Compensating for population sampling in simulations of epidemic spread on temporal contact networks

Abstract

Data describing human interactions often suffer from incomplete sampling of the underlying population. As a consequence, the study of contagion processes using data-driven models can lead to a severe underestimation of the epidemic risk. Here we present a systematic method to alleviate this issue and obtain a better estimation of the risk in the context of epidemic models informed by high-resolution time-resolved contact data. We consider several such data sets collected in various contexts and perform controlled resampling experiments. We show how the statistical information contained in the resampled data can be used to build a series of surrogate versions of the unknown contacts. We simulate epidemic processes on the resulting reconstructed data sets and show that it is possible to obtain good estimates of the outcome of simulations performed using the complete data set. We discuss limitations and potential improvements of our method.

Évaluer le risque épidémique à partir de données lacunaires

“…B. Si on retire une certaine fraction des personnes du réseau, et donc les contacts qu'elles ont avec le reste de la population (en pointillés), et qu'on réalise les simulations sur le réseau restant, on observe que, plus la fraction « inconnue » est grande, plus on sous-estime la probabilité d'avoir de grandes épidémies. C. En utilisant la méthode présentée dans [7] pour réintroduire cette fraction « inconnue », on peut construire une version alternative mais plausible des contacts manquants (en rouge) à partir des données partielles. On retrouve alors une très bonne évaluation du risque épidémique, même pour des fractions inconnues grandes (jusqu'à 50 %).…”

“…Dans notre étude [7], nous avons considéré trois contextes distincts : un lycée, dans lequel le réseau de contacts est fortement structuré par les classes et sa dynamique par les horaires des cours ; une confé-rence, pour laquelle le réseau n'est pas structuré et la dynamique des contacts peu contrainte par des horaires ; un immeuble de bureau, structuré en départements, mais sans emploi du temps fixé. …”

Measuring social networks in primates: wearable sensors vs. direct observations

“…[1-3, 3-8].6 With the increasing deployment of digital devices, new ways of collecting data, 7 combined with new network analysis tools, have made possible the development of 8 quantitative measures of these relationships and patterns in modern human societies, 9 January 17, 2020 1/20leading to the emergence of computational social science in the last decades [9]. For 10 instance, social relationships have been inferred and studied using various data sources 11 ranging from phone calls [10], e-mails [11,12], online interactions [13], to face-to-face 12 interactions measured by wearable sensors [14][15][16][17][18][19][20].…”

“…Wearable sensors data A subgroup of 13 baboons, consisting only of juveniles and 127 adults (all individuals at least 6 years old) were collared with leather collars. The 128 collars were fitted with wearable proximity sensors (RFID tags) developed by the 129 SocioPatterns collaboration (http://www.sociopatterns.org/), already used in many 130 studies involving humans [14,16,18,20,40,44,51], and recently also animals [39]. In our 131 setting, each sensor was secured in a customized box specially designed with a 3-D 132 printer to contain the sensor and a long-life battery connected to it.…”

“…On the other hand, wearable sensors do not yield 52 information on the type of behavioral interactions and they do not register contacts 53 with individuals not wearing any sensor, such as very young individuals or out-group 54 members for instance. The quality of the collected data might also depend on 55 infrastructure constraints and potential technical failures, so that sampling issues must 56 also be considered carefully [20,40]. 57 Thus, data obtained from direct observations and from wearable sensors 58 infrastructures have very different nature and could in principle lead to very different 59 descriptions and understanding of the social bonds between individuals and of the 60 resulting social networks [41,42].…”