When speed matters, take the helping hand
Image credit: REUTERS/Henry Nicholls/File Photo
Although the U-turn on the NHS tracing app is welcome, those in charge were warned of problems months ago
One of the most promising outcomes of the entire Covid-19 pandemic has been the way in which scientists and technologists have chosen to work together to try to fight the disease on numerous fronts. They have created shared databases, pooled resources on machine learning and other technologies and have been publishing results at breakneck speed.
Though, inevitably, some results wind up having to be retracted or redone and a lengthy analysis will be needed over the next few years to find out what really worked and what did not, the net result is likely to demonstrate some of the fastest progress made in epidemiology and virus treatment that the world has ever seen. The cooperation made the individual decisions by the French and UK governments to go their own way on contact-tracing apps seem more absurd with every passing day.
The NHS in the UK seemed to have adopted a policy of 'not invented here' (NIH) some time ago, despite being a keen user of private technology for clinical testing right up to genome sequencing over the years. Researchers working on IT for medicine reported trying to talk to the NHS about digital technologies for Covid-19 programmes back in March. From the beginning of April, a popular webinar series organised by the European ELLIS machine-learning research lab hosted talks by leading AI and computer scientist Yoshua Bengio and others around the world talking about their work on tracing apps. And they were building on top of existing work themselves. “We are starting from the codebase of the MIT Media Lab, an app called SafePaths,” Bengio explained then.
The aim of Bengio’s group was to build an app for use by the Canadian health services. Provinces of Canada started evaluations of the Mila in mid-May. The main question at the beginning of June was whether the AI-heavy approach used by Mila would provide better results than a more straightforward design based on statistical analysis. But there was at least a choice. In June. Not, according to successive ministerial briefings, in September, October or as the delays have progressed, the end of this year.
Michael Lewis, professor of life-science innovation at the University of Birmingham, was among those critical of the approach that the NHS and the government had taken in pushing ahead with the centralised app and in rejecting advances from the tech community. In contrast to the approach taken for drug trials and vaccine research, even picking one course of action looked a mistake.
“Why one app? We should have gone on multiple tacks. Why not go with the best test-and-trace technology that emerges; use as many digital technologies as we can. We need to get something out quickly and we need all the tools we can muster to drive the virus down,” Lewis said.
One issue with the UK’s approach was that it took a different approach from most of the apps developed around the world. The terminology around the choices – centralised versus decentralised – is slightly misleading as even the decentralised version involves reporting some information back to one or more servers. However, the key to the decentralised system that is supported by the code that Apple and Google have made available to many of the groups working on apps is that the most personally sensitive data is kept on users’ own handsets. The servers act as clearing houses for peer-to-peer transactions that check whether two handsets have passed close enough to each for long enough to count as a 'contact' and that some information on infection risk has to be passed.
The NHS went with the centralised option where the contact data is all stored in a common database. Though it is meant to be anonymised data, one early decision caused raised eyebrows. The NHS said it wanted to store that data for 20 years, ostensibly to help to deal with future coronavirus epidemics. Privacy researchers such as Michael Veale of University College, London pointed out that the data would only be pseudo-anonymised – that is, stripped of explicit personal identifiers. There remained a risk that, armed with additional data, users with access to the database could track the interactions of individuals over the period they used the app.
Lewis pointed out one crucial problem for any test-and-trace app: that it relies on users deciding to download and use it. Any tracing app of this nature, like a vaccine, only works if a sufficiently large enough cohort deigns to use it. Opting for the privacy-problematic approach would most likely depress user numbers without providing much of a benefit to future researchers or to users. There are some advantages of a centralised system in that it can deal better with situations where a user is tagged as positive for a while but then receives a confirmed-negative test. However, early trials indicated that such amber alerts were confusing to begin with and might be skipped in the final rollout.
Now, with the replacement of Matthew Gould on the NHSX app project by former Apple executive Simon Thompson, the centralised approach has finally been ditched in favour of one that uses the Apple-Google technique to support a decentralised app. The good news is that the NHSX team has been running two codebases so that it is not starting from scratch, which should lead to a rollout within months. But in an environment where time is of the essence, the NHS may well be three months behind countries that were willing to accept help willingly offered.
Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.