The epidemiological impact of the NHS COVID-19 app

Chris Wymant, Luca Ferretti, Daphne Tsallis, Marcos Charalambides, Lucie Abeler‐Dörner, David Bonsall, Robert Hinch, Michelle Kendall, Luke Milsom, Matthew Ayres, Chris Holmes, Mark Briers, Christophe FraserNature 2021 · DOI

Abstract

Digital contact tracing (DCT) was deployed globally during the COVID-19 pandemic to augment manual efforts. This study provides a large-scale empirical validation of the NHS COVID-19 app in England and Wales. Using mechanistic modelling and statistical matched-neighbour regression, we estimate that the app prevented between 284,000 and 594,000 cases during the late 2020 pandemic waves. Our findings suggest that for every 1% increase in app uptake, infections were reduced by 0.8% to 2.3%, highlighting DCT as a potent public health intervention.

Introduction

The rapid spread of SARS-CoV-2 necessitated innovative public health responses. While manual contact tracing is a cornerstone of infectious disease control, its inherent delays and difficulty in identifying non-household contacts limited its efficacy. Digital contact tracing, utilizing the Google-Apple Exposure Notification (GAEN) framework, was proposed as a scalable, privacy-preserving alternative. However, its real-world impact remained a subject of intense debate. This paper quantifies the epidemiological impact of the NHS COVID-19 app in England and Wales between September and December 2020.

Data and Methodology

The study integrated multiple datasets, including privacy-preserving app telemetry, official public health records, and demographic data from the Office for National Statistics (ONS).

Figure 1: Spatial and temporal distribution of app uptake and COVID-19 cases. (a) App uptake by Lower Tier Local Authority (LTLA). (b) Total cases per 100,000. (c) Distribution of app uptake across LTLAs. (d) Daily number of cases in England and Wales across different study phases.

To ensure robust estimation, two complementary analytical approaches were employed:

  1. Mechanistic Modelling: This method estimated averted cases by analyzing the number of notifications sent, the Secondary Attack Rate (SAR)—the probability a notified contact subsequently tests positive—and estimated quarantine adherence.
  2. Statistical Matched-Neighbour Regression: This approach controlled for confounding variables by comparing pairs of geographically adjacent LTLAs with similar socio-economic profiles but differing levels of app adoption. This allowed researchers to isolate the "app effect" from other factors like local lockdowns.

Key Insights

Technical Parity and Reach

A critical finding was that the app's SAR (approximately 6%) was comparable to that of manual tracing (6.9%). This validates the Bluetooth-based proximity detection as an accurate proxy for risky exposure. Furthermore, the app excelled at identifying "random" contacts—such as those on public transport—who are typically unreachable via manual interviews.

Impact of Adoption

The effectiveness of the app scaled significantly with population uptake. On average, every primary user who consented to notify their contacts through the app prevented one subsequent case, effectively breaking a transmission chain. The analysis suggests a strong non-linear benefit: a 1% increase in uptake could reduce total cases by up to 2.3%.

Results

During the three-month study period, the app was used by approximately 16.5 million people (28% of the population). The two analytical models yielded consistent results:

  • Cases Averted: Between 284,000 (modelling) and 594,000 (statistical analysis).
  • Deaths Averted: An estimated 4,200 to 8,700 lives were saved.
  • Efficiency: The app sent 1.7 million notifications, averaging 4.2 notifications per index case, demonstrating a high degree of efficiency for the public health infrastructure.

Discussion and Limitations

While the study provides strong evidence for the efficacy of DCT, certain limitations persist. As an observational study, residual confounding—such as app users being more likely to follow other health guidelines—cannot be entirely dismissed. Additionally, quarantine adherence was estimated via surveys, which may be prone to bias.

Future developments in DCT should explore "backwards contact tracing" to identify outbreak sources and integrate more closely with rapid testing regimes. Nevertheless, this work establishes the NHS COVID-19 app as a vital tool in the pandemic response, proving that digital interventions can achieve significant public health outcomes while maintaining user privacy.

Summary updated

BibTeX
@inproceedings{wymant2021epidemiological,
  author = {Chris Wymant and Luca Ferretti and Daphne Tsallis and Marcos Charalambides and Lucie Abeler‐Dörner and David Bonsall and Robert Hinch and Michelle Kendall and Luke Milsom and Matthew Ayres and Chris Holmes and Mark Briers and Christophe Fraser},
  title = {{The epidemiological impact of the NHS COVID-19 app}},
  booktitle = {Nature},
  year = {2021},
  doi = {10.1038/s41586-021-03606-z},
}
Figure