Why is elimination of COVID-19 in Australia apparently off the table?

Recent days have witnessed the apparent success of Australia’s efforts to control the spread of COVID-19 through mass public health measures including Social Distancing, Quarantine, banning of public gatherings, travel and economic restrictions. These measures and their timing are being wonderfully catalogued and captured by Behrooz Hassani and his colleague Yuri Song here.

As can be seen above, they are also producing some tentative, ‘don’t get your hopes-up yet‘, kinds of signs that the spread of COVID-19 in Australia is coming under control, leading some obviously frustrated free-market pundits and public figures to join (arguably isolation-crazed) calls to ‘release the restrictions!‘.

However, avoiding hyperbole, David Crowe of the Sydney Morning Herald summarises the dilemma faced by policy-makers in Australia well in his Good Friday column, indicating that the Australian Government has at once rejected herd immunity but they have also not promised elimination; their objective is to “manage the deaths without overloading the hospitals“.

Writing in The Conversation and AFR, respectively, both Michelle Grattan and Phillip Coorey come to much the same conclusion; that the Government is not moving toward efforts to eliminate COVID-19 from Australia and is instead seeking a suppression pathway of unknown duration. This is in direct contrast to the New Zealand approach, which seeks to eliminate the disease quickly and completely through comprehensive lockdown measures, which appear to be showing signs of success.

Regarding the question of elimination and whether it is also a viable option for Australia, Coorey quotes Chief Medical Officer, Brendan Murphy who says,

“That is one of the options available. The issue, though, is that then you don’t have any immunity in the population and you really have to control your borders in a very aggressive way and that might be for a long time.”

In doing so, he echoes the concerns of Professor James McGraw who reasonably wonders aloud whether elimination is feasible, globally.

So the policy blockage for adopting elimination in Australia is that a) the majority of the population remains susceptible should new cases arrive, and b) we might have to control the borders for a long time? Right. But rather than being issues, could these elements not be regarded as features (or at least acceptable trade-offs)?

I’ll start with the second point first. Regarding the border restriction, this doesn’t sound either 1) impossible, or 2) politically unpopular – especially for a conservative Government given the Australian public’s general views on border security and also if NZ eradicated the disease and trans-Tasman travel was at least made available. As other countries also demonstrated that they had eradicated the virus, they (& we) could also gradually open-up to international travel again.

But staying on borders, the idea of ‘we have to open up sometime!’ seems to come up a lot. But ask yourself this – how long will it take for the current isolation restrictions to be lifted under a supression regime? 9 months? A year? Will it really be any different? Simon Birmingham doesn’t seem to think so. And wouldn’t we all prefer to at least have unrestricted movement within our own country in the meantime? But if you are really still stuck on the issue of borders, the Visa system could be adjusted (as it already has been) to suit the new reality:

How so? It would require certified non-infection before inbound flight and enforceable isolation on arrival. Just a more complex visa process, really.

— Hugh Riminton (@hughriminton) April 11, 2020

So, borders? Not really an issue that would appear to add delay or complication to the situation we already find ourselves in.

Regarding Murphy’s concern about maintaining a susceptible population, the stated and accepted goal of social and economic measures presently being taken to ‘flatten the curve‘ is so the hospital and ICU system can cope. Fair enough. But, this still means that there is an underlying expectation that elimination is off the table and herd immunity will be acheived at some (very, very far off but controlled) point in the future as the virus creeps, rather than blasts, through the population.

However, unless herd immunity is achieved, a significant proportion of the community will continue to be susceptible and illness and death will result, albeit at a lesser rate than in the ‘let it rip’ scenarios. Here, Murphy’s and Australia’s concerns about border restrictions remain unresolved, and significant social and economic restrictions will still need to continue for an unknown duration. Let’s remember – at just over 6300 confirmed cases in Australia, we still have at least 14,993,700 new cases to go….

In fact, the Government’s own modelling advice released on Tuesday suggests a timeline for peak ICU demand somewhere around November (how much we are expected to consider this modeling as a scenario exercise vs a prediction vs an education campaign, though, I am still unsure of). The contributors are totally up-front about this and I appreciate the difficulty in trying to manage the information around this, too.

However, the difficulty with the public policy advice that comes from these published curves is they automatically limit the options available to Government and able to be discussed with the public. They do so, because the the R0 value under each scenario is set in the model parameters a-priori under the assumption that implemented social and economic restriction measures are 25% or 33% effective in reducing the reproduction rate, R0 (see below). These figures are reasonably chosen on the best available evidence, which is obviously scarce because COVID-19 has never happened to us before, and when you take a look at it, actually not very best at all. But no problem, you work with the best estimates you have.

However, by pre-supposing the reduced R0 in a top-down model rather than observing it in a bottom-up modeling approach using an agent-based modeling architecture, we would consider that we are missing a potential trick and therefore, a policy option available to Government and Australia. What if the social distancing and quarantine measures are actually far more effective than is being suggested and advised? What if our policies could actually ‘crush’ the curve rather than flatten it? What if they already are? I mean, look at what actually seems to be happening… What if we put the goal of elimination back on the table?

In the remainder of this post I am going to show a few charts that are taken from 100 runs of our Agent-Based Model on policy responses available to respond to COVID-19 in Australia. You can read about the model here.

There are many differences in this type of modelling to what might be regarded as ‘typical’ SIR or SEIR models. Primarily, though, the main difference is that we attempt to build an artificial society that behaves in a way that is analogous in the most important ways our actual society works in relation to this disease. We give the society basic rules of behaviour (e.g., how much they interact, pass on viruses, travel, etc. ) then set the society free inside the model and watch what happens. The reason we run the model hundreds of times and look at average behaviours is because no two model runs are exactly the same – just like in a normal society, there is heterogeneity, randomness, good luck and bad luck, prosocial and antisocial behaviour. For more information on the approach, see here and also this description of differences, advantages and disadvantages between SIR and ABMS in disease modeling, here.

Like any modeling effort, we *** give*** the**usual* **caveats* **and* **disclaimers*** and state up front that this is not completed work – we continue to calibrate, iterate, include and exclude variables and functions. However, we invite you, the reader to not only tell us what is wrong with it, but to contribute to making it better – our model is totally transparent and available for download in html (runs a bit slow) or Netlogo. The code, the results, alternative scenarios, the lot. If you think you have a better implementation or assumption, please, don’t just snark from the sidelines, be our guest and get involved. Help make both it and our public response better. Email me if you’d like to help.

Model assumptions from one run (remember, the model can be tuned to whatever you might think is reasonable and run again) are listed below:

• 85% of people social distancing 85% of the time
• Social policy restrictions are implemented at day 65 after January 15th – around March 21
• Incubation period is 5 days +/- 1 day before being symptomatic
• Illness duration is an average of 15 days
• Identified COVID-19 cases are isolated after the incubation period has lapsed and reported at a mean of 10 days after initial infection
• 90% of people comply with isolation orders
• Age-ranges for the population are in deciles and current to Australian census data
• reinfection rate is 0
• Early stage R0 is around 3.0 – it is not predetermined but calculated dynamically as the population infection evolves (see below), dropping over the course of disease progression in the community and policy implementation
• Superspreaders (10% of the population) exist prior to travel restrictions being implemented
• People have an average number of close contacts per day of around 0.75 pre policy implementation and 0.25 post policy
• The disease transmission rate is 50% per close contact
• People are potentially infectious through the period of their illness
• If hospital beds are available (max 65,000), patients can be quarantined
• If ICU beds are available, they are allocated (max 4200) (this might be greater now irl)
• 5% of people who become infected require ICU

OK – The first chart I am going to show is an estimate of how long, if we continue with implementation of our current restrictions, it might take to get to zero new daily cases.

As you can see, as we move into August the mean new cases really bottoms out, with almost no new cases being recorded into September. A quick check against Australia’s actual daily case count here, shows you that even our figures are slight over estimates of current official numbers.

In the next chart, we estimate COVID-19 related ICU bed demand over time. This broadly mirrors the new case demand.

Next we estimate active infections – so those people who are current infected in the model and have not recovered or died.

The average number of days post policy implementation of daily case elimination is 89 (SD, 27.2 days), or a date of June 19th (with a big margin of error of about a month either side of that and a bit more).

The last graph I want to show is the R0 (or probably more accurately Rt value) over the course of the simulation. ( I can’t give you a pretty graph today, but here’s a sample from a single, typical run). You can see that it is highly variable at the beginning of the simulation before dropping significantly in the latter stages. It records null values when there are no cases, so the timeline is not accurate – don’t worry, I’ll update this in the next little while when I output the next run.

Our results are broadly in line with those offered by the University of Sydney’s Complex Systems Group headed by Prof Mikhail Prokopenko, despite both being developed totally independently. Interestingly, they also used an agent-based architecture and concluded that there was a real possibility that cases could be eliminated in around 100 days if we all stick to the plan. Maybe both our models are wrong? Time will tell.

Summary

So, I offer these results in good faith and to open again the possibility of elimination of COVID-19 in Australia as a policy option. This most closely resembles the Endgame C scenario proposed by the Grattan Institute in March. On our reckoning, though, and with continued effort, we might already be on the way there.

We have a real opportunity to push to zero quickly. Rather than easing off, why wouldn’t we take it?