Limited data on infectious disease distribution exposes ambiguity in epidemic modeling choices (2401.15190v1)

Abstract: Traditional disease transmission models assume that the infectious period is exponentially distributed with a recovery rate fixed in time and across individuals. This assumption provides analytical and computational advantages, however it is often unrealistic. Efforts in modeling non-exponentially distributed infectious periods are either limited to special cases or lead to unsolvable models. Also, the link between empirical data (infectious period distribution) and the modeling needs (corresponding recovery rates) lacks a clear understanding. Here we introduce a mapping of an arbitrary distribution of infectious periods into a distribution of recovery rates. We show that the same infectious period distribution at the population level can be reproduced by two modeling schemes -- host-based and population-based -- depending on the individual response to the infection, and aggregated empirical data cannot easily discriminate the correct scheme. Besides being conceptually different, the two schemes also lead to different epidemic trajectories. Although sharing the same behavior close to the disease-free equilibrium, the host-based scheme deviates from the expected epidemic when reaching the endemic equilibrium of an SIS transmission model, while the population-based scheme turns out to be equivalent to assuming a homogeneous recovery rate. We show this through analytical computations and stochastic epidemic simulations on a contact network, using both generative network models and empirical contact data. It is therefore possible to reproduce heterogeneous infectious periods in network-based transmission models, however the resulting prevalence is sensitive to the modeling choice for the interpretation of the empirically collected data on infection duration. In absence of higher resolution data, studies should acknowledge such deviations in the epidemic predictions.