental importance to those monitoring potential local outbreaks and trying to control rates of infection.

Several models have been built to try to determine the proportion of smartphone users needed for these means of contact tracing to be beneficial, and how successful a popular system could be in terms of managing local Covid-19 outbreaks and spread. There is also the important question as to whether an app could partly or completely replace any elements of traditional contact tracing, an activity which is normally labour-intensive and costly.

Among the national smartphone contact tracing apps, Germany’s remains the benchmark, and was delayed in order to adopt the decentralised model using Apple and Google’s frameworks, thus to give a good guarantee of protecting the privacy of its users.

According to figures obtained by the BBC, of the 83 million people in Germany, only around 16 million have downloaded this app since it launched in June. That’s less than 20% of the whole population, and probably around a quarter of all smartphone users. Its adoption has been far below the percentages envisaged by those modelling the benefits of such apps, which normally start to become significant once adoption exceeds 50% and rises towards 80%.

Because the German app has respected data privacy, Public Health authorities have gained almost no information about Covid-19 outbreaks from the app. They know that about 500 users of the app have tested positive – that’s an insignificant proportion – and no one can find out how many contacts have been successfully traced as a result. There is also no record of how many exposures resulted in false alarms, nor of missed diagnoses. There are similar problems with Switzerland’s app, and a lack of data for Ireland’s too.

Adoption of smartphone contact tracing apps has also been very poor in Japan (6% of the population), Italy (7%), and France (3%). To date, no national smartphone contact tracing app has been demonstrated to have had any significant benefit in controlling outbreaks, or significantly reducing the incidence of Covid-19. Only draconian access to personal data, as used in South Korea, seems to have brought any positive results.

There’s another serious problem which hangs over both centralised and decentralised contact tracing apps: the reliability of distance estimates using Bluetooth signal attenuation. This isn’t new, but had become overshadowed by the issue of privacy protection. Sceptics argue that anomalous propagation of Bluetooth signals result in too many false positives and negatives. One of the potential advantages of a centralised model is that it would allow the use of machine learning to adapt thresholds to minimise error, but now that all the major actors seem to have abandoned that model, it’s unclear whether Bluetooth signal attenuation will ever be reliable enough for this purpose – something which seems to have been quietly ignored.

Conventional contact tracing methods also haven’t been faring well. In Europe, modern social mobility and travel increase the total number of contacts, and result in many of them being anonymous and untraceable. Although national figures have appeared impressive, areas most prone to local outbreaks are often those in which most contacts are anonymous, and people are least likely to engage with contact tracing services. The danger is that local outbreaks can grow to the point where they become very hard to control, and threaten to return to the explosive rise and spread of infection which nations in Europe experienced in March and April.

Governments in Europe and North America need to consider critically why they are failing to engage the public. Lack of trust is one obvious issue which has limited the adoption of smartphone apps, but those apps don’t appear to have been designed to be effective for Public Health either. It would appear that no government has thought this out properly yet.

Postscript

A more recent report from the BBC suggests that Ireland’s contact tracing app is working. Well, that all depends what you mean by that last word.

According to the figures given, it has been downloaded by 1.2 million users of a population of 5 million: that’s 26% of the population of Ireland, which is far below the proportions deemed to be effective in modelling (60-80%). During the two weeks which it has been in use, it is claimed to have resulted in 91 “close contact exposure alerts”, which is remarkably few. It also reveals that, whatever the people of Ireland have been told, their app is collecting centralised data on them. Whether that will affect its use remains to be seen, but for the moment – whatever its developers might claim – there’s absolutely no evidence that their app has overcome any of the problems which I have described above, and it falls far short of the adoption rate to make it a valuable tool for controlling Covid-19 infection.

The Irish app might appear more promising than others, but badly needs numerical evidence to confirm whether that promise has been realised.