Preserving friendships in school contacts: An algorithm to construct synthetic temporal networks for epidemic modelling

High-resolution temporal data on contacts between hosts provide crucial information on the mixing patterns underlying infectious disease transmission. Publicly available data sets of contact data are however typically recorded over short time windows with respect to the duration of an epidemic. To inform models of disease transmission, data are thus often repeated several times, yielding synthetic data covering long enough timescales. Looping over short term data to approximate contact patterns on longer timescales can lead to unrealistic transmission chains because of the deterministic repetition of all contacts, without any renewal of the contact partners of each individual between successive periods. Real contacts indeed include a combination of regularly repeated contacts (e.g., due to friendship relations) and of more casual ones. In this paper, we propose an algorithm to longitudinally extend contact data recorded in a school setting, taking into account this dual aspect of contacts and in particular the presence of repeated contacts due to friendships. To illustrate the interest of such an algorithm, we then simulate the spread of SARS-CoV-2 on our synthetic contacts using an agent-based model specific to the school setting. We compare the results with simulations performed on synthetic data extended with simpler algorithms to determine the impact of preserving friendships in the data extension method. Notably, the preservation of friendships does not strongly affect transmission routes between classes in the school but leads to different infection pathways between individual students. Our results moreover indicate that gathering contact data during two days in a population is sufficient to generate realistic synthetic contact sequences between individuals in that population on longer timescales. The proposed tool will allow modellers to leverage existing contact data, and contributes to the design of optimal future field data collection.Author summary: Face-to-face contacts occur between individuals throughout day-to-day activities. These contacts form a network of opportunities for the spread of diseases, such as COVID-19 or influenza. Modelling the transmission of respiratory viruses along such networks can provide insights on the efficacy of mitigation measures. To be actionable, such models must be fed with realistic contact data covering the time scale of an epidemic. However, high-resolution data on temporal contacts are most often limited to recording windows of a few days. In this manuscript, we propose two methods (the friendship-based and class-mixing-based approaches) to synthetically extend high-resolution contact data recorded in a secondary school. The class-mixing-based approach takes into account the organisation of students in classes, the overall activity patterns dictated by the school timetable, and the observed contact duration distributions. The friendship-based approach is further optimised to preserve friendships among the students, which induce repeated contacts with correlated durations. Using an agent-based model describing the spread of COVID-19 in a school, we show that taking into account friendships when creating synthetic contact data leads to transmission chains among students that differ substantially from those obtained with class-mixing-based contacts. Both types of contacts instead predict similar transmission pathways between classes. Our manuscript thus provides tools to generate improved model inputs from existing data with limited temporal scales, and an evaluation of the impact of preserving friendships on simulated COVID-19 outbreaks.