Published on April 3rd, 2020 📆 | 5571 Views ⚑0
Pandemics: Do we need an app for that?
Right now, with huge numbers of infected individuals and a limited testing capacity, the US has no way of knowing who’s at risk for a SARS-CoV-2 infection. The ultimate goal of socially isolating, however, is to reduce the levels of infection so that we can do what’s called contact tracing: figuring out everyone who an infected individual has been in contact with and isolating and testing them. If implemented effectively, this will catch newly infected people before they become contagious, keeping the virus from spreading.
That process, however, relies on contact tracing being efficient and accurate enough to identify anyone at risk before they move on and infect multiple new people. A new study by a group of Oxford researchers suggests that SARS-CoV-2 is simply too infectious for this to work well. The team isn’t without a solution, though: a smartphone app that caches contact information and alerts all contacts as soon as a positive test result happens.
Without a trace
Contact tracing is, in principle, really simple. Once an infected individual is identified, they’re interviewed to ask where they’ve come into contact with other people for a while. In reality, it’s a nightmare. People’s memories are faulty, and it can be difficult to reconstruct everywhere they’ve been. And it’s one thing if they know they visited a few friends or family members; it’s something else if they rode a bus or stopped by a large store. Identifying who was even in the same place at that time can take days if not weeks.
Can it actually work with a fast-moving disease like SARS-CoV-2? To find out, the Oxford team turned to an epidemiological model to explore how contact tracing and other measures might influence the spread of the virus.
Epidemiological models of a disease are based at least in part on the properties of the disease itself, like how long it remains asymptomatic and when people become infectious. In this case, the researchers used a combination of their model and real-world data to estimate some properties of the virus itself. The real world data came from a number of contact traces that had been reported by other researchers. These were plugged into the model, which then created estimates for four key properties: transmission from people who don’t yet have symptoms, transmission from those who never develop symptoms, transmission from those with symptoms, and infections caused by viruses left on surfaces by those with infections.
Once they had values for these properties, the model was run to get a picture of how control efforts would alter the trajectory of a coronavirus outbreak. Specifically, they looked at the simplest plan: isolate people with symptoms, perform contract tracing with them, and quarantine anyone who is determined to be at risk based on the contact tracing.
But it didn’t work. Contact tracing simply takes too long, and the coronavirus is too contagious.
But things look better if you could accelerate the contact tracing, so the researchers tried out an idea that essentially makes it instantaneous: a smartphone app. The app simply runs in the background and registers other phones it ends up in proximity to, caching the information. If the user receives a positive coronavirus test (yes, we know that’s actually a negative), then everyone on who has been cached can receive a notification that they should self-isolate and get tested.
The actual success depends on how well we do at getting known infected people into quarantine and getting everyone who is a potential contact to isolate themselves. But if we assume a 70 percent success rate on both of those, then the function of the phone app is obvious. Even under a model that assumes very quick manual contact tracing (two days until contacts are isolated), the virus will infect enough people to keep the pandemic growing. In contrast, with instant contact tracing, the 70/70 quarantine/isolation condition is enough to cause new infections to decline.
Obviously, this raises some significant privacy concerns. To be effective, the app not only has to simply register something about nearby phones—Bluetooth IDs, for example—it has to exchange enough information that the phone can be contacted if testing reveals a potential risk. In addition, the system would have to be integrated into the testing services so that a positive test result produces near-instantaneous alerts.
Realistically, all of that requires a centralized database that integrates both data from the application users and the testing service. This should allow the app’s individual users to remain anonymous and would make it easier for security best practices to be followed. It does, however, create a single point of failure, both for functionality and security.
How well the public would accept those trade-offs is probably a matter of cultural expectations. China, with its large state infrastructure and citizens used to lower levels of privacy, has already experimented with mobile-application-based control of infection risks. Contrast that with the US, where there’s not even any centralized authority managing the testing, and the citizens tend to mistrust governmental intrusions. Not only will the decision to pursue such a system be more complicated, but it’s not even obvious what agency would run it or how they’d start integrating all the disparate testing systems.
As for the model the idea is based on, the usual caveats still apply: any model requires a lot of assumptions about the properties of the disease, and an evaluation of data that remains incomplete during this rapidly expanding pandemic. That said, there’s no question that the challenges and time involved in the standard approach to contact tracing raise legitimate questions about how soon we can turn to it in countries where infections are still spreading largely uncontrolled, like Italy, Spain, and the US. The authors’ suggested approach may allow some countries that can manage the challenge of setting up a system like to be able to switch to contact tracing sooner than expected.
For more consideration on the privacy issues, there’s a more extensive discussion in an earlier issue of Science. If you think the whole thing is a good idea, MIT has already created an application that implements some of the ideas discussed in this paper.
Science, 2020. DOI: 10.1126/science.abb6936 (About DOIs).