ESPE Yearbook of Paediatric Endocrinology

Nature. 2022;605(7910):423-5. doi: 10.1038/d41586-022-01332-8.PubMed ID: 35581494

Brief summary: This Nature comment was prompted by the very wide range of calculated values for the COVID-19 infections ‘reproduction number’ R, produced by different modelling teams despite access to the same datasets on UK’s emerging ‘2nd wave’ in October 2020. Values of R ranged from 115 new infected individuals (infected from 100 individuals infected at baseline) with a lower confidence limit of 96 (i.e. slightly declining infections) to 166 with an upper limit of 182 (i.e. almost doubling). They argue that no one answer is correct, rather the range of values indicates true uncertainty in the data.

This concept may seem bizarre to you – if so, I strongly encourage you to read this eloquent and well-illustrated article in full. While rarely (I hope) some published statistics might be due to use of invalid analyses or even deliberate ‘massaging’ of the data, these authors show that a range of valid statistical approaches exist to answer the same question with the same datasets. Choice, classification, and type of adjustment for potential confounders is one simple example of the possible differences. Another may be the decision to use non-parametric tests for skewed data, or to transform values to allow parametric tests. Indeed, in other ‘multi-analyst’ projects, independent statisticians rarely use the same procedure.

In October 2020, the UK was concerned about the possible re-emergence of COVID-19 (the ‘2nd wave’). The Scientific Pandemic Influenza Group on Modelling asked 9 teams to calculate the reproduction number R for COVID-19 infections. As well as differing modelling approaches, another source of variability between teams was likely their decision to prioritise the abundance of data available (on deaths, hospital admissions, testing rates). No two teams produced the same estimates.

The authors describe that such multi-team approaches are commonly used in the fields of high-energy physics, astronomy and climate modelling. They argue for the adoption of similar working in other fields where we place great importance on a single result, such as medicine, psychology, materials science, ecology, etc\..