Who infects whom in an epidemic?

Epidemiological surveillance systems often provide data on specific characteristics of an infected population. For instance, sex, geographical location, socioeconomic level and age of the registered individuals. This allows us to study the population divided into groups. However, information on the dynamics of infected people classified by group is not usually exploited when analyzing the evolution of an epidemic. In this work, we propose a tool to analyze how the spread of an epidemic is heterogeneous among different population groups. Based on records of infected individuals we identify synchronicity and causality interactions among population groups. Describing this dynamics and which population groups are the first focus of infection is essential for decision makers. We represent time series by population group and their degree of similarity using a weighted graph, and we apply a community detection algorithm to partition this graph. Each community is composed of synchronized age groups. The direction of interaction among different communities is identified using sample cross-correlation in a domain that can indicate causality. This is illustrated by considering age groups and using datasets of COVID-19 in Jalisco, Mexico and influenza A(H1N1) in the USA. In both cases, the proposed methodology detected which age groups show synchronized behavior across time, and which age groups influence the subsequent appearance of epidemic outbreaks in other groups.