Guidance

Epidemiological Modelling Frequently Asked Questions

Published 24 December 2021

1. Why do we need modelling?

• Infectious disease models are used to understand the spread of a disease through populations. We use these to investigate how the COVID-19 epidemic could unfold in the UK, and what the consequences could be for the UK’s population and public services.

• The model parameters include all of the main factors that influence the path of the epidemic – for example how infectious the virus is, how long people are infectious for, how susceptible people of different ages are to infection and to serious disease, and the effectiveness of vaccines.

• Models are useful tools when dealing with uncertainty about the future. They can help decision-makers because they can be used to look at different future scenarios – for example if policy interventions were made, new treatments or vaccinations became available or if a new variant emerged – by changing the modelling inputs and assumptions.

• For more information on how models work see the Introduction to epidemiological modelling.

2. Do the models produce forecasts?

• No. The outputs from the epidemiological models used for medium to long term planning (weeks to months) are definitely not forecasts. True forecasts, such as for the weather, only look a few days ahead because beyond that the level of uncertainty becomes too great. The epidemiological models typically look several months into the future.

• The models produce scenarios rather than forecasts. They are most useful for comparing scenarios (for example, how might things change if policies are brought in to reduce the number of contacts each person has) rather than focusing on the absolute numbers of each scenario. Scenario modelling is also a useful tool to establish general principles, for example that the earlier interventions are brought in the less time they are likely to be needed for.

• The inputs to the scenarios are a set of modelling assumptions, which are estimates of all key factors such as vaccine effectiveness in a given age-group and the infectiousness of the virus. A subset of these input variables – such as the average number of contacts each person has in a day – are influenced by policy and changing these assumptions is how the impacts of interventions are modelled. The output shows what this might mean in terms of infections, hospitalisations and deaths. Scenarios are often depicted in the form of a graph with a central estimate and an accompanying confidence interval that captures any and all uncertainty for each scenario. The wider the confidence interval, the more uncertain the future is in a particular scenario.

3. Isn’t the COVID-19 modelling always too pessimistic?

• No. But media coverage generally highlights the biggest number from the most pessimistic of a range of scenarios, which will usually be the very outer limit of the range of possibilities.

• Comparisons between old modelling scenarios and actual data should, however, be made with caution, remembering that these are scenarios rather than predictions, and will always be highly uncertain. And it is important to compare like with like. Scenarios make important assumptions that affect the results, for example that no further government interventions are made. Comparing this with what happened after the government did intervene is misleading.

• Some scenarios - so-called Reasonable Worst Case scenarios – are meant to be pessimistic. They describe a bad outcome that is unlikely but not impossible. This is an important tool for planners of public services where underestimating needs and pressures would cause significant problems.

4. Who decides what scenarios are modelled? Aren’t they just cherry-picked to make the point Ministers, SAGE or the modellers want to make?

• The academic modelling teams supporting the COVID-19 response are independent researchers, with international reputations for the high calibre of they work. They are not civil servants, but they support the COVID-19 response by taking requests from government for modelling.

• These requests are, rightly, for scientific advice (for example “what would reducing restrictions on mixing between households in hospitality settings do to the number of COVID-19 infections?”) rather than policy advice (for example, “should we re-open hospitality?”).

• Modellers and policymakers discuss possible scenarios to ensure they are most useful and relevant for policy purposes. As independent researchers, they are also free to do whatever other modelling they like and to publish it in academic journals.

• For any given scenario, modelling teams will use their professional judgement on the basis of the available evidence to choose the most appropriate assumptions to use in the modelling. In the case of modelling the effect, for example, of reducing restrictions on mixing between households, they will estimate what the likely effect of this would be on infections as it means people would meet up indoors more often. (Unsurprisingly, people’s behavioural response to policy changes is often one of the hardest elements to capture in the modelling). Policymakers are not permitted to influence this process, or to ask for a “better” or “worse” outcome.

• There is, however, complete transparency about the assumptions made, which are published, as are the models themselves, for everyone to challenge and to encourage others to bring forward relevant evidence. This is how science works and it helps weed out any errors and counter any biases.

5. Data from South Africa shows that Omicron is less severe than Delta - so why didn’t you include this in the modelling? Isn’t this an example of cherry-picking and deliberate pessimism?

• No. The modelling considered by SAGE on 16 and 20 December 2021 included a wide range of different assumptions for the intrinsic severity of Omicron, from equally severe to Delta down to 10% as severe. The models also assumed that people who have immunity from vaccination or prior infection would experience milder disease, which is important as Omicron is able to infect more such people than Delta. This would also reduce apparent severity.

• A range of scenarios for Omicron’s severity were included because of the uncertainty about how different this might be compared to Delta. It has taken time to get more robust evidence on the severity of Omicron compared to Delta, and this modelling was produced before the new data emerged. It is encouraging that the data from South Africa show a lower rate of hospitalisation from Omicron than was seen in the Delta wave earlier this year, but it is difficult to know how much this is a reflection of, for example, the immunity in the population from this Delta wave and also from vaccination rather than a difference in intrinsic severity. Data from the UK are now also emerging that support Omicron having lower severity than Delta but it is not yet clear precisely how much lower.

• How important is this estimate of severity for the overall impact of the Omicron wave? There is no question that, were the intrinsic severity of Omicron to be, say, 50% of Delta, it would be good news. It would mean, all other things being equal, half as many people needing to go into hospital. The main challenge with a fast-growing variant such as Omicron is, however, how quickly it is spreading. (Omicron infections as of mid-December 2021 were doubling approximately every 2 days). 50% lower severity would be cancelled out in one doubling, so hospitalisations would rise just as quickly, merely two days later than otherwise.