Waning immunity, media fear-mongering and fast-paced predictions. The University of Otago's Matthew Parry answers some questions about Covid modelling.
► A lot of official modelling projections are quite detailed, what are the main points for the public to keep in mind?
I think it is important to realise that making projections at the early stage of an outbreak is very difficult. This is because random events can have a large effect on what happens next – for example, whether a person who was infectious travelled to work on a crowded bus or drove to work instead.
Modellers do account for this in their models, but it means there is initially a great deal of uncertainty in the projections. This is another important thing to keep in mind – projections are not just a single number, say of positive cases, at some time in the future. Rather most models give a range or “fan” of possible futures. As we get more data, we start to see where in the fan we are.
What we can expect as the case numbers grow in Aotearoa is that the relative uncertainty in the projections will start to decrease. This is because random events will become less influential. Because we have not had a full-blown Covid-19 epidemic in Aotearoa, this is actually new territory for our modellers.
► How quickly do modellers need to analyse and respond to changing case numbers in order to generate new predictions?
There are two key ingredients to a model of an epidemic in a population. The first is the fundamental description of what is going on i.e. the epidemic process. For example, an infected person has varying levels of infectiousness over a period of time before isolating. The second ingredient is parameters. For example, how long is an infected person infectious for? What is the probability they will end up isolating?
In the early stages of Covid-19 modelling, the fundamental description of the models changed quite quickly in order to answer more detailed questions. Currently, however, the fundamental description is somewhat settled.
On the other hand, the parameters of the model often need to be fine-tuned. This is because each variant is different. For example, there is some evidence that individuals infected with Omicron become infectious faster than individuals with Delta. The only way to determine the value of the parameters is with data.
So, to answer your question, modellers will typically update their models as soon as new data comes in.
► Are you concerned people will lose trust in modelling as the recent predicted numbers have not hit?
That is always a concern. I think part of the problem is that the media often focus on a single 'headline' number, so the model’s range of possible futures is overlooked or deemed too hard to communicate.
Also lost is that model predictions typically come with qualifying statements like: “if things continue like this then…” or “if we assume the value of this parameter can be estimated from overseas data…”. People often forget that things didn’t continue like that because we had a lockdown or because we went out and got vaccinated – or it turned out that the experience overseas was not relevant to Aotearoa’s context.
► Do you think the way Covid-19 modelling is being communicated in the media risks being fear-mongering?
By and large, no. I think most commentators are being very responsible and realise that modelling is used to aid decision-makers by giving them a range of possible scenarios.
However, there is always a danger that model output can be misinterpreted or misused. For example, we had predictions of 80,000 cases in Aotearoa by Waitangi Day weekend — based on model output from overseas modellers. Although this was quickly debunked, it made for alarming headlines.
In my experience, modellers both here in Aotearoa and overseas are very cautious about their predictions. They realise that a lot is at stake. Typically, they make sure their models and outputs are given as much prior scrutiny as the time allows.
► How does modelling in Aotearoa compare to other countries in terms of how it is done and how it stacks up in terms of accuracy?
As I mentioned before, unlike most parts of the world, we haven’t yet had a full-blown Covid-19 epidemic. This means it is hard to compare like-for-like predictions. However, it is worth pointing out that much of the modelling done here has been peer-reviewed internationally. The models used in Aotearoa are very similar to those used overseas but they have been carefully adapted to our situation.
► Do you think we'll need further boosters to keep cases on the low side, as other modellers have suggested?
This seems very plausible. One reason is that, unlike many other countries, we have extremely low levels of prior immunity in the population. Because of this, boosters remain the best way to up our protection against Omicron.
► Why is Covid-19 modelling important to have?
I can think of three immediate reasons why modelling is important. The first is because we want to know how waning immunity might affect the course of the epidemic, especially going into winter. How quickly waning occurs is an important model parameter.
The second reason is that we have to be prepared for any new variants that may crop up at any time and change the nature of the epidemic, just as Omicron has.
Finally, as we prepare to open up our borders to the rest of the world, modelling will help us understand the impact of imported cases on the epidemic in Aotearoa.
