This paper develops a new statistical framework for studying how infections spread through a population whose contact patterns change over time, even when the data are incomplete. The authors focus on the realistic problems that make inference hard: the true time when someone became infected is often unknown, the contact network may be observed only at times or with error, people can be infected from outside the observed group, and symptoms do not always mark the moment of infection. Their goal is to tie all of these moving parts together in one coherent probability model.

At a high level, the researchers built a joint model that treats disease progression, the changing contact network, and the observation process as interacting continuous-time random processes. They couple a standard disease model called SEIR — susceptible, exposed, infectious, removed — with a network whose links can change depending on individuals’ disease status. They also add explicit models for how symptoms and contacts are observed, which allows the framework to include intermittent observations and measurement error in recorded contacts.

The main technical contribution is the derivation of a complete-data event-history likelihood for this combined epidemic-and-network process under partial observation. In plain terms, the authors write down the full probability of all events (infections, recoveries, link changes, and observations) that could produce the data, even when some events are hidden. That complete-data likelihood gives a principled starting point for standard statistical tools, including likelihood-based and Bayesian inference performed by imputing the missing events (a technique called data augmentation).

Why this matters: current methods often treat these complications separately or make strong simplifying assumptions, such as that the network is fully observed, the population is closed, or symptom onset equals infection time. By putting everything in one model, the framework clarifies how information about disease and contact dynamics combine to determine which parameters can be estimated from partial data. The authors also show that many existing epidemic-network models fit inside their framework as special cases, so it unifies a range of approaches.