Towards a statistical framework for epidemic trajectory modelling? South West England as a case study
19th June 2020, 2:00 pm – 3:00 pm
Meeting link will be distributed the day before the talk. Or contact Song Liu (email@example.com) for the link.,
We developed a regional model of COVID-19 infection dynamics from NHS data in the South West of England, for use in estimating the number of infections, deaths and required acute and intensive care beds. These results are interesting and not in complete agreement with the national picture, due to local demographic details. The model is an SEIR-style deterministic compartmental model with many parameters. We used latin hypercube sampling to search for good parameters, and attempt to account for data quality issues, such as poor reporting of infections. See https://www.medrxiv.org/content/10.1101/2020.06.10.20084715v1 for details.
But what is the "right" thing to do, from a statistical perspective? The second half of this session will be used as a discussion forum, asking about possible frameworks for fitting such models. How should prior knowledge - for example, parameters estimated on different infections - be integrated? How does identifiability fit into statistical best practice?
In short, what is the "gold standard"? What options exist that are sufficiently robust and standardized such that they could be rolled into NHS trusts across the country? And should we work towards a new methodology for the "next epidemic"?