Model based COVID infection fatality ratio estimates should be used very cautiously

Bristol team emphasises the need for testing random samples of the population to establish disease prevalence and properly measure the size of the epidemic and disease severity.

The comment on a paper by Verity et al. (2020) from the Imperial College infectious diseases modelling team will appear in Lancet Infectious Diseases. It is written by Bristol’s Professor Simon Wood, Professor Peter Green and Dr Matteo Fasiolo, as well as Professor Ernst Wit from Università della Svizzera italiana, Switzerland, and was sent to the Lancet and the Imperial team on 10th April.

The Verity et al paper estimates the infection fatality ratio (IFR) for COVID-19 from a number of data sets. The team reviewed the models and data used and conclude that the data contain very limited information about IFR, while some of the model assumptions are very strong, suggesting that the IFR estimates should be treated with great caution for planning. They argue that this emphasises the urgent need for testing of random samples of the population to establish disease prevalence. Until such data are available, they propose using somewhat less problematic alternative IFR estimates. The Imperial projection of half a million deaths would be reduced by nearly 50 per cent using the alternative estimates, although the Bristol team believe life years lost to be a more sensible measure of epidemic impact.

Professor Wood said: “Sound epidemic management requires solid information about disease severity and epidemic size. We make the point that complex statistical models of inadequate data are a poor substitute for direct statistical assessment of epidemic size and hence disease severity.”

Further information

Read the comment at: