The real world has a habit of confounding even the most meticulously crafted models.
On Monday, researchers from a consortium led by the Doherty Institute held a media briefing regarding updated “sensitivity analyses” of its original Covid-19 modelling.
There were two major changes: figures for a much larger active caseload were included at day zero of the model to account for relaxation of restrictions in the middle of an outbreak and incorporating less optimistic assumptions about test, trace, isolate and quarantine capacity. What the team has yet to do, but signalled is on the cards, is to update the model to reflect the increased severity of the Delta variant regarding hospitalisations, intensive care admissions and death.
In its first round of work, prepared for government and released in early August, the Doherty team based its model on an outbreak of 30 people and ran dozens of scenarios from this starting point to track how community transmission might behave over a six-month period. So much of the national plan to begin opening up between 70 and 80 per cent double-dose vaccination coverage was predicated on this model.
But what if there are thousands of cases in the community instead, as was already the case in New South Wales and now in Victoria? How would an outbreak spread then?
“So we undertook a sensitivity analysis where we seeded outbreaks at 70 and 80 per cent with hundreds or thousands of [active] cases,” Professor Jodie McVernon said during the press conference on Monday.
“And when we say hundreds it was 300 to 1000 cases and for thousands it was 1000 to 4500. We then said, if we start with this many cases, do we kind of break the assumptions? Overall our conclusions were robust.
“However, there was one important distinction which was at the 70 per cent [vaccination] threshold, starting off with thousands of cases meant that there was the opportunity for cases to grow more rapidly in that window, and that led to what we call overshoot. The ultimate size of the epidemic was bigger, in that situation.”
To prevent that from happening, vaccines need to be boosted by “medium restrictions” – including stay-at-home orders – until at least the 80 per cent vaccine threshold is reached. In NSW that is expected about October 17. In Victoria it will be about November 5.
“So the logical conclusion – and this maps back to our earlier recommendations – was that jurisdictions should look at the situation they’re in, they should look at how their infections are being controlled with the measures they have in place,” McVernon said.
“Those who are coming into the 70 per cent mark with thousands of infections should aim to keep case numbers as low as possible by continuing to suppress epidemic growth.”
What is clear from the Doherty update is that its initial assumption – that optimal test, trace, isolate and quarantine measures could be achieved as cases climbed from a low starting point – simply could not hold at a rate of many hundreds or more than a thousand cases a day.
“At these higher case numbers, it’s also recognised that test, trace, isolate, quarantine (TTIQ) responses are likely to be partially, rather than optimally, effective,” the Doherty team said in a statement released alongside the model update.
In the national common operating picture – a sort of weekly dashboard monitoring every jurisdiction’s handling of Covid-19 outbreaks – the most recent data shows a devastating breakdown in the ability of NSW Health staff and partners to keep on top of a soaring workload.
For the week ending September 16, just 68 per cent of positive Covid-19 cases in the state were told of their results within one day of the test collection. This is the first time in this outbreak that the reported figure to the dashboard has been below 100 per cent. Similarly, the rate at which close contacts are notified has dived. Precisely how bad things have become is difficult to assess because, for the first time in this Delta outbreak, this data was not provided to the monitoring dashboard by authorities.
There is an opportunity in the future to completely redesign how the TTIQ system works. That might mean abandoning some of the less “efficient” measures that made sense when chasing Covid zero but offer less “value” in a transition to suppression. Members of the Doherty consortium have started this work, but it is too early to report in detail.
“For a lot of states, contact tracing has not been about reducing transmission per se, it’s been about reducing the probability that there is one leftover case out there and that’s a very different situation,” Professor Nick Golding, of the Telethon Kids Institute and Curtin University, told reporters on Monday.
“So hundreds, literally hundreds, of contact tracers assigned to one case to run down many, many different contacts and contacts of contacts… those are all things that are very labour intensive.
“While they help with the aim of getting to zero, that is not an efficient use of resources for when you just want to get the majority of people into isolation before they can start spreading. If you can accept that you can miss a few, then you can target your resources to get far more people into isolation earlier.”
McVernon is more direct: “We know we can’t keep tracing at that level forever.”
In its updated analyses, one of the Doherty group’s most optimistic scenarios involves 80 per cent vaccination coverage, baseline public health and social measures that mirror restrictions in Sydney in March this year, and a “low” outbreak number of between 10 and 1000 cases in the community. Under this model, which also assumed optimal TTIQ, there would be 5627 symptomatic infections, 175 hospitalisations, 31 intensive care admissions and 26 deaths over a period of six months.
Change any one of those variables, however, and the numbers fluctuate significantly. If all variables remain the same but the outbreak is seeded in the thousands, the number of symptomatic infections soars to 276,761 over six months, with 9764 hospitalisations, 1909 ICU admissions and 1678 deaths.
When asked by The Saturday Paper if the team also modelled the impact on ambulance services, a key part of managing hospital capacity, Professor McVernon said no. “We look at certain things in terms of what might happen in terms of capacity,” she said, “but we haven’t explicitly considered those services, no.”
In less than two weeks, Sydney is likely to hit its vaccination target of 70 per cent double dose for its eligible population. The health system will be at its worst point in modern history about the same time.
On Thursday, Liverpool Hospital, in the city’s west, reported there had been an outbreak of 24 Covid-19 cases across multiple wards, including in the ICU. The state recorded 1063 new daily cases and reported 1244 Covid patients in hospital, with 233 in ICU. Paramedics in the city are essentially managing a caseload of more than 10,000 people in the community, although the official figures are now blended together and difficult to disaggregate.
There is good news on the way, however.
James McCaw, a professor of mathematical biology at the University of Melbourne, told reporters on Monday that the bending of the curve in NSW is “predominantly a vaccine effect”. Victoria is starting to see early signs of a similar bend, he said, although its coverage thresholds for the vaccine are a few weeks behind NSW.
“You can see a slow and steady decline in the transmission potential in NSW,” McCaw said, “and that’s almost entirely due to the vaccine rollout.”
Curiously, the Burnet Institute – the second major outfit modelling Covid-19 for state governments in Australia – disagrees.
In a September 14 paper, which has been criticised online by other modellers and health professionals, the Burnet Institute said it was curfews and other tough restrictions in the 12 local government areas of concern in Sydney that “has worked to halt the rise in cases”.
The proof for this link, the researchers said, was a “hinge point” in the chart that measures daily case numbers. These tough restrictions were introduced on August 23, on top of other longstanding measures in the 12 council areas, and within a week the hinge point appeared.
What is not considered, however, is the priority vaccination access given to adults under the age of 40 in those same council areas a week before the new restrictions came into force. When asked about this, the Burnet Institute’s public health lead, Professor Mark Stoové, said the team’s conclusion was that “it was too early for vaccine to have had that much of an impact”.
“They have worked, there is a flattening there (of) about 9 per cent, but it just wasn’t sufficient to explain what we actually saw in relation to case numbers going down in relation to those 12 LGAs,” he said.
“It’s actually borne out by the other LGAs as well, so when we modelled the influence of vaccine coverage it actually had the same impact in all the local government areas outside of those 12 with restrictions.”
But models can run up against the real world in more ways than one. When Melbourne experienced an outbreak of Delta – unlike an earlier incursion of the less virulent Kappa variant – the state had just emerged from stay-at-home orders. Within hours of the first eight cases being identified, Melbourne residents had returned to the state’s tough restrictions, including outdoor exercise and travel limits and mandatory masks. A 9pm-5am curfew followed two weeks later.
Although these measures undoubtedly reduced the number of daily infections, they did not lead to a “hinge point” in the curve and were unable to stop the outbreak from becoming the fastest growing of any Covid cluster in Australian history. On Thursday, Victoria recorded 766 new cases, higher even than the peak recorded in last year’s second wave.
“All I can say is different cities have different outbreaks,” Stoové says. “Sometimes it is on the basis of geography, of the movement of the population, or just randomness in the differences between where cases occur. And, to be honest, there is a degree of luck here.”
Stoové ran into a firestorm of criticism last weekend when he attempted to explain a separate set of modelling conducted by Burnet, in which a worst case scenario for NSW predicted 7700 cases a day by the end of August “if nothing changed”.
Burnet, which has “disease elimination” as one of its core platforms, has loaned many of its modellers to the new OzSAGE group – a collection of researchers from various fields advocating for “an exit strategy from this pandemic with the best possible health, social and economic outcomes”. Its critics, such as former deputy chief medical officer Dr Nick Coatsworth, have called the amalgam a Covid zero smokescreen and accused it of spreading alarm.
Stoové says one of the implications of the criticism he faced last weekend was that Burnet “predetermine the policies we want to model”. He says this is not the case.
“When we are commissioned to do the modelling, we will model what the government asks us to model, and the primary interest is the capacity of the health system,” he says. “When we are doing some model, on the side, I guess, it’s more about informing community dialogue and we model things that we think are plausible responses to increasing cases.”
The sticking point for many researchers, however, is that even when using so-called stochastic models to account for random variables, human behaviour is very hard to predict with anything approaching accuracy.
“What we are not doing is plugging into our model prescriptive assumptions about how different, particular measures will affect transmission,” Professor Nick Golding, who worked with the Doherty team, says.
“I don’t think anyone, any modeller, can get a really good handle on exactly how each measure affects transmission because it is modified by human behaviour. People behave differently at different times in response to these things.”
What is much easier to measure is the effect of vaccines. When looking at both NSW and Victoria, there has been a consistent erosion of the effective reproductive rate of the virus in these outbreaks. Professor James McCaw is adamant that this is primarily driven by uptake in vaccinations in both states. When asked, his response is simple: “Yes.”
This article was first published in the print edition of The Saturday Paper on September 25, 2021 as "Covid figures: Splits emerge in government health models".
For almost a decade, The Saturday Paper has published Australia’s leading writers and thinkers. We have pursued stories that are ignored elsewhere, covering them with sensitivity and depth. We have done this on refugee policy, on government integrity, on robo-debt, on aged care, on climate change, on the pandemic.
All our journalism is fiercely independent. It relies on the support of readers. By subscribing to The Saturday Paper, you are ensuring that we can continue to produce essential, issue-defining coverage, to dig out stories that take time, to doggedly hold to account politicians and the political class.
There are very few titles that have the freedom and the space to produce journalism like this. In a country with a concentration of media ownership unlike anything else in the world, it is vitally important. Your subscription helps make it possible.
Select your digital subscription