Data-driven mechanistic framework with stratified immunity and effective transmissibility for COVID-19 scenario projections

Przemyslaw Porebski, Srinivasan Venkatramanan, Aniruddha Adiga, Brian Klahn, Benjamin Hurt, Mandy Wilson, Jiangzhuo Chen, Anil Vullikanti, Madhav Marathe, Bryan LewisEpidemics 2024 · DOI

TL;DR

This paper introduces UVA-adaptive, a flexible, data-driven mechanistic modeling framework designed for real-time COVID-19 scenario projections. It integrates real-world data on immunity, variants, and vaccination to provide adaptive, high-resolution forecasts, crucial for informing public health policy during a dynamic pandemic.

What Problem Does This Paper Address?

During the COVID-19 pandemic, policymakers at state and federal levels relied heavily on modeling to guide responses. While many custom models exist, they often struggle to keep pace with the rapidly evolving pandemic landscape – changing variants, vaccine uptake, and public behavior. Sustained, accurate projections require a framework that is not only robust but also adaptive, capable of integrating new data and evolving complexities like stratified immunity and multiple strains. This paper addresses the need for such an integrated and adaptive modeling system.

Methods

The core of UVA-adaptive is built upon the PatchSim metapopulation SEIR (Susceptible-Exposed-Infectious-Recovered) model. The key innovation lies in its data-driven, mechanistic approach. Instead of relying solely on fixed parameters, UVA-adaptive uses a calibrated, time-varying effective transmissibility parameter, β(t), as the primary mechanism to capture the combined effects of interventions, behavioral changes, and evolving pathogen characteristics.

This framework incorporates:

  • Stratified Immunity: Accounts for varying levels of immunity based on vaccination status (number of doses) and prior infection history.
  • Multi-Strain Dynamics: Extends the model to handle multiple SARS-CoV-2 variants, incorporating their immune escape properties.
  • Adaptive Trajectory Fitting: A data-driven pipeline continuously updates projections using real-time data.
  • Scenario Definition via Transmissibility: Scenarios are defined by adjusting β(t) and other parameters, allowing exploration of "what-if" situations.
  • Post-processing for Outcomes: Simulated infections are translated into reported cases, hospitalizations, and deaths using estimated rates and time lags.

Datasets

The model utilizes a variety of real-time datasets for calibration and projection:

  • Seroprevalence Surveys (CDC): Used to estimate case-ascertainment rates.
  • Home Test Surveys (e.g., Rader et al., 2022): Incorporated to adjust case-ascertainment rates, especially with the rise of at-home testing.
  • Variant Prevalence Data (GISAID, outbreak.info, CDC): Used to model the spread and impact of different SARS-CoV-2 strains.
  • Vaccination Data: Tracks uptake and effectiveness.
  • Reported Cases, Hospitalizations, and Deaths: The primary outcome data for calibration and evaluation.
  • Mobility Reports (Google LLC): Used for correlation analysis with transmissibility.
  • Mask-wearing Trends (Delphi Group at Carnegie Mellon University): Used for correlation analysis.
  • Weather Data (NOAA): Used for correlation analysis.

Key Insights and Ideas

  • Transmissibility as a "Catch-All" Parameter: The paper highlights the power of using a single, time-varying transmissibility parameter (β(t)) to implicitly capture a multitude of complex, unmodeled factors like behavioral changes, NPI effectiveness, and even some aspects of variant evolution. This simplifies the model while retaining significant predictive power.
  • Synergy of Mechanistic and Data-Driven Approaches: UVA-adaptive demonstrates that combining the explainability of mechanistic models with the adaptability of data-driven methods creates a powerful tool for dynamic forecasting.
  • Importance of Stratified Immunity: The model's ability to differentiate immunity based on vaccination and infection history is crucial for accurately projecting future waves, especially as new variants emerge with varying immune escape properties.
  • Correlation with Societal Signals: The finding that calibrated transmissibility correlates with mobility and mask-wearing data suggests that these societal factors are indeed strong drivers of epidemic dynamics, and the model implicitly captures their influence.

Key Results

The UVA-adaptive model was successfully used for regular projections for the Virginia Department of Health (VDH) and the US COVID-19 Scenario Modeling Hub (SMH) across multiple rounds.

  • The calibrated transmissibility parameter β(t) was shown to correlate with the evolution of the pathogen and associated societal dynamics.
  • The model's ability to capture the immunity profile of the population was demonstrated, with comparisons to breakthrough case data showing reasonable agreement, though some discrepancies were noted for boosted populations during the Omicron wave.
  • The framework allowed for scenario-based projections, including counterfactual analyses of interventions and variant emergence.

Limitations and Future Work

The authors acknowledge several limitations:

  • Lack of Explicit Age Stratification: The model does not explicitly stratify by age, which limits its ability to capture age-specific transmission dynamics and the differential impact of NPIs on various age groups. Future work could incorporate age-specific contact matrices.
  • Limited Connectivity Between Regions: Inter-regional transmissions, especially for novel variants, are primarily handled through external seeding, which might not fully capture complex spatial spread.
  • Behavioral Effects Modeled Indirectly: Behavioral changes (like NPIs or adaptive behavior) are not explicitly modeled but are incorporated as changes in β(t). For projections where past data is insufficient, external evaluation might be needed.
  • Parameter Estimation Dependence: The calibration procedure relies heavily on external data sources and estimations for parameters like immune escape and strain-specific vaccine efficacies, which were not always updated dynamically.
  • Uncertainty Estimation: While factorial design is used to explore parameter uncertainty, it doesn't include uncertainty from parameters not included in the design (e.g., waning immunity).

Future work could involve developing more explicit age-structured models, improving inter-regional transmission modeling, and refining the estimation of parameters related to vaccine efficacy and waning immunity.

Summary updated

Abstract

Scenario-based modeling frameworks have been widely used to support policy-making at state and federal levels in the United States during the COVID-19 response. While custom-built models can be used to support one-off studies, sustained updates to projections under changing pandemic conditions requires a robust, integrated, and adaptive framework. In this paper, we describe one such framework, UVA-adaptive, that was built to support the CDC-aligned Scenario Modeling Hub (SMH) across multiple rounds, as well as weekly/biweekly projections to Virginia Department of Health (VDH) and US Department of Defense during the COVID-19 response. Building upon an existing metapopulation framework, PatchSim, UVA-adaptive uses a calibration mechanism relying on adjustable effective transmissibility as a basis for scenario definition while also incorporating real-time datasets on case incidence, seroprevalence, variant characteristics, and vaccine uptake. Through the pandemic, our framework evolved by incorporating available data sources and was extended to capture complexities of multiple strains and heterogeneous immunity of the population. Here we present the version of the model that was used for the recent projections for SMH and VDH, describe the calibration and projection framework, and demonstrate that the calibrated transmissibility correlates with the evolution of the pathogen as well as associated societal dynamics.
BibTeX
@inproceedings{porebski2024data,
  author = {Przemyslaw Porebski and Srinivasan Venkatramanan and Aniruddha Adiga and Brian Klahn and Benjamin Hurt and Mandy Wilson and Jiangzhuo Chen and Anil Vullikanti and Madhav Marathe and Bryan Lewis},
  title = {{Data-driven mechanistic framework with stratified immunity and effective transmissibility for COVID-19 scenario projections}},
  booktitle = {Epidemics},
  year = {2024},
  doi = {10.1016/j.epidem.2024.100761},
}
Figure