Collaborative outbreak modelling for decision support: evaluating trade-offs from multi-model combination

Infectious disease modelling is a useful tool for supporting outbreak control, offering to interpret the complex uncertainty of epidemiological dynamics. This uncertainty allows for many approaches, choices, and interpretations during modelling work, producing a wide variety of modelling results. Multi-model collaborations create comparability across this diversity, and the opportunity for evidence synthesis. This often includes a quantitative combination of numerical model results. This thesis evaluates collaborative modelling work during the COVID-19 response in the UK and Europe. Four papers draw from the UK’s Scientific Pandemic Infections group on Modelling (SPI-M), and the European COVID-19 Forecast and Scenario Hubs. First, I found that outbreak detection may be confounded by combining estimates of the reproduction number, aggregating over relevant heterogeneity. Next, I evaluated ensemble projections of both short- and long-term COVID-19 incidence, characterising predictive performance, representation of uncertainty, and policy relevance. I then identified tensions in the structural sustainability of modelling work for outbreak response. A thematic analysis draws out shared challenges in collaborative outbreak modelling. Modelling collaborations are vulnerable to sampling biases that may limit the validity of multi-model combinations, while also facing varied and competing stakeholder needs. Meanwhile, collaborative decision support may face a fundamental trade-off between offering consensus versus context: creating a single evidence synthesis risks losing insight into heterogeneous epidemic dynamics. A further trade-off challenges collaboration with capacity: multi-model combinations depend on consistent model components, despite collaborators’ constrained capacity during emergency response. Selecting an appropriate strategy to resolve these tensions likely depends on the purpose, timing, and scale of outbreak decision-making. Future work should explore the validity of epidemiological inference from multi-model combinations, and clarify both capacity constraints and stakeholder needs at the science-policy interface.