Leveraging Real-Time Population-Scale GPS Data to Forecast and Identify Components of Epidemic Dynamics
Introduction: Large-scale GPS data (from mobile phones) provides detailed insights into mobility and contact patterns (proximity of devices) in the population. Sensing real-time contact patterns is a representation of infection dynamics and therefore they are key to forecasting infection rates of SA...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-03-01
|
| Series: | International Journal of Infectious Diseases |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1201971224007264 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Introduction: Large-scale GPS data (from mobile phones) provides detailed insights into mobility and contact patterns (proximity of devices) in the population. Sensing real-time contact patterns is a representation of infection dynamics and therefore they are key to forecasting infection rates of SARS-CoV-2—in contrast to the retrospective nature of infection surveillance. This study leverages Bayesian mixed media modeling to provide probabilistic forecasts of epidemic trends. Our model emphasizes the significance of contact duration—a key driver in the probability of infection. The aim is to distinguish between contact-driven and variant-driven epidemics to facilitate more targeted interventions. Methods: We utilize a consistent population-scale panel of GPS location data donors from the German population. As in [1], we break down R_t (dynamic reproduction rate) into contact-driven and variant-driven components. Our method improves upon this by including contact duration: we categorize contacts as either stable (home/work) or random, and assess their impact on R_t. Employing the Bayesian Mixed Media Model and its carryover effect [2], we model the time lag between contact and reporting to accurately forecast R_t. Results: The GPS data from around 1% of the German population since 2019, with 70000000 samples per day, yield accurate insights into contact patterns. We quantified the delay between real-time contact and the R_t at 15 ± 1 days, therefore we forecast R_t ∼2 weeks into the future, with the prognosis demonstrating a stronger correlation between the contact-driven component during 2020. Additionally, the variant-driven component accurately captures changes in the transmissibility of SARS-CoV-2.Our analysis categorizes the types of contacts—such as stable/recurrent and random/one-time contacts. Despite stable contacts constituting less than 1% of all contacts, they account for approximately 50% of the total contact duration. Random contacts, albeit representing 99% of all contacts, contribute to the remaining 50% of the total contact duration and are responsible for trends in R_t. Discussion: While GPS data is not easily available, it significantly enhances our understanding of real-time epidemic trends compared to retrospective methods such as infection surveillance and surveys. We showed that transmissibility decomposed from R_t exhibited notable increases with the emergence of alpha, delta, and omicron variants, and subsequent declines following mass vaccination efforts, highlighting the impact of both viral evolution and increased population immunity on the pandemic's dynamics. Conclusion: Large-scale GPS data provides detailed insights into contact patterns (i.e. behavioural aspects), it offers a robust framework for predicting and managing epidemic trajectories. This methodology quantifies delayed effects of contacts on R_t and shows how random contacts provide clearer insights into R_t trends. It allows us to explore the effect of viral transmissibility without relying on molecular biology, enhancing our ability to monitor public health interventions. [1] Schulz, Steven, et al. https://doi.org/10.1101/2023.03.02.23286502 [2] Jin, Yuxue, et al. https://research.google/pubs/bayesian-methods-for-media-mix-modeling-with-carryover-and-shape-effects/ |
|---|---|
| ISSN: | 1201-9712 |