Professor Anne Mills, of the London School of Hygiene and Tropical Medicine, introduced the 29th Bradford Hill Memorial Lecture on Tuesday 30th June 2020, recognising the profound impact of Sir Austin (Tony) Bradford Hill on the field of Medical Statistics.
This year’s speaker, Professor Christl Donnelly, of Oxford University and Imperial College London, has vast practical experience of applying methods in the context of infectious diseases, including SARS, MERS, Ebola and Covid-19. Her very relevant topic was 'Real-time analysis of Covid-19 – epidemiology, statistics and modelling in action'.
Professor Donnelly began with some of the questions related to Covid-19 we might ask, and explained how statistics, epidemiology and modelling can help to address these questions. For instance, how many cases are there? To estimate the number of cases in Wuhan early in the epidemic, researchers at Imperial used the number of confirmed cases exported from Wuhan, the average time to detection of a case, the catchment population, and the number of daily flights from Wuhan airport. On 17 January, when there were three cases outside China, they estimated about 1,700 in Wuhan with 95% confidence interval 400-4,500, although at that stage only about 60 cases had been reported.
One important question is how fast is the epidemic growing. We can answer this by estimating the now famous reproductive number, the average number of people each infected person will go on to infect in a susceptible population; as we all now know if this is below one the epidemic will not spread. Using their estimated number of cases in Wuhan, and a range of plausible values for the size of the original zoonotic infection, the Imperial team estimated a worryingly high reproductive number of 2.6 (2.1-3.5).
The question 'How serious is Covid-19?' can be addressed by estimating the case fatality ratio (the proportion of people included as cases who die), or the infection fatality ratio (the proportion of people infected with the virus who die). While it seems a simple task to estimate these quantities, it is in fact quite challenging. As many people infected with Covid-19 have no or very mild symptoms, they may never be included as cases. This is in contrast to SARS, where almost everyone infected with the disease has symptoms severe enough to require hospitalisation. Another challenge is estimating these ratios while the epidemic is ongoing. A simple estimate of the total number of fatalities so far divided by the number of cases will underestimate the true ratio early in the epidemic, as patients’ outcomes have not yet been observed. As this simple ratio increases over time, there is a danger it will be misinterpreted that the virus is becoming more deadly, and Professor Donnelly showed us sample headlines making this claim for SARS. Early in the epidemic, a better estimate of the case fatality ratio for SARS, where the distributions of time to death and time to recovery are similar, was the number of deaths divided by the number of deaths plus recoveries. An alternative, originally suggested by Professor Sir David Cox for SARS, is an interpolation of Kaplan-Meier curves for the death and recovery processes. This method is currently being used for Covid-19 to estimate a fatality ratio of 35%(34-36%) for hospitalised cases. The infection fatality ratio is considerably lower, a recent estimate being 0.66% (0.39-1.33%).
Estimates of the reproductive number, infection fatality ratio, and the distributions of the periods that the virus is incubating and in transmission are necessary for modelling. Individual-based simulations of transmission have been used to explore scenarios of unmitigated, mitigated, and supressed epidemics, and to make projections under different interventions. Modelling groups at LSHTM, Imperial and elsewhere have found that with a reproductive number greater than 2.5, there would be a lot of spread and substantial numbers of deaths.
Professor Donnelly demonstrated the potential of using aggregated movement data (for example provided by Facebook and O2 for the UK) linked to transmission data, to predict changes in the number of cases from changes in movement. There is a clear decrease in the reproductive number to under one with the reduction in mobility following public health interventions. As movement returns, keeping the reproductive number below one will require other measures, such as social distancing and masks.
What does the future hold? We will have more data on COVID-19, such as data on antibody testing. Appropriate data will, for instance, give us a better understanding of risks for different groups. There has been a remarkable impact on the process of scientific research, with thousands of Covid-19 preprints published online, along with code and data. Greater transparency runs the risk that problems are not identified by a peer review process, and findings may be misinterpreted. Communication of findings is important, and public interest leads to more need and opportunity for public engagement about the role of statistics, epidemiology, and modelling of infectious diseases. Professor Donnelly ended her talk with a mention of Florence Nightingale, the first woman fellow of the RSS and one of the earliest proponents of evidence-based medicine. The celebrations of her 200th birthday this year have been put on hold. Finally, drawing inspiration from Bradford Hill, and the current pandemic, Professor Donnelly reflected on the role of scientists – to present what-if scenarios with our best possible estimates and levels of uncertainty.
There were many questions for Professor Donnelly following the lecture. Topics included data quality, inter-country comparisons, risk factors, modelling assumptions, effects of interventions, communicating Covid-19 statistics, and the impact on the scientific peer-review process. Professor Stephen Evans (LSHTM) fielded the questions, transmitted the virtual applause, and thanked Professor Donnelly for a very insightful and timely lecture.
Download the presentation (PDF).
Written by Nicola Fitz-Simon, a lecturer in biostatistics at the National University of Ireland Galway and secretary of the Medical Section.