Many countries are relying on Corona apps to identify the contacts of infected persons and isolate those affected in order to break the infection chains. Germany has now also published an app that uses Bluetooth technology to warn people if they have been in the vicinity of infected people. One of the first scientific studies on the subject has been published already in 2014. Computer scientist Manuel Cebrian showed together with Kate Farrahi and Remi Emonet that smartphone data can help with contact tracing during an epidemic. Today Manuel Cebrian heads the Digital Mobilization research group in the Center for Human and Machines at the Max Planck Institute for Human Development. In the Interview he explains if contact tracing needs technical support and why it can work even if not everyone installs a tracing app.
All over the world great hope is now placed in Corona Tracing Apps. What do you think about it?
Manuel Cebrian: We know that contact tracing, together with the medical care and isolation of sick patients is crucial for containing an epidemic. But a growing body of scientific evidence is making the case that human contact tracing might not be fast or accurate enough once the epidemic takes off. Especially for an elusive killer disease like Covid-19, where asymptomatic people might infect a substantial number of people. To help contact tracing get faster and more accurate, I think we need some extra technologically enabled solution.
How can computer science contribute to this?
My research interest here is to explore how high-resolution digital social-sensing, through for example collecting digital online data or analyses of social networks, can be used in disaster management and emergency response, from terrorism to biological weapons, to climate-change disasters to crowd panics, to energy blackouts or cyber-attacks. We showed, for example, that you could perform a quasi-real time estimate of damage from hurricanes and other natural disasters by exploiting publicly available information that people post on social media. We also studied predicting the spread of the flu in a community by using machine learning over Bluetooth data provided by mobile phones. This led us to believe that Bluetooth data could help with epidemic studies. When I heard from a colleague who is a biophysicist how important contact tracing is in infectious diseases, I wanted to evaluate if we could apply our experience in network science and digital social-sensing to make contact tracing faster.
What was your approach?
Around 2011 we had the opportunity to use a vast amount of real-world smartphone data for epidemic and contact-tracing models. The data, which had been previously collected by colleagues of ours included 72 volunteer students from a US university, whose smartphone usage was recorded over 9 months. The data had subsequently been used in quite a few studies. For our study, we focused on anonymized Bluetooth interactions and phone call interactions between these students, with the Bluetooth data used as an indication of face-to-face interaction, which is conducive to an epidemic.
First, we ran a standard epidemic model through the network of contacts, represented by the student’s Bluetooth network. We saw that the simulated epidemic spread through the population and we recovered a typical epidemic curve. Then we added contact tracing to this simulation, assuming that when somebody is traced, they cannot infect anyone. In Covid-19 terms that would be: You are quarantined. What we saw is that even with moderate levels of contact tracing it would push down the peak of the epidemic. So, the first test was not a surprise: If you have a perfect digital representation of face-to-face interactions, contact tracing works.
But not everybody can – or will – install tracing apps.