On the use of multiple compartment epidemiological models to describe the dynamics of influenza in Europe

Inbar Seroussi, Nir Levy, Daniela Paolotti, Nir Sochen, Elad Yom‐TovarXiv 2019 · DOI · arXiv

Abstract

We develop a multiple compartment Susceptible-Infected-Recovered (SIR) model to analyze the spread of several infectious diseases through different geographic areas. Additionally, we propose a data-quality sensitive optimization framework for fitting this model to observed data. We fit the model to the temporal profile of the number of people infected by one of six influenza strains in Europe over 7 influenza seasons. In addition to describing the temporal and spatial spread of influenza, the model provides an estimate of the inter-country and intra-country infection and recovery rates of each strain and in each season. We find that disease parameters remain relatively stable, with a correlation greater than 0.5 over seasons and strains. Clustering of influenza strains by the inferred disease parameters is consistent with genome sub-types. Surprisingly, our analysis suggests that inter-country human mobility plays a negligible role in the spread of influenza in Europe. Finally, we show that the model allows the estimation of disease load in countries with poor or non-existent data from the disease load in adjacent countries. Our findings reveal information on the spreading mechanism of influenza and on disease parameters. These can be used to assist in disease surveillance and in control of influenza as well as of other infectious pathogens in a heterogenic environment.

Introduction

Epidemiological modeling often faces a "Goldilocks" problem. Simple models, like the classic Susceptible-Infected-Recovered (SIR) framework, are computationally efficient but fail to capture the complexity of multiple viral strains moving across interconnected borders. Conversely, highly complex models often break down when fed real-world surveillance data, which is notoriously noisy, incomplete, or inconsistent across different countries. This study addresses two main gaps: modeling heterogeneity (multiple strains and regions) and managing data quality without introducing manual filtering bias.

Methodology: The mcSIR Model and Lagrangian Optimization

The researchers introduced the multi-compartment SIR (mcSIR) model. Unlike a standard SIR model that uses scalar values for infection ($\beta$) and recovery ($\gamma$) rates, the mcSIR model utilizes tensors. This allows the representation of a multidimensional system where different infectious sub-populations (countries) interact under the influence of multiple viral strains.

To fit this model to real-world data, the authors proposed a Lagrangian Optimization Framework. This variational technique treats the epidemic's evolution as a physical system seeking a "stationary solution." Crucially, the framework includes Lagrange multipliers ($\lambda_i$) for each country. During optimization, the algorithm automatically assigns lower weights to countries with noisy or missing data. This data-quality sensitive approach allows the model to utilize all available information while naturally muting the influence of unreliable sources.

Datasets

The study validated the model using two primary data types:

  • WHO Surveillance Data: Weekly reports of six influenza strains (including H1N1 and H3N2) across 24 European countries from 2012 to 2018.
  • Human Mobility Data: To test the impact of travel, the authors used air and train travel statistics from Eurostat, alongside a proxy for total movement derived from 50 million GPS-tagged Twitter messages across Europe.

Key Results and Performance

The mcSIR model demonstrated significant improvements over the standard SIR model. It achieved an average Pearson correlation of $r = 0.57$ with WHO data across all seasons and strains. The Lagrangian framework successfully identified and de-prioritized noisy data, with a Spearman correlation of $\rho = 0.52$ between the learned weights and the quality of the fit.

Comparison of Pearson correlation distributions for the standard SIR model versus the proposed mcSIR model.

Furthermore, disease parameters remained relatively stable over time, showing a correlation greater than 0.5 across different seasons. This stability suggests that these models possess strong predictive power for future outbreaks.

Insights into Viral Dynamics

The Mobility Surprise

In previous studies of the US, human mobility (air travel and commuting) explained a significant portion of disease spread. However, in Europe, the authors found no significant correlation between mobility patterns and the spread of influenza. This suggests that European flu dynamics might be driven more by local reservoirs or synchronized environmental triggers rather than travelers carrying the virus across borders.

Biological Consistency

When the model inferred disease parameters, the authors clustered the strains based on these values. The resulting clusters matched known genomic subtypes (Type A vs. Type B), providing a biological "sanity check" for the model's accuracy.

Dendrogram showing the clustering of influenza strains based on inferred disease parameters for the 2013 season.

Predicting the "Invisible"

Because the model accounts for spatial connectivity, it can perform "data completion." The authors successfully predicted the infection curve for the UK using only data from its neighbors and historical parameters, achieving a 0.96 correlation with the actual WHO data.

Conclusion and Future Work

While robust, the model currently assumes strains do not interact (ignoring cross-reactivity) and does not account for demographic variations or vaccination rates. Future research will focus on incorporating these factors and utilizing finer-grained spatial data to further investigate the role of mobility in disease transmission.

Summary updated

Abstract

We develop a multiple compartment Susceptible-Infected-Recovered (SIR) model to analyze the spread of several infectious diseases through different geographic areas. Additionally, we propose a data-quality sensitive optimization framework for fitting this model to observed data. We fit the model to the temporal profile of the number of people infected by one of six influenza strains in Europe over $7$ influenza seasons. In addition to describing the temporal and spatial spread of influenza, the model provides an estimate of the inter-country and intra-country infection and recovery rates of each strain and in each season. We find that disease parameters remain relatively stable, with a correlation greater than $0.5$ over seasons and stains. Clustering of influenza strains by the inferred disease parameters is consistent with genome sub-types. Surprisingly, our analysis suggests that inter-country human mobility plays a negligible role in the spread of influenza in Europe. Finally, we show that the model allows the estimation of disease load in countries with poor or none existent data from the disease load in adjacent countries. Our findings reveal information on the spreading mechanism of influenza and on disease parameters. These can be used to assist in disease surveillance and in control of influenza as well as of other infectious pathogens in a heterogenic environment.
BibTeX
@misc{seroussi2019use,
  author = {Inbar Seroussi and Nir Levy and Daniela Paolotti and Nir Sochen and Elad Yom‐Tov},
  title = {{On the use of multiple compartment epidemiological models to describe the dynamics of influenza in Europe}},
  year = {2019},
  eprint = {1906.08631},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1906.08631},
}
Figure