Guiding epidemic intervention decisions across space and time

Deciding when and how to intervene during an unfolding outbreak is challenging due to unreliable data, uncertain intervention effects and extreme sensitivity to timing. For example, counterfactual analyses suggest that introducing COVID-19 lockdowns just one week earlier could have halved deaths in some settings. While epidemic models are widely used to inform policy, the tools available to robustly trigger and evaluate interventions remain underdeveloped. In this talk, I present two advances toward more reliable real-time epidemic response. First, I show how applying standard outbreak indicators across spatial scales can delay detection and distort risk assessment. To address this, I develop the risk-averse reproduction number [1,2], derived using E-optimal design, to enable earlier and more reliable outbreak warnings. Second, I highlight critical gaps in methods for guiding intervention decisions that must balance competing health and societal objectives. I introduce EpiControl [3,4], a real-time, decision-support framework that uses model predictive control. EpiControl proposes optimal interventions that adapt to evolving data and remain robust to unanticipated dynamics (e.g., novel disease variants). [1] K Parag, U Obolski. Risk averse reproduction numbers improve resurgence detection. PLOS Comput Biol 19 (2023). [2]K Parag, M Santillana, A Cori, U Obolski. The R = 1 threshold can misclassify epidemic stability. Commun Phys (2026). [3] S Beregi, K Parag. Optimal algorithms for controlling infectious diseases in real time using noisy infection data. PLOS Comput Biol 21 (2025). [4] S Beregi, S Bhatia, A Cori, K Parag. EpiControl: a data-driven tool for optimising epidemic interventions and automating scenario planning to support real-time response. medRxiv (2025). [Includes EpiControl R package].