So what can Australia's modelling tell us about the months ahead and how does ours compare to other countries?
How does modelling actually work?
The modelling of infectious disease spread and containment is part of a larger field of computer modelling that is also used to forecast weather, simulate flight paths and develop drugs, among other things.
It uses mathematics and computer science to create simulated populations in which people are given certain behaviours and risks. Different variables or parameters can be fed into the model to see how changes in policy or behaviours affect outcomes.
Since the start of the pandemic, a variety of COVID-19 models have been produced in Australia, all designed differently and offering different answers based on the data and formulas used.
They are constantly updated as new data becomes available and as new scenarios eventuate, which in an outbreak, can happen quickly.
The Doherty Institute modelling that has informed the federal government’s national plan, for example, has been updated to account for current Delta outbreaks, 12- to 15-year-olds accessing the vaccine, and smarter test, trace and isolate methods.
The modelling has been informed by a consortium led by Professor Jodie McVernon that includes experts from around Australia. Chris Baker, a research fellow in statistics for biosecurity risk analysis at the University of Melbourne, is one of them.
Dr Baker says modelling is all about collaboration.
“As modellers, we definitely welcome what other modellers are putting out. It’s important to interrogate where models are different and what’s driving those differences,” he said.
The information they provide helps governments predict demand on public health systems and decide how and when to increase or reduce restrictions.
Models also allow us to track progress. Comparing Doherty’s current modelling released in its recent sensitivity analysis with its first report shows Australians how far they’ve come, Dr Baker says.
“If you look back at the dates of when we expected to reach certain vaccination coverage thresholds, we’re ahead of that now.”
Professor Catherine Bennett, chair of epidemiology at Deakin University, describes models as “fantastic” tools to try and understand the pandemic and suggest how things might play out under different scenarios.
Now they’ve entered the public sphere, they’ve also become part of the messaging, she says.
“People start to understand how this all works and how their actions translate into benefits," she said. "If people are seeing the modelling and they're understanding it, you have a better chance of using it as a really important educational tool as well.”
Prof Bennett says models aren’t about predicting the future - something she describes as a “risky thing to do”. Their value lies in the comparison of different scenarios at each step in a pathway.
“That’s useful because you’re seeing rising control of your vaccine coverage offset against easing restrictions. We can see how well the model fits as we go.”
What limitations do models have?
Experts warn it’s important to not rely too heavily on models or to over-interpret them.
Dr Baker says they're not designed to forecast the number of infections or deaths in the future.
“It’s about how we can use a model to think through what policies we want to make. It’s a subtle, but important distinction,” he says.
He also says the public should bear in mind there’s a lag before new data is reflected in modelling.
“There’s often a delay from when new data emerges and a new scientific picture is formed, to when that becomes absorbed in the modelling, to when it can become public.
“By the time the public sees it, things can look a little bit out-of-date.”
Epidemiologist Margaret Hellard is a deputy director at the Burnet Institute, whose modelling informs New South Wales and Victoria’s roadmaps out of the pandemic.
Prof Hellard cautions that models are one tool among a suite that should be used to inform COVID-19 responses.
It’s also important people don’t come to expect the scenarios laid out in them because models are “not an exact science”, she says.
“There’s often a misunderstanding on how we’re supposed to use models - by the general public and the media,” she said. “Media uses them to put pressure on the government, who then feel under pressure to make decisions that aren’t always the best ones.”
People may want numbers but models can only offer ranges, she says.
“There is no absolute certainty in a global pandemic. Models are helpful tools but they’re still assumptions. The reality of what you have on the ground is what you have to deal with.”
What assumptions do models make?
The Burnet Institute uses scenario-based modelling for NSW and Victoria, which assesses the likely impacts of restrictions and the easing of them.
The scenarios it assumes can be more optimistic than what may eventuate or more pessimistic, though the reality is usually somewhere in between, Prof Hellard says.
On its website, the Burnet Institute lists its optimistic assumptions as schools being able to achieve a 50 per cent reduction in transmission risk through ventilation and other measures; no waning of vaccine immunity over time; ongoing compliance with restrictions; vaccinated people continuing to quarantine for 14 days if identified as contacts; 95 per cent compliance with vaccine mandates; and vaccines being delivered equally across all sub-population groups.
At the same time, its negative assumptions assume that the current epidemic growth rate will continue (taking into account declines due to vaccine immunity) and that there will be no impact of seasonality.
“We sincerely think seasonality is probably really important but nobody can give us the measure and if it’s not certain, you don’t put it in the model,” Prof Hellard said. “We’re really clear to governments that this is likely to be a pessimistic projection but we don’t know by how much. So we tell them to factor in a gain of seasonality and the implications of that.”
In addition, most models assume that people live in a city and not in a whole range of communities, and that vaccination rates are evenly distributed.
This means they're unable to calculate how COVID-19 will work in different areas.
They can also attribute only basic properties to individuals such as age, household structure and participation in different contact networks.
Prof Bennet says it’s a shame that models can’t factor in things such as ethnicity, attitude to testing and attitude to vaccination, because these details “really matter in a pandemic” and could “fundamentally change” outcomes.
“If you step into a situation where you have the worst case scenarios - a vulnerable group of people, low vaccination rates, mixing in a way that increases transmission risk - and the virus gets there, that’s not going to fit a model that’s based on average vaccination rates, even if they’re at 90 per cent," she said.
Models' inability to offer geographic specificity could become more limiting as NSW and Victoria emerge from lockdowns and the health response needs to become more localised.
Furthermore, some key data remains unavailable for modelling. We still don’t know exactly how vaccination affects incubation and infectious periods, Prof Bennet says.
“We couldn’t live without modelling but equally, we’ve got to know how to work with it so we don’t lose sight of its strengths or be annoyed by its limitations - or ask too much of the model or the modellers,” she says.
How does Australia's modelling differ from other countries?
One of the biggest public criticisms of the Doherty modelling is how it appears to predict scenarios in conflict with those that have eventuated overseas.
For example, it suggests relatively low numbers of infections and deaths once Australia reaches 80 per cent vaccination coverage with ongoing restrictions in place. Meanwhile, the UK has witnessed consistently high numbers of infections and deaths at that vaccination target.
Dr Baker says Doherty modellers have relied heavily on modelling produced overseas, which have shown Australia how vaccines have been affecting transmission and severe disease, and how new variants are spreading. The consortium worked closely with UK modellers when creating its report.
He says the different scenarios suggested for Australia reflect our unique pandemic experience.
“Because of the strong suppressant we’ve had over the last 18 months, we’re never going to see the level of deaths and severe cases the UK has had," he said.
“None of the policies that we even modelled or even considered in Australia for when we get to 80 per cent match up to what the UK has done. They had essentially no restrictions at around 80 per cent double dose.
“We still have pretty stringent capacity restrictions and density requirements that are still going to be in place at 80 per cent. I expect there will be a lot less [infections and deaths] than what we’ve seen overseas just due to those continuing restrictions.”
Dr Baker says the scenario in New Zealand is much more comparable to Australia, where modelling is also similar.
Prof Bennet agrees it’s hard to compare the context of Australia to that of the UK.
“Comparing a country three times the size of Australia in terms of population and a fraction of the size geographically is a difficult direct comparison to make.”
Norway and Denmark make easier comparisons, she says, both being the approximate size of Greater Sydney in population, with similar daily case numbers. Yet they have only a fraction of people hospitalised with the virus by comparison.
Both countries have just opened up, with Norway at around 70 per cent double dose vaccination and Denmark at around 75 per cent, according to Our World in Data.
Prof Bennet says Australia’s experience of opening up at 80 per cent will “probably be somewhere in the middle” between the UK and Norway and Denmark.
“And that’s right where the modelling put us," she said.
What does modelling tell us about Australia's future?
Dr Baker says there is one “absolute certainty” in Doherty's modelling.
“If Delta’s here and we can increase vaccination coverage, we will need less restrictions to keep things under control," he said.
Doherty sensitivity analysis released on 18 September says modelling showed “the epidemic came under control” when vaccination coverage increased beyond 80 per cent.
But at that point, there’s still a lot of work to do, Dr Baker says. From a modelling perspective, the response at 80 per cent and 90 per cent is still being worked out, particularly when it comes to international borders and contact tracing.
“When we’re no longer trying to extinguish every transmission chain, we don’t need to trace every single contact of every single case,” he said.
Dr Baker believes the vaccines’ effectiveness and Australia’s escalating rollout “should offer us a lot of hope”.
Prof Bennet says it’s hard to model too far into the future because the parameters fed into models will continue to change.
She agrees that one of the best things about models is the hope and reassurance they offer.
“If people see the control and how that was predicted in the model, that reassures people that the effort has been worth it," she said.
“People can also look at the worst-case scenario of the modelling, which tells us about what has been prevented. Modelling tells us a bit about what could have been.”
See more in the Vaccine in Focus series here