Accounting for overdispersion in case counts is essential when generating prediction intervals for early infectious disease outbreaks, as ignoring it causes empirical coverage probability to deviate considerably from nominal levels.
The investigation focused on epidemic situations where an infectious disease spreads to an unexpectedly high number of cases, creating major consequences for the health system and society at large. The authors examined this early outbreak window by distinguishing between the initial period of approximately exponential infection growth and the subsequent second early phase where cumulative infection numbers begin to flatten.
To compare performance across these distinct outbreak stages, the team evaluated three statistical modeling frameworks: Gaussian non-linear least squares, Poisson regression, and negative binomial regression. Each model type was evaluated using two different mean functions to describe the progression of infections. These consisted of an exponential mean function to capture early rapid expansion and a Gompertz growth function to capture the phase where cumulative infections begin flattening.
The methodological analysis incorporated a phase-specific comparison of model fits alongside prediction intervals for new infections. The researchers assessed both coverage accuracy and interval width to address uncertainty in predicting case numbers.
In addition to comparing conventional models, the study presented an improved version of well-known parametric bootstrap prediction intervals tailored for the exponential growth phase. The researchers also addressed modeling the subsequent flattening phase, delivering what is to their knowledge the first implementation of parametric bootstrap intervals for models using the Gompertz growth function as their mean function.
Subject of Research: Prediction intervals and model comparisons for early outbreak case forecasts
Article Title: Prediction intervals for early outbreak models
Article References: Kück, F., Friede, T., & Leha, A. (2026). Prediction intervals for early outbreak models. BMC Medical Research Methodology. https://doi.org/10.1186/s12874-026-03064-z
Image Credits: AI Generated
DOI: 10.1186/s12874-026-03064-z
Keywords: prediction intervals, early outbreak models, infectious disease outbreaks, exponential growth phase, Gompertz growth function, overdispersion, coverage accuracy, interval width, Poisson regression, negative binomial regression, Gaussian non-linear least squares, parametric bootstrap, case count predictions, epidemic modeling, mean functions, uncertainty quantification, cumulative infections
News Source: Kristina Jarvis. (November 17, 2026). Accounting for Overdispersion Improves Prediction Interval Coverage in Early Outbreak Models. Scienmag.



