Modelling with SPEED: a Stochastic Predictor of Early Epidemic Detection

The frequency of emerging infectious disease outbreaks continues to rise, necessitating predictive frameworks for public health decision-making. This study introduces the Stochastic Predictor of Early Epidemic Detection (SPEED) model, an adaptation of the classic Susceptible-Infected-Recovered model, employing a Gillespie-like algorithm to simulate early-stage stochastic disease transmission. SPEED incorporates individual-level detection probabilities based on the infection time and the lag from GP consultation to lab confirmation. The model dynamically adjusts to public health responses by enhancing testing and reducing detection times once a single case has been identified. SPEED serves two key functionalities. First, as a statistical inference tool refining reproduction number estimates following the detection of a small number of cases. SPEED inference uses specified prior distributions for the reproduction number to provide reliable posterior estimates. Second, to simulate epidemic scenarios under specified values of the reproduction number in order to construct a distribution of the time to subsequent detections. The model is used to evaluate how second case timings can rule out higher values of the reproduction number. Comparisons with simulations under heightened surveillance scenarios demonstrate the model's utility in assessing response efficacy on the initial outbreak spread. Our results demonstrate SPEED applied to a single case of influenza A(H1N2)v, detected through routine flu surveillance on the 23rd November 2023.