Monitoring the COVID-19 epidemic with nationwide telecommunication data

Joel Persson, Jurriaan F. Parie, Stefan FeuerriegelProceedings of the National Academy of Sciences 2021 · DOI

TL;DR

This paper uses nationwide telecommunication data from Switzerland to analyze the impact of COVID-19 policies on human mobility and how this mobility, in turn, predicts case growth. The findings show that policy measures significantly reduced mobility, and this reduction is a strong predictor of future COVID-19 cases, offering a valuable tool for real-time epidemic surveillance and evidence-based policymaking.

What Problem Does This Paper Address?

During the COVID-19 pandemic, governments worldwide implemented various policy measures (e.g., lockdowns, bans on gatherings, school closures) to curb the spread of the virus. A critical challenge for public health officials was to understand the effectiveness of these measures in changing human behavior, specifically mobility, and how these behavioral changes translated into epidemic dynamics. While telecommunication data offered a rich source of information on human movement, its utility for near real-time epidemic surveillance and assessing policy impact needed empirical validation. This paper addresses the gap in understanding the direct link between specific policy measures, nationwide human mobility patterns, and subsequent COVID-19 case growth, aiming to provide a scalable and timely tool for epidemic management.

Methods

The study employs a two-stage modeling approach. First, a difference-in-difference (DiD) analysis is used to estimate the causal effect of five different policy measures (bans on gatherings >100 and >5 people, school closures, venue closures, border closures) on human mobility. This method allows for the estimation of policy impacts while controlling for unobserved time-invariant factors and time-varying factors independent of policy measures. Second, regression models are used to estimate the predictive relationship between mobility and reported COVID-19 cases, incorporating a forecast window of 7 to 13 days to account for incubation and reporting delays. The models are specified as hierarchical negative binomial regression models to handle overdispersion in the data. Additionally, a mediation analysis is conducted to decompose the total effect of policy measures into direct effects (not mediated by mobility) and indirect effects (mediated by mobility).

Datasets

The primary dataset used is anonymized, aggregated telecommunication data from Switzerland, covering approximately 1.5 billion trips between February 10 and April 26, 2020. This data was provided by Swisscom Mobility Insights. Additionally, the study utilized data on reported COVID-19 cases, hospitalizations, and deaths from the Swiss Federal Office of Public Health (FOPH), as well as data on policy measure implementation dates at the cantonal level. For validation purposes, a representative panel survey on mobility in Switzerland (Intervista AG) was also used.

Key Insights and Ideas

  • Mobility as a Real-Time Epidemic Indicator: The study strongly validates the use of telecommunication data for near real-time epidemic surveillance. The observed reduction in mobility directly predicts future case counts, highlighting its potential as an early warning system.
  • Policy Effectiveness Varies: Not all policies had the same impact on mobility. Bans on gatherings of more than 5 people were the most effective in reducing overall trips, followed by venue closures and school closures. This granular understanding is crucial for targeted interventions.
  • Mobility is a Key Mediator: The mediation analysis reveals that a significant portion of the effect of policy measures on COVID-19 cases operates through changes in human mobility. This underscores the importance of mobility reduction as a strategy for epidemic control.
  • Cross-Border Mobility Impact: Border closures primarily affected commuting trips, suggesting that their impact was more on cross-border workers than on internal Swiss mobility.

Key Results

  • Human movement in Switzerland dropped by 49.1% during the study period (Feb 10 - Apr 26, 2020) compared to the same period in 2019.
  • Bans on gatherings of more than 5 people were estimated to reduce mobility by 24.9%.
  • A 1% reduction in human mobility predicted a 0.88–1.11% reduction in daily reported COVID-19 cases over a 7 to 13-day forecast window.
  • The mediation analysis showed that policy measures operated to a large degree through mobility, with indirect effects for venue closures accounting for about a third of their total effect.

Limitations and Future Work

The authors acknowledge several limitations. First, the findings are subject to the accuracy of COVID-19 case data. Second, the models are parsimonious, and external factors not captured by the variables might influence the results. Third, while the DiD analysis allows for causal interpretation of policy effects on mobility, the link between mobility and cases is primarily predictive. Fourth, the study is limited to the first wave of the epidemic in Switzerland, and future research should explore other countries and time periods to confirm external validity.

Future work could involve extending the analysis to other countries and time periods. The authors also suggest that telecommunication data can be further leveraged for more detailed epidemic surveillance and policy impact assessment. The study's findings on the mediating role of mobility also open avenues for exploring more nuanced behavioral adaptations beyond just mobility changes.

Summary updated

Abstract

In response to the novel coronavirus disease (COVID-19), governments have introduced severe policy measures with substantial effects on human behavior. Here, we perform a large-scale, spatiotemporal analysis of human mobility during the COVID-19 epidemic. We derive human mobility from anonymized, aggregated telecommunication data in a nationwide setting (Switzerland; 10 February to 26 April 2020), consisting of ∼1.5 billion trips. In comparison to the same time period from 2019, human movement in Switzerland dropped by 49.1%. The strongest reduction is linked to bans on gatherings of more than five people, which are estimated to have decreased mobility by 24.9%, followed by venue closures (stores, restaurants, and bars) and school closures. As such, human mobility at a given day predicts reported cases 7 to 13 d ahead. A 1% reduction in human mobility predicts a 0.88 to 1.11% reduction in daily reported COVID-19 cases. When managing epidemics, monitoring human mobility via telecommunication data can support public decision makers in two ways. First, it helps in assessing policy impact; second, it provides a scalable tool for near real-time epidemic surveillance, thereby enabling evidence-based policies.
BibTeX
@inproceedings{persson2021monitoring,
  author = {Joel Persson and Jurriaan F. Parie and Stefan Feuerriegel},
  title = {{Monitoring the COVID-19 epidemic with nationwide telecommunication data}},
  booktitle = {Proceedings of the National Academy of Sciences},
  year = {2021},
  doi = {10.1073/pnas.2100664118},
}
Figure