Q&A: How does Nphet's modelling work and how accurate is it?
Chief Medical Officer Dr Tony Holohan said that advice was given to Government based on Nphet modelling projects which indicate there could be more than 200,000 cases of the virus next month. Picture: Colin Keegan, Collins Dublin
A host of new restrictions were announced on Tuesday in response to the high number of Covid-19 cases being reported.
Under the measures announced:
- Bars, restaurants and nightclubs will have a midnight closing time from tomorrow;
- There will be a return to working from home;
- Digital Covid certs must be presented in theatres and cinemas (they were not introduced for hairdressers and gyms as had been expected)
- Entire household must self-isolate for five days following a positive case.
Chief Medical Officer Dr Tony Holohan said that advice was given to Government based on Nphet modelling projects which indicate there could be more than 200,000 cases of the virus next month.
Health Minister Stephen Donnelly said the modelling presented by Nphet on Monday evening was "very stark".
But what is modelling and how do Nphet use it to guide how we move forward in the pandemic?
Professor Turlough Downes, from the School of Mathematical Sciences at Dublin City University, says a model gives us an idea of how things work such as weather or climate or, in this case, a virus like Covid-19.
"For example, we know that if somebody walks into a room and they have Covid and nobody else has Covid then there will be an increase in the number of cases of Covid in that room because that person will infect people."
With this basic understanding, you can then look at what it would mean if that happens every time somebody walks into a room. What overall effect is that going have?
In terms of Nphet and Covid, a model has been built beginning with some basic observations and moving on to work out how the system works overall.
"We try to say, ok if that understanding is right then what do we think will happen in the long-term?"
Over the months since Covid arrived on our shores, there have been more observations and these are refined and built upon.
Modelling is not a way to predict the future but rather forecast what may happen given certain parameters.
For example, if the transmissibility of the Delta variant is a little bit more than first thought or happens in slightly different ways these parameters are put into the model.
"We take the worst case scenario of all those parameters, what we think will make the worst case scenario and then that is what is published as the pessimistic scenario. Then you say, well what about the other side?"
You input these other parameters and work out what the best case scenario would be.
"In neither of those cases are we actually trying to predict what would actually happen because it there will be a certain amount of uncertainty. For example, what proportion of cases are going to be hospitalised?"
Prof Downes goes on to explain: "You take the pessimistic and the optimistic and you're saying that these are our bounds. That is the worst that can happen and this is the best that will happen."
All of this, however, is based on the the underlying assumptions in the model being correct.
Sometimes there can be factors at play that the modelling team are not aware of. An example of this would be some of Nphet's modelling earlier in the year which may not have factored in the idea of waning immunity because it was not something that we had a lot of information on early in the vaccination roll-out.
Covid-19 is very sensitive to changes, says Prof Downes. If you change the input even just a little bit, the output change is huge.
"It is really, really, sensitive to some of these parameters which is why some of the models seem kind of ludicrously wrong occasionally."
According to Prof Downes, these things can be worked into models.
The European Model, which is a sophisticated model, runs full-scale simulations using things like mobile phone data, travel bookings, traffic data etc.
"They run the model and they say, well if this number of people are infected and the distribution around the community is like this, how many more are going to get infected?"
With this type of model, things such as seasonality can be factored in quite easily but Prof Downes notes that is it very computationally intensive to run these very large models.
"In the kinds of models that are used - say that NASA and other national agencies around Europe and the world would use - they are a bit more simplistic than that.
"What they basically say is if 10% of the people are infected then let's assume those 10 are evenly mixed throughout the country. Which we know isn't the case...We don't infect in a uniform way. So these models are going to miss things like seasonality."
He added: "Under those circumstances, the best thing that you can do is say that the likelihood of the probability of a person infecting another person goes up in winter and down in summer. Even if that's not technically what's happening that is the kind of way that you can make that model do what you need it to do.
"Inevitably because you are simplifying things so much, it's not going to be a great approximation of what actually happens. You can work it in a little bit but it's not going to be brilliant."
Speaking to the on Cork's 96FM, Prof Downes said that when it comes to Covid-19, we are in a situation of uncertainty.
"What we can say though, is that if you look at the growth rates say in hospitalisations and ICUs now, if that keeps going as it is then by Christmas we will have the same number of people in ICU as we had last January.
"It is not unreasonable to say that that will keep going. There's nothing obvious that will stop that from happening unless we bring in mitigations."
