Quantifying the importance and location of SARS-CoV-2 transmission events in large metropolitan areas

Alberto Aleta, David Martín-Corral, Michiel A. Bakker, Ana Pastore y Piontti, Marco Ajelli, Maria Litvinova, Matteo Chinazzi, Natalie E. Dean, M. Elizabeth Halloran, Ira M. Longini, Alex Pentland, Alessandro Vespignani, Yamir Moreno, Esteban MoroProceedings of the National Academy of Sciences 2022 · DOI

TL;DR

This paper uses privacy-enhanced mobility data and an agent-based model to pinpoint where and when SARS-CoV-2 transmission occurred during the first wave in NYC and Seattle. It reveals that a small percentage of individuals and smaller, everyday settings (like workplaces and grocery stores) were disproportionately responsible for infections, not just large gatherings.

What Problem Does This Paper Address?

Understanding the precise dynamics of SARS-CoV-2 transmission is crucial for designing effective, yet minimally disruptive, public health interventions. While non-pharmaceutical interventions (NPIs) like lockdowns and social distancing have been implemented, their impact is often assessed by broad epidemiological metrics (e.g., reduction in case growth rate). There's a need for a more granular understanding of where and when transmission events occur, especially considering the phenomenon of superspreading events (SSEs), where a small number of individuals or locations are responsible for a large proportion of infections. This paper aims to fill this gap by modeling transmission at a fine-grained level, integrating mobility data with an agent-based approach to understand the role of different settings and individual behavior.

Methods

The researchers developed a detailed agent-based model (ABM) that simulates SARS-CoV-2 infection spread within synthetic populations representing the New York City (NYC) and Seattle metropolitan areas. The core of their methodology lies in integrating anonymized, time-resolved mobility data from millions of individuals. This data was used to construct a weighted, multilayered temporal network. This network comprises four layers: households, schools (static), and dynamically changing workplace and community layers.

The community layer was further detailed by incorporating location data from public APIs (Foursquare) to represent various points of interest (POIs). Transmission probabilities between individuals were estimated based on their co-location and the duration of their stays in these settings, moving beyond simple homogeneous mixing assumptions. The model incorporates a stochastic, discrete-time compartmental model to simulate individual infection progression (susceptible, latent, infectious symptomatic/asymptomatic, recovered, deceased). Crucially, the model was calibrated against weekly death data to estimate the effective reproduction number ($R_t$). They also conducted counterfactual simulations to assess the impact of NPI timing.

Datasets

  • Mobility Data: Privacy-enhanced, time-resolved mobility data from over 0.5 million individuals in the NYC and Seattle metropolitan areas (February-June 2020), collected through a GDPR-compliant framework. This data represents anonymized user location data.
  • Location Data (Points of Interest): Data from a large database of 375,000 locations in NYC and 70,000 locations in Seattle obtained from the Foursquare public API.
  • Demographic Data: Census and demographic data for the NYC and Seattle metropolitan areas.
  • Epidemiological Data: Reported surveillance data on weekly COVID-19 deaths for NYC and Seattle, used for model calibration. CDC reported prevalence data was also used for comparison.

Key Insights and Ideas

  • The "18/80 Rule" of Transmission: A small fraction of individuals (around 18%) are responsible for a disproportionately large share of infections (around 80%). This highlights the significant impact of individual-level heterogeneity in transmission.
  • Small Settings Matter Most: Contrary to the focus on large gatherings, the bulk of SARS-CoV-2 transmission occurred in smaller, everyday settings like workplaces, grocery stores, and food venues. This suggests that interventions targeting these common locations could be highly effective.
  • Dynamic Risk Landscape: The "dangerous" places for transmission are not static. The importance of different venues shifts over time, influenced by both the pandemic's progression and the implementation of NPIs. This underscores the critical role of behavioral adaptation.
  • Place + Behavior = Risk: A location's risk is not inherent but a complex interplay between the characteristics of the setting itself and the behavior of the individuals who frequent it. This challenges the idea of universally "high-risk" venues.
  • NPI Timing is Critical: Even small shifts in the timing of NPI implementation can have a substantial impact on the trajectory of the epidemic, potentially doubling peak deaths if delayed.

Key Results

  • Superspreading Events (SSEs): Approximately 10% of transmission events were estimated to be SSEs. While mass gatherings are a risk factor, the majority of SSEs occurred in smaller settings.
  • Impact of NPIs: The introduction of NPIs led to a dramatic decrease in contacts, with $R_t$ dropping below one within a week in both NYC and Seattle. A one-week delay in NPI implementation in NYC could have doubled the peak number of deaths.
  • Setting-Specific Transmission: Before NPIs, infections were prevalent in community and workplace layers. After NPIs, the household layer saw an increase, especially in NYC. In the community layer, food/beverage and shopping venues were key transmission sites before NPIs, with grocery stores becoming more prominent after restrictions.
  • Heterogeneous Infection Distribution: The distribution of secondary infections was highly skewed, fitting a negative binomial distribution with dispersion parameters ($k$) of 0.16 for NYC and 0.23 for Seattle. This confirms significant overdispersion.

Limitations and Future Work

  • Data Granularity: While the mobility data is detailed, it's an indirect proxy for actual transmission. The model doesn't capture the nuances of specific contact types (e.g., masked vs. unmasked, conversational vs. brief).
  • Outdoor vs. Indoor Risk: The model assumes a reduced transmission probability outdoors, based on existing evidence, but further research quantifying this difference is warranted.
  • Behavioral Nuances: The model doesn't differentiate behavior within contacts (e.g., friends vs. strangers), which could influence transmission.
  • Exclusion of Nursing Homes: Nursing homes, a significant source of severe outcomes, were not explicitly modeled, though their impact on population-level transmission dynamics is acknowledged as potentially less central than other settings.
  • Future Work: The authors suggest that real-time tracking of transmission events could be instrumental for evaluating and defining targeted mitigation policies. Their methodology could be extended to other locations and future pandemics. The findings also suggest optimizing targeted measures, such as extending operating hours for certain venues or promoting smart working.

Summary updated

Abstract

Detailed characterization of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) transmission across different settings can help design less disruptive interventions. We used real-time, privacy-enhanced mobility data in the New York City, NY and Seattle, WA metropolitan areas to build a detailed agent-based model of SARS-CoV-2 infection to estimate the where, when, and magnitude of transmission events during the pandemic's first wave. We estimate that only 18% of individuals produce most infections (80%), with about 10% of events that can be considered superspreading events (SSEs). Although mass gatherings present an important risk for SSEs, we estimate that the bulk of transmission occurred in smaller events in settings like workplaces, grocery stores, or food venues. The places most important for transmission change during the pandemic and are different across cities, signaling the large underlying behavioral component underneath them. Our modeling complements case studies and epidemiological data and indicates that real-time tracking of transmission events could help evaluate and define targeted mitigation policies.
BibTeX
@inproceedings{aleta2022quantifying,
  author = {Alberto Aleta and David Martín-Corral and Michiel A. Bakker and Ana Pastore y Piontti and Marco Ajelli and Maria Litvinova and Matteo Chinazzi and Natalie E. Dean and M. Elizabeth Halloran and Ira M. Longini and Alex Pentland and Alessandro Vespignani and Yamir Moreno and Esteban Moro},
  title = {{Quantifying the importance and location of SARS-CoV-2 transmission events in large metropolitan areas}},
  booktitle = {Proceedings of the National Academy of Sciences},
  year = {2022},
  doi = {10.1073/pnas.2112182119},
}
Figure