Nationwide rollout reveals efficacy of epidemic control through digital contact tracing

The global spread of SARS-CoV-2 prompted the development of numerous digital contact tracing applications. While studies, such as Ferretti et al., suggested the potential for these apps to keep the reproduction number below 1, conclusive evidence of their effectiveness remained lacking. Concerns regarding privacy also arose. Existing assessments of these apps often relied on incomplete data from Exposure Notification Systems (ENS) and population surveys. This study addresses these limitations by analyzing a unique real-world dataset from the rollout of the Smittestopp app in Norway during the spring of 2020. This dataset, encompassing millions of contacts from 12.5% of the adult population, allows for a direct measurement of the app's real-world performance and its potential contribution to pandemic control. The study aims to determine the efficacy of digital contact tracing and its dependence on app uptake, offering valuable insights for public health policy.

The paper references prior research by Ferretti et al. which suggested that an effective digital contact tracing system could keep the reproduction number below 1. However, it notes the lack of conclusive evidence regarding the real-world effectiveness of digital contact tracing apps and the privacy concerns surrounding their use. The study highlights the challenges in accurately assessing efficacy due to limitations in data availability from deployed systems, particularly those using the Exposure Notification System (ENS), which prioritizes privacy by limiting data visibility. The authors mention that previous studies used a combination of incomplete ENS data and population surveys, which this study aims to improve upon.

The study utilized anonymized data from the Smittestopp app's rollout in Norway. Smittestopp used Bluetooth Low Energy (BLE) to detect nearby phones and measure signal strength to approximate distance. Data was centrally collected and analyzed to identify contacts, acknowledging the asymmetric nature of contact detection (detection in one direction doesn't imply reciprocal detection). The analysis focused on close encounters lasting at least 15 minutes. Over 26 million close encounters were recorded. A machine learning model (random forest classifier) was used to estimate the number of close contacts not identifiable by manual contact tracing. Technological efficacy was calculated using a model that considers probabilities of detection between different phone types (iOS and Android), incorporating app uptake rates among these groups. The study's efficacy estimates were integrated into an established pandemic spread model (Ferretti et al.) to assess the potential impact of digital tracing on controlling the pandemic's spread. The analysis also examined the identification of 'risky' close contacts and the characteristics of highly connected individuals (those with numerous contacts) to understand the potential of digital tracing to detect and control superspreading events.

The study's key findings include: 1. A technological tracing efficacy of 80% was achieved, indicating that Smittestopp successfully identified a large proportion of close contacts. 2. At least 11% of the identified close contacts could not have been found through manual contact tracing, demonstrating the app's ability to reach contacts otherwise missed. 3. The efficacy of digital contact tracing is strongly dependent on app uptake, with higher uptake rates leading to increased efficacy. However, even moderate uptake (e.g., 40%) can have a significant impact, potentially controlling a pandemic with a reproduction number of 1.5. 4. The app effectively identified individuals with excessive contacts, highlighting its potential to contain superspreading events. 5. The random forest classifier used to identify unknown contacts achieved an accuracy of 89%, estimating that at least 11% of risky close contacts were random and likely missed by manual tracing. 6. Analysis using an epidemiological model showed that while achieving complete pandemic control through app-based contact tracing alone requires extremely high uptake rates, a significant impact can be realized with more moderate uptake rates, especially when coupled with other control measures.

The findings demonstrate that digital contact tracing using smartphones can significantly enhance traditional methods, reaching contacts that might otherwise be missed. The strong dependence on app uptake highlights the importance of public health campaigns to increase adoption. While complete control requires high uptake, even moderate uptake can provide substantial benefits when combined with other pandemic control measures like social distancing. The ability to identify individuals with a high number of contacts is a key advantage, potentially allowing for targeted interventions to prevent superspreading events. The study acknowledges that the results are specific to the Smittestopp app, which predates the ENS standard; however, the authors anticipate that ENS-based apps would achieve comparable or better efficacy. The study's use of a comprehensive dataset allows for a more robust evaluation of digital contact tracing's effectiveness compared to previous studies reliant on incomplete data or surveys.

This study provides strong evidence for the efficacy of digital contact tracing apps, demonstrating their capacity to identify a significant proportion of close contacts missed by manual tracing and aid in controlling superspreading events. While high uptake is crucial for maximizing impact, even moderate rates can make a substantial difference when used alongside other control measures. Further research should focus on improving app uptake strategies and integrating digital contact tracing into broader public health initiatives. The development of more privacy-preserving yet equally effective technologies is also important for maximizing the effectiveness of this critical tool.

The study's data originates from Smittestopp, which predates the ENS standard. While the authors expect comparable or superior performance from ENS-based apps, this hasn't been empirically verified. The study did not directly link app-detected contacts to infection data, preventing a direct assessment of the impact on infection rates. Furthermore, the estimates of the fraction of random contacts are conservative. Finally, the study acknowledges that the app’s efficacy is strongly dependent on the level of app uptake in the population.