Projecting social contact matrices to different demographic structures.

Sergio Arregui,Joaquín Sanz,Alberto Aleta,Yamir Moreno,Arne Traulsen

doi:10.1371/journal.pcbi.1006638

Abstract

The modeling of large-scale communicable epidemics has greatly benefited in the last years from the increasing availability of highly detailed data. Particullarly, in order to achieve quantitative descriptions of the evolution of epidemics, contact networks and mixing patterns are key. These heterogeneous patterns depend on several factors such as location, socioeconomic conditions, time, and age. This last factor has been shown to encapsulate a large fraction of the observed inter-individual variation in contact patterns, an observation validated by different measurements of age-dependent contact matrices. Recently, several works have studied how to project those matrices to areas where empirical data are not available. However, the dependence of contact matrices on demographic structures and their time evolution has been largely neglected. In this work, we tackle the problem of how to transform an empirical contact matrix that has been obtained for a given demographic structure into a different contact matrix that is compatible with a different demography. The methodology discussed here allows to extrapolate a contact structure measured in a particular area to any other whose demographic structure is known, as well as to obtain the time evolution of contact matrices as a function of the demographic dynamics of the populations they refer to. To quantify the effect of considering time-dynamics of contact patterns on disease modeling, we implemented a Susceptible-Exposed-Infected-Recovered (SEIR) model on 16 different countries and regions and evaluated the impact of neglecting the temporal evolution of mixing patterns. Our results show that simulated disease incidence rates, both at the aggregated and age-specific levels, are significantly dependent on contact structures variation driven by demographic evolution. The present work opens the path to eliminate technical biases from model-based impact evaluations of future epidemic threats and warns against the use of contact matrices to model diseases without correcting for demographic evolution or geographic variations.

Highlights

During recent years, models on disease transmission have improved in complexity and depth, integrating high-resolution data on demography, mobility and social behavior [1, 2]
We evaluate the impact of an eventual future outbreak by simulating a SEIR scenario in the countries and regions analyzed
Our results show that using outdated or imported contact matrices that do not take into account demographic structure or its evolution can lead to largely misleading conclusions

Summary

Introduction

Models on disease transmission have improved in complexity and depth, integrating high-resolution data on demography, mobility and social behavior [1, 2]. The complete knowledge of the network of contacts through which an epidemic spreads is usually unreachable or impossible to implement, and for modeling purposes it is useful to remain at the coarse level of age-groups. Under this view, the population under study is divided into different groups according to its age distribution and different contact rates are assumed among these groups. Relevant examples are influenza-like diseases [6,7,8,9,10], pertussis [11], tuberculosis [12, 13], and varicella [14] They are instrumental for modeling and implementing more efficient interventions [15, 16]

Methods

Results

Discussion

Conclusion