Predicting the SARS-CoV-2 effective reproduction number using bulk contact data from mobile phones

Sten Rüdiger, Stefan Konigorski, Alexander Rakowski, Jonathan Antonio Edelman, Detlef Zernick, Alexander Thieme, Christoph LippertProceedings of the National Academy of Sciences 2021 · DOI

Abstract

Traditional methods for estimating the effective reproduction number ($R$) of SARS-CoV-2 rely on clinical case reports, which suffer from significant reporting lags. This study introduces the "Contact Index" (CI), a metric derived from de-identified GPS data of approximately 1.3 million mobile phones in Germany. By applying graph sampling theory, the researchers demonstrate that the CI—which accounts for the heterogeneity of human interactions and "superspreading" potential—is a highly accurate predictor of $R$ with a 17-day lead time. This metric provides a proactive tool for public health interventions, identifying behavioral shifts before they manifest as infection surges.

Introduction

Managing a pandemic requires timely data. The effective reproduction number ($R$) is the gold standard for tracking viral spread, but it is inherently retrospective. The delay between infection, symptom onset, testing, and official reporting means that by the time $R$ rises, the window for early intervention has often closed. Furthermore, simple mobility metrics or average contact counts often fail to capture the "superspreading" events that drive exponential growth. This research proposes the Contact Index (CI) as a real-time behavioral proxy that captures the network structure of human interactions.

Methodology: From GPS to Contact Networks

The study analyzed GPS tracking data from a representative panel of 1.15 to 1.4 million devices across Germany between March and November 2020.

To define a "contact," the researchers discretized the geographic area into 8m x 8m tiles. A contact event was recorded when two distinct devices occupied the same tile within the same two-minute window.

Schematic of contact definition (A) and the resulting population interaction network (B).

Unlike traditional models that focus on the average number of contacts per person (mean degree), this approach utilized graph theory to calculate the Contact Index (CI). The CI is based on the ratio of the second moment to the first moment of the degree distribution. This mathematical formulation gives significantly more weight to "hubs"—individuals with an unusually high number of contacts—thereby quantifying the potential for superspreading within the population network.

Key Results: Predictive Power and Lead Time

The researchers compared the CI against official $R$ values provided by the Robert Koch Institute (RKI). A primary finding was the strong temporal correlation between behavioral changes and subsequent shifts in $R$.

Comparison of Contact Index (CI) and reproduction number (R) over time (A, B) and their correlation (C).

The 17-Day Lead Time

The CI demonstrated a maximum Pearson correlation coefficient of 0.82 (95% CI: 0.78–0.86) when shifted forward by 17 days. This lag accounts for the biological incubation period (approx. 5 days) and the administrative delays in the testing and reporting pipeline (approx. 12 days). This 17-day window offers policymakers a critical early-warning system.

CI vs. Mean Contacts

A crucial insight was that the CI remained a better predictor than the simple mean number of contacts. Following the first lockdown, while the average number of contacts recovered to 75% of pre-pandemic levels, $R$ remained low. The CI correctly reflected this because, although people were moving more, they were not re-forming the large, dense clusters (hubs) seen earlier in the year.

The Critical Threshold

The study identified a critical threshold value ($CI_{crit} \approx 38$). When social behavior pushed the CI above this level, $R$ was statistically likely to exceed 1.0, signaling the onset of exponential growth. Interestingly, the introduction of mandatory masks did not fundamentally alter the linear relationship between CI and $R$, though it may have shifted the absolute risk levels.

Discussion and Limitations

While the CI is a powerful predictive tool, it has limitations. GPS data at an 8m resolution is a proxy for physical proximity and cannot distinguish between indoor and outdoor settings or the presence of physical barriers like walls. Additionally, the CI primarily measures internal connectivity and may not account for $R$ increases driven by imported cases from international travel.

Despite these limitations, the Contact Index provides a robust, real-time dashboard for monitoring pandemic-relevant social behavior. By focusing on the network structure of contacts rather than just volume, it allows for more nuanced and timely public health decisions.

Summary updated

Abstract

Over the last months, cases of SARS-CoV-2 surged repeatedly in many countries but could often be controlled with nonpharmaceutical interventions including social distancing. We analyzed deidentified Global Positioning System (GPS) tracking data from 1.15 to 1.4 million cell phones in Germany per day between March and November 2020 to identify encounters between individuals and statistically evaluate contact behavior. Using graph sampling theory, we estimated the contact index (CX), a metric for number and heterogeneity of contacts. We found that CX, and not the total number of contacts, is an accurate predictor for the effective reproduction number R derived from case numbers. A high correlation between CX and R recorded more than 2 wk later allows assessment of social behavior well before changes in case numbers become detectable. By construction, the CX quantifies the role of superspreading and permits assigning risks to specific contact behavior. We provide a critical CX value beyond which R is expected to rise above 1 and propose to use that value to leverage the social-distancing interventions for the coming months.
BibTeX
@inproceedings{rudiger2021predicting,
  author = {Sten Rüdiger and Stefan Konigorski and Alexander Rakowski and Jonathan Antonio Edelman and Detlef Zernick and Alexander Thieme and Christoph Lippert},
  title = {{Predicting the SARS-CoV-2 effective reproduction number using bulk contact data from mobile phones}},
  booktitle = {Proceedings of the National Academy of Sciences},
  year = {2021},
  doi = {10.1073/pnas.2026731118},
}
Figure