Monitoring real-time transmission heterogeneity from incidence data

Yunjun Zhang, Tom Britton, Xiaohua ZhouPLOS Computational Biology 2022 · DOI

TL;DR

This paper introduces a novel, flexible, and generic model to estimate real-time transmission heterogeneity of infectious diseases using only incidence data. This approach overcomes limitations of existing methods that require complex, often unavailable, data and fail to capture temporal changes, offering a more accessible and dynamic way to monitor epidemic spread.

What Problem Does This Paper Address?

Understanding how infectious diseases spread is crucial for effective public health interventions. A key aspect of this spread is "transmission heterogeneity," meaning that some infected individuals infect far more people than others (superspreading events). This heterogeneity can significantly impact epidemic dynamics, making control measures less efficient and prolonging outbreaks. While methods exist to identify this heterogeneity, they often rely on detailed data like contact tracing, viral genomes, or spatial information, which are not always available or easy to generalize. Furthermore, most existing methods assume a constant level of heterogeneity throughout an epidemic, failing to capture its dynamic, real-time changes. This paper addresses the need for a simpler, more accessible, and dynamic method to monitor transmission heterogeneity using readily available incidence time series data.

Methods

The core of this paper is the development of the "Instant-individual heterogeneity" (IIH) model. This model extends the standard renewal process framework for epidemic modeling by introducing the concept of "instant-individual reproduction number" (IIRN). The IIRN accounts for the variation in infectiousness both between different individuals and at different points in time.

Key methodological aspects include:

  • Instant-Individual Reproduction Number (IIRN): This is the central innovation, representing the expected number of secondary cases generated at time t by an individual infected at time s. It's modeled as a random variable, allowing for individual-specific and time-varying infectiousness.
  • Gamma Distribution for Heterogeneity: The IIRN is assumed to follow a Gamma distribution, with the parameter k (dispersion number) directly quantifying transmission heterogeneity. A smaller k indicates higher heterogeneity.
  • Joint Likelihood Function: The model derives a joint likelihood function for both the reproduction number (Rₜ) and the dispersion number (kₜ), allowing for simultaneous estimation of these key epidemiological parameters from incidence data.
  • Time-Windowed Estimation: To capture real-time changes, the model uses a sliding time window to estimate Rₜ and kₜ, providing a temporal trend of transmission dynamics.
  • Robustness to Biases: The authors specifically test and demonstrate the model's robustness to common issues like underreporting and misspecification of the serial interval.

Datasets

The paper utilizes several real-world epidemic datasets for validation:

  • Static Scenario Datasets:
    • COVID-19 in Hong Kong, China (2020)
    • COVID-19 in Tianjin, China (2020)
    • MERS in South Korea (2015)
    • Measles in Canada (2019)
  • Time-Varying Scenario Datasets:
    • Ebola epidemic in Freetown, Sierra Leone (2014-2015)
    • COVID-19 in Georgia, USA (2020)
    • COVID-19 in Hong Kong, China (2020-2021)
  • Omicron Variant Data:
    • COVID-19 in South Africa (2021-2022)

The paper also uses simulated data generated under various conditions of reproduction number, dispersion number, and reporting rates to rigorously evaluate the model's performance.

Key Insights and Ideas

  • Incidence Data is Sufficient: The most significant insight is that detailed contact tracing or genomic data isn't always necessary to estimate transmission heterogeneity. Incidence time series alone can provide valuable insights.
  • Heterogeneity is Dynamic: The model's ability to estimate real-time heterogeneity (kₜ) reveals that it's not static but changes over the course of an epidemic, which is critical for understanding epidemic evolution.
  • Omicron's Potential for Superspreading: The analysis of Omicron in South Africa suggests it might exhibit greater transmission heterogeneity than Delta, potentially contributing to its rapid spread.
  • Model Robustness is Key: The demonstrated resilience of the IIH model to underreporting and serial interval misspecification makes it a practical tool for real-world epidemiological data, which is often imperfect.
  • Distinguishing from Simpler Models: The paper clearly shows how the IIH model can reveal significant heterogeneity where simpler "instant-level heterogeneity" models fail, often overestimating k and thus underestimating heterogeneity.

Key Results

  • Accurate Estimation: The IIH model accurately estimated both reproduction number (R) and dispersion number (k) in simulations across various scenarios, with accuracy improving with longer time windows of data.
  • Real-World Consistency: Estimates for historical epidemics (MERS, Ebola, COVID-19, Measles) were consistent with findings from studies using contact tracing or spatial data, validating the model's applicability.
  • Omicron Heterogeneity: The study found that the Omicron variant in South Africa had a lower dispersion number (k) than the Delta variant during the same period, indicating potentially higher superspreading potential.
  • Sensitivity Analysis: The model showed robustness to underreporting (down to 60% missing cases) and was less affected by serial interval standard deviation than its mean.
  • Time-Varying Trends: For epidemics like Ebola and COVID-19 in Georgia, the model revealed significant temporal trends in heterogeneity, with k often falling below 1, indicating substantial superspreading.

Limitations and Future Work

  • Spatial Heterogeneity: The current model aggregates data over a study area and thus overlooks spatial heterogeneity within local scales. Future work could incorporate this by considering population distribution and social network structures.
  • Data Delay: The estimation of real-time heterogeneity is inherently delayed due to the time it takes for infections to be confirmed and reported. Future work could address this by reconstructing infection curves from incidence data or optimizing data accumulation window lengths.
  • Distributional Assumptions: The model uses a Gamma distribution to characterize heterogeneity. However, for scenarios like ongoing vaccination, a bimodal distribution might be more appropriate. Future work could explore alternative distributions.
  • Prior Information: The study used "non-informative" priors for parameter estimation. Incorporating more informative priors could potentially improve estimation accuracy.
  • Model Complexity: While the IIH model is simpler than some alternatives, extending it to more complex transmission models (e.g., SIR models) and exploring real-time heterogeneity within those frameworks is a future direction.

Summary updated

BibTeX
@article{zhang2022monitoring,
  author = {Yunjun Zhang and Tom Britton and Xiaohua Zhou},
  title = {{Monitoring real-time transmission heterogeneity from incidence data}},
  journal = {PLOS Computational Biology},
  year = {2022},
  doi = {10.1371/journal.pcbi.1010078},
}
Figure