Stages of COVID-19 pandemic and paths to herd immunity by vaccination: dynamical model comparing Austria, Luxembourg and Sweden

Françoise Kemp, Daniele Proverbio, Atte Aalto, Laurent Mombaerts, Aymeric Fouquier d’Hérouël, Andreas Husch, Christophe Ley, Jorge Gonçalves, Alexander Skupin, Stefano MagnimedRxiv 2021 · DOI

This paper, published in early 2021 during the critical transition from lockdowns to the first vaccination campaigns, provides a rigorous framework for understanding how social behavior and immunization rates interact. By extending the classic SEIR model to include undetected cases and healthcare bottlenecks, the authors offer a roadmap for reaching herd immunity across three European nations with vastly different pandemic strategies.

TL;DR

Researchers developed an extended SEIR model to compare the pandemic trajectories of Luxembourg, Austria, and Sweden. The study quantifies how social distancing and vaccination must work in tandem, estimating that while vaccines are a game-changer, herd immunity was unlikely to be reached before mid-summer 2021 without maintaining significant social restrictions.

What Problem Does This Paper Address?

By late 2020, the world faced an "interregnum" period: vaccines were approved, but supply was low, and "pandemic fatigue" was threatening social distancing compliance. Public health officials needed to know:

  1. How many people were actually infected (including undetected cases)?
  2. How much have social interventions actually reduced transmission?
  3. What vaccination rate is required to finally "open up" without crashing the healthcare system?

The authors tackle these questions by comparing three distinct national strategies: Luxembourg (high testing), Austria (strict lockdowns), and Sweden (a more relaxed, "trust-based" approach).

Methods

The authors built an extended Susceptible-Exposed-Infectious-Removed (SEIR) model. While a standard SEIR model is relatively simple, this version is significantly more granular, featuring 16 distinct compartments.

The extended SEIR model structure, featuring 16 compartments to track the progression from exposure to recovery or death, including healthcare-specific paths.

Key innovations include:

  • Undetected vs. Detected Branches: The model splits the "Infectious" path. Detected cases move toward quarantine and hospitals, while undetected cases continue to spread the virus until recovery or death.
  • Healthcare Progression: It explicitly models the flow of patients through regular hospital wards, ICUs, and "longer stay" hospitalizations.
  • Dynamic Social Interaction ($\rho$): Instead of a fixed transmission rate, the authors introduced a time-dependent parameter, $\rho(t)$, which represents the fraction of pre-pandemic social interactions. This parameter was adjusted at specific dates corresponding to policy changes.
  • Bayesian Calibration: They used Markov Chain Monte Carlo (MCMC) methods to estimate parameters, allowing them to quantify uncertainty and account for "parameter degeneracy."

Datasets

The study utilized publicly available time-series data from February to December 15, 2020, for Luxembourg, Austria, and Sweden. The data points included total confirmed cases, daily new cases, hospital and ICU occupancy rates, and cumulative deaths. Prevalence data from antibody studies were used to calibrate the "undetected" fraction.

Key Insights and Ideas

  • The "Iceberg" of Infection: The model reveals that Sweden had the highest fraction of undetected cases, while Luxembourg—thanks to its "Large Scale Testing" program—had the highest fraction of its population actually infected (estimated at ~18% by Dec 2020).
  • Quantifying the "New Normal": In December 2020, the model estimated that social interactions across all three countries were roughly 1/3 of pre-pandemic levels. This provided a concrete metric for the "cost" of keeping the effective reproduction number ($R_{eff}$) near 1.
Estimated social interaction levels (left) and effective reproduction numbers (right) across Luxembourg, Austria, and Sweden.
  • The Vaccination-Social Synergy: A major finding is that vaccination does not allow for an immediate return to normal. If social measures were dropped too early, even a robust vaccination campaign would be overwhelmed by a massive "exit wave" of infections among the remaining susceptible population.

Key Results

  • Herd Immunity Thresholds: Based on estimated $R_0$ values (ranging from 3.16 in Austria to 4.0 in Sweden), the herd immunity threshold was calculated at roughly 68% to 75% of the population.
  • Required Vaccination Rates: To reach herd immunity by mid-April 2021, Luxembourg needed ~2,700 vaccinations/day, while the larger populations of Austria and Sweden required ~45,000/day.
  • The Timeline: If a country could vaccinate its entire population within one year, herd immunity would likely be reached by July (Luxembourg) or August (Austria/Sweden) 2021.
  • The "Pessimistic" Scenario: The model showed that if social interactions returned to October 2020 levels too quickly, ICU capacity would be breached long before vaccination could provide a safety net.

Limitations and Future Work

The model treats each country as a closed system, which is a limitation for regions with high cross-border mobility like Luxembourg. Additionally, the study assumes lasting immunity from infection or vaccination; future work would need to incorporate waning immunity or the emergence of new variants. Finally, while the model is granular in its healthcare paths, it is not age-stratified, which is a key factor in COVID-19 risk and vaccination prioritization.

Summary updated

Abstract

Abstract Background Worldwide more than 72 million people have been infected and 1.6 million died with SARS-CoV-2 by 15th December 2020. Non-pharmaceutical interventions which decrease social interaction have been implemented to reduce the spread of SARS-CoV-2 and to mitigate stress on healthcare systems and prevent deaths. The pandemic has been tackled with disparate strategies by distinct countries resulting in different epidemic dynamics. However, with vaccines now becoming available, the current urgent open question is how the interplay between vaccination strategies and social interaction will shape the pandemic in the next months. Methods To address this question, we developed an extended Susceptible-Exposed-Infectious-Removed (SEIR) model including social interaction, undetected cases and the progression of patients trough hospitals, intensive care units (ICUs) and death. We calibrated our model to data of Luxem-bourg, Austria and Sweden, until 15th December 2020. We incorporated the effect of vaccination to investigate under which conditions herd immunity would be achievable in 2021. Results The model reveals that Sweden has the highest fraction of undetected cases, Luxembourg displays the highest fraction of infected population, and all three countries are far from herd immunity as of December 2020. The model quantifies the level of social interactions, and allows to assess the level which would keep R eff ( t ) below 1. In December 2020, this level is around 1/3 of what it was before the pandemic for all the three countries. The model allows to estimate the vaccination rate needed for herd immunity and shows that 2700 vaccinations/day are needed in Luxembourg to reach it by mid of April and 45,000 for Austria and Sweden. The model estimates that vaccinating the whole country’s population within 1 year could lead to herd immunity by July in Luxembourg and by August in Austria and Sweden. Conclusion The model allows to shed light on the dynamics of the epidemics in different waves and countries. Our results emphasize that vaccination will help considerably but not immediately and therefore social measures will remain important for several months before they can be fully alleviated.
BibTeX
@misc{kemp2021stages,
  author = {Françoise Kemp and Daniele Proverbio and Atte Aalto and Laurent Mombaerts and Aymeric Fouquier d’Hérouël and Andreas Husch and Christophe Ley and Jorge Gonçalves and Alexander Skupin and Stefano Magni},
  title = {{Stages of COVID-19 pandemic and paths to herd immunity by vaccination: dynamical model comparing Austria, Luxembourg and Sweden}},
  year = {2021},
  doi = {10.1101/2020.12.31.20249088},
  howpublished = {medRxiv},
}
Figure