select search filters
roundups & rapid reactions
factsheets & briefing notes
before the headlines
Fiona fox's blog

expert reaction to data from the MRC Biostatistics Unit at the University of Cambridge nowcasting and forecasting COVID-19

The MRC Biostatistics Unit at the University of Cambridge have released new data which is nowcasting and forecasting the COVID-19 outbreak in the UK. 


Prof Brendan Wren, Professor of Microbial Pathogenesis, London School of Hygiene & Tropical Medicine, said:

“The modelling data from the University of Cambridge appears encouraging and suggests the number of new COVID-19 cases is dropping significantly particularly in London.  The study has high confidence levels but is a model and needs to be backed up real time diagnostic tests.  Currently, there are around 3,000 positive tests reported daily (new and existing cases) and we need to determine what are the current reservoirs of infection.  Who is reporting positive, why are they positive and who are their contacts?  A lot of those reporting positive are likely to be health workers.  We need a targeted testing, tracing and tracking procedure then the R number will consistently decrease throughout the country.”


Prof Matt Keeling, Professor of Populations and Disease, University of Warwick, said:

“I’d like to express a note of caution regarding the figures quoted from the University of Cambridge’s MRC Biostatistics Unit.  I’ll start by saying that I have the greatest respect for both the researchers and the science from this world leading unit – but as ever science is about thorough scrutiny by one’s peers.

“On the whole I would generally agree with many of the findings reported by the Unit – R is generally about 0.7-0.8 in many regions of the country, although potentially slightly higher in the devolved nations.  However, I feel the R value quoted for London is extremely low – Warwick University predictions put the value closer to 0.6; this still predicts a faster decline in London than the rest of the country but not as dramatic as predicted by the MRC Biostatistics Unit, and therefore we would predict far more infections.  The difference between 0.4 and 0.6 might sound like scientists arguing over small details, but these translate to very substantial differences in the rate of decline of cases (or the time for the epidemic to halve).

“So what is generating these differences?  The MRC Biostatistics Unit has considered deaths due to coronavirus as their main quantity of interest – in London this is falling far more rapidly than other measures of the outbreak such as hospital admissions.

“What this study highlights is that London is experiencing a subtly different epidemic to other regions of the country, with a more rapid decline and different relationships between ICU occupancy, hospital occupancy and deaths.  This is clearly something that needs to be understood in more detail, as it may be important for how different areas of the country exit from lock-down.

“I am extremely worried about the media message that ‘London could be coronavirus free in days’.  At the moment there is way too much uncertainty in factors such as asymptomatic transmission or infection in the health-care worker population to make this claim.  Londoners have made a brilliant effort in adhering to lock-down advice, and bringing the infection under control.  If people think London is coronavirus-free that could be dangerous, and could lead to complacency, undermining all the struggles and sacrifices that everyone has made so far.  A relaxation of vigilance could easily see R increasing above 1, and a second epidemic wave.”


Prof Paul Hunter, Professor in Medicine, UEA, said:

“It is not clear from the easily available documentation what methods and models this group have been using to produce their estimates and predictions.  From their website they appear to have based their models on Bayesian models of Influenza A/H1N1.  Whether influenza models are directly applicable to COVID19 is still an area for debate.

“Looking at their results the Cambridge group estimate that as of 10th May that there were 6,540,000 (5,060,000–8,480,000) cumulative infections in England representing almost 12% of the population.  This report has become available at about the same time as the ONS produced its data covering the two week period 27th April to the 10th of May.  The Cambridge model would predict some 220,000 new cases during this same period.  The ONS reported that 148,000 people in England had the coronavirus (COVID-19) (95% confidence interval: 94,000 to 222,000).  However this figure is not estimating the same thing as the Cambridge group.  Within the 148,000 will be quite a few people who had become infected before 27th April and were still shedding the virus.  Consequently the number of new infections in the ONS data will be smaller still.  Another discrepancy is in the age distribution.  The Cambridge group predict a strong age distribution with a high attack rate (18%) in school children and a low rate in older people, especially those over 75.  The ONS report finds no such strong age distribution though the confidence intervals are large.

“Whilst the confidence intervals around both studies are high there does seem to be some importance differences in conclusions drawn from modelling and from surveillance of actual infections in the real world.  Both approaches are subject to uncertainty and potential biases and it remains to be seen which gives the more precise estimates of actual disease burden.  The results of antibody testing when available will provide additional useful information.

“The results of both studies provide valuable and useful information in the management of the current epidemic, providing we do not make the mistake of assuming that any one study is necessarily accurate.  A full view of the epidemic will come from multiple studies and sources of information that at times may well appear conflicting.  But this is how science work as it reaches consensus.  In the meantime we must be careful about making statements about actual numbers of new cases per day without pointing out the large uncertainty.  As you try and drill down to the regional level the relative uncertainty is likely to be even greater.”


Dr Thomas House, Reader in Mathematical Statistics, University of Manchester, said:

“This study uses robust methods to estimate the trajectory of the epidemic in different regions.  The authors make careful use of death data, which avoids some common sources of bias.  Due to the delays between onset of COVID and death, and the fact that Government measures have recently changed with effects on R that remain to be seen, these results are not intended to be used for forecasting.  It is not the case that we can tell from this data that the number of new cases in London is down to 24 – this was modelling fitted to deaths and predictions will change due to the changes announced last Sunday.”


Prof Sheila Bird, Formerly Programme Leader, MRC Biostatistics Unit, University of Cambridge, who was not involved in this work, said:

“Press coverage of the important recent dissemination of work by MRC Biostatistics Unit and Public Health England is rightly interested – we’ve all been waiting for sight of SPI-M’s inputs to SAGE.  However, excitement has to be moderated by the cautions that the Cambridge team clearly set out: they have yet to have access to all COVID-mention deaths, which are registered (after some delay – inevitably) by Office for National Statistics, and their work is reliant on the patient having been COVID-confirmed (via testing).

“The back-calculation element from confirmed COVID fatalities of the Cambridge/PHE work means that the well-estimated reproduction numbers pertain to transmission rates which give rise to this type of fatality, not all COVID-mention deaths.  That is one key reason for caution.

“There is another, which is that extrapolation beyond the time-frame of the observed data (tests, hospital admissions, deaths) has to be assumption-led and is liable to change when informed by data in the latter part of May – and especially so when some loosening of lockdown has been signalled.

“Precisely for this reason, the Cambridge/PHE team has put down a marker for how things looked before any relaxation came into effect.  And it is a very good and appropriate that they did so.  Hence, over-excitement about what may, or may not, transpire later in June is premature.

“The ONS’s Infection Survey results hint that under-testing remains an issue.”


All our previous output on this subject can be seen at this weblink:


Declared interests

Prof Sheila Bird: “SMB is a former Programme Leader at MRC Biostatistics Unit, Cambridge and so the work discussed is by my colleagues.”

None others received.


in this section

filter RoundUps by year

search by tag