epymorph turns spatial epidemic model comparisons into controlled forecasting experiments

epymorph improves forecasting of hospital admissions during the 2025-2026 U.S. influenza season, but only when using an empirically-derived contact matrix alongside age stratification.

• Why it matters: Metapopulation models often mix transmission, movement, and geography in hard-to-reproduce code, making it difficult to isolate which assumptions drive forecast performance.
• What they did: The Python package epymorph builds spatial models from separately specified transmission, movement, and geographic components, retrieves each component’s required data, and estimates time- and space-varying parameters with Bayesian filters feeding probabilistic forecasts.
• The result: In 2025-2026 influenza, age stratification improves hospital-admission forecasts only with the empirically-derived contact matrix; in the 2020-2021 SARS-CoV-2 winter wave, commuting data improves county forecasts while a state-level model best fits state forecasts.

AI summary of the abstract · may contain errors · not medical advice · read the paper.