Making Tracks: 
 The Landscape of Digital Contact Tracing

As the viral wave of COVID-19 spread through societies across the globe, we quickly determined that contact tracing - a key part of all public health toolkits - could be assisted or automated by using mobile apps powered by Personal Data Accounts.

While hailed as an idea of great potential, the use of these apps rightly gave rise to huge public anxiety on the private and ethical use of the data that would be collected. Different contact tracing app architectures soon emerged, some prioritizing maximally private collection and analysis of data, others prioritizing governments and public health authorities' ability to gather and process as much data as they could.

In April 2020, we set up our first global Hackathon, Hackfromhome, which had 822 participants from 65 countries, and partners from across the academic, medicine, tech, and corporate life spheres. Our aim to generate ideas for privacy-preserving tools to fight the virus.

Out of this came the Safetrace project, an idea from a Case Western Reserve University-led team to use Personal Data Accounts. Dataswift's private analytics functionality and an innovative network model helped them create an app that could privately store COVID-19 relevant data while also allowing for anonymized central computation of users' infection risk based on who they had met.

Working closes with Case, Dataswift and the HAT-LAB won funding from Innovate UK to develop the idea further - and a new app, ShareTrace, was born.

As part of our Innovate UK funding, we researched the landscape of contact tracing apps and identified over 130 contact tracing apps. We gathered comprehensive information on 105 of these apps and focused on identifying their main architectures and assessing their success in achieving widespread adoption among the target populations.

Our results highlighted the relative success and popularity of decentralized and Bluetooth based approaches. This success is in very large part due to the widespread use of the Google Apple Exposure Notification API by public health authorities developing apps for use against COVID-19, and the requirements these tech giants enforced upon public health authorities seeking to use it. Despite concerns on the epidemiological efficacy of privacy-maximizing GAEN based apps, they are now the most widespread system in use.We see strong potential for systems like Sharetrace, which combine the benefits of both decentralized and centralized architectures whole also placing control of the data in the hands of individual users

• 65% of apps used Bluetooth, while 19% used GPS to pick up contact history. 12% used both.
• In general, we found that decentralized, Bluetooth based apps were considerably more successful in attaining downloads. 
• Our findings showed that, on average, daily download rates amongst Internet users of a country were over twice as high for decentralized apps (0.40%) when compared with centralized ones (0.20%).
• Apps with over 150 days since release have had greater penetration on average if they were Bluetooth-based rather than GPS-based. Nevertheless, three out of four of the apps with the greatest penetration are GPS-based. Among apps with under 100 days between launch and latest download figures (including many Bluetooth-based GAEN apps developed by European governments), penetration levels are considerably higher on average among Bluetooth-based apps than GPS-based apps, the majority of which are concentrated around 0-5% penetration.
• On average, GAEN apps were downloaded twice as much per day as non-GAEN apps amongst Internet users in a given country; 0.42% for GAEN apps compared to 0.21% for the non-GAEN apps.
• One interesting aspect of the data we gathered is the correlation between a country's size and the proportion of its Internet-using population that downloaded the official DCT app for their territory; in smaller countries and jurisdictions, a far higher proportion of the population downloaded their app.
• Delving deeper into the data, we found that nations with under half a million inhabitants had an average penetration of 52.66%, well over double the average penetration across all state-backed national applications of 22%. Unusually, we also found that the 5-6 million range had an average penetration of 41.82%, with a range of 25.7% to 57.65%, indicating that this result is not due to a hugely successful outlier.