Transmission Dynamics of Eastern Equine Encephalitis: Global Sensitivity Analysis and SHAP Parameter Importance in an Age-Structured Vector-Host Model

Abstract

Eastern equine encephalitis virus (EEEV) is a deadly arboviral pathogen with 30% severe case fatality. EEEV exhibits pronounced 2–3 year cyclical outbreak patterns in the northeastern United States, linked to shifts in mosquito feeding preferences between hatch-year and adult avians. We developed an age-structured vector-host model incorporating differential feeding patterns of Culiseta melanura mosquitoes on European Starlings and American Robins. Global sensitivity analysis revealed mosquito biting rate (a) as the dominant driver in transmission, with avian infectivity (δ) and exposure (α) playing secondary roles. Pairing tree-based machine learning algorithms with SHAP analysis on 100,000 parameter sets identified parameter hierarchies that govern cyclic transmission. Adult avian mortality (µ_a_) was identified as the key parameter underlying cyclic and stable transmission patterns. SHAP further revealed these patterns to be grounded on opposite sides of the same epidemiological mechanism. Stable endemics emerge from demographic stability paired with transmission optimization, while cyclic endemics emerge from demographic instability paired with minimal transmission optimization. Numerical simulations illustrated critical threshold dynamics at 0.2 ≤ α ≤ 0.4, where heightened hatch-year exposure triggers demographic instability responsible for observed 2–3 year cycles, while balanced exposure (α ≈ 0.5) leads to stable endemics. These mechanisms provide a foundation for targeted surveillance and control interventions.