Using outlier detection methods to incorporate highly heterogeneous infection rates into compartment models

The article addresses the problem of superspreading events (SSEs), which represent extreme and rare bursts of disease transmission that standard compartmental models assuming population homogeneity fail to capture. The authors present a new framework that treats SSEs as statistical outliers in case count time series and incorporates them into SIR-type models through pulse terms. This approach separates anomalous SSE-driven transmission from background spread, thereby reducing bias in estimating transmission rates. The method was validated on synthetic data and subsequently applied to COVID-19 outbreaks in Hong Kong and the German district of Gütersloh. Results showed improved model fits and more robust estimates of background transmissivity. The modular framework is adaptable to various diseases and data contexts.