Emerging from lockdown:

modelling, outputs and assumptions

1
Around the world, countries are struggling with the pandemic.
Some have never emerged from their first wave
14-day new case average
• COVID-19 has caused 26 million global infections and 1 million deaths
• Many countries have never emerged from their first wave
• Early action in Victoria enabled control to be quickly regained

COVID-19 has had a tragic impact on people around the world. To date
over 26 million have been infected. Almost a million have died.

In many countries the virus was allowed to spread for weeks, or even
months, before containment measures were put in place. Stratospheric
levels of demand for intensive care occurred and much could not be met.

In Victoria, cases peaked at just 104, of which 48 were acquired through

local community transmission. Some of these people will live with
permanent disability arising from COVID-19. 20 lives were tragically lost.

Modelling published in April showed the much more catastrophic

outcome that would have happened in the absence of intervention.
58,000 Victorians would have been infected on the worst day of the
pandemic. 10,000 would have needed intensive care. Many would have
died.

Source: Our World in Data (2020).

Note: These countries have increased their testing capacity, meaning their first and second waves are not perfectly comparable.
Other countries have emerged from their first wave but exited restrictions
too quickly, and are now experiencing resurging infections
14-day new case average
• Victoria is one of many countries now experiencing a second wave
• Most have arisen from countries exiting restrictions too quickly
• Many countries that aggressively suppressed avoided a second wave

Victoria has recently experienced a second wave of infections. Outside

Australia, New Zealand, and Taiwan, which have pursued aggressive
suppression and elimination strategies, these have been common.
Second waves have typically occurred following premature easing of
restrictions, with jurisdictions underestimating just how challenging it is to
maintain control with even low numbers of infections.
Examples of resurgence include:
Israel, which gained control over new COVID-19 outbreaks after their first wave in
April. But schools were opened on May 17, which soon led to outbreaks in
classrooms that spread to homes.
Spain, which used stringent stay-at-home orders until May. In June, restrictions were
removed, and international tourism was encouraged. This soon led to a second wave.
European countries have since banned travel to Spain. Its hospitals are reaching
capacity again. And provinces have just been given the power to re-enter lockdown.
France, which after strict lockdowns in March and April is now facing a second wave.
Masks are being made compulsory. Some schools remain closed. Local lockdowns
are being considered. A national lockdown has not been ruled out.
Source: Our World in Data (2020).
Note: These countries have increased their testing capacity, meaning their first and second waves are not perfectly comparable.
Stage 4 restrictions enabled Victoria to turbo-charge its exit from Wave 2

• Less stringent Stage 3 restrictions have proven ineffective in Victoria

• Moving to Stage 4 more than doubled the rate of decline in cases Victorian 14-day new case average
• Staying in Stage 4 longer means total exit from restrictions sooner
Stage 4
introduced
Thanks to the incredible sacrifices made by Victorians during Stage 4
restrictions, we are beginning to suppress the spread of coronavirus.
Stage 3 restrictions helped slow the growth of coronavirus cases, but
even with masks, cases were taking 49 days to halve.
The introduction of Stage 4 restrictions helped us speed things up, and
cases are now halving every 18 days: more than twice as fast.
Keeping Stage 4 restrictions until case numbers are low enough to safely
reopen will enable all Victorians to get back to COVID-normal, faster.
Victoria will not be in a safe position to re-open in mid September

• On September 4th, we had a 14-day average of 116 cases (1,624 total)

• In mid-September, we will have an average of 63 cases (882 total) Victorian 14-day new case average
• With so many cases in the community, re-opening at this point will risk
a resurgence, undoing all of the gains achieved from lockdown.

Coronavirus can quickly get out of hand, and the national strategy to
make sure we do not have 1,000s of daily cases is to suppress
community transmission.
University of Melbourne modelling finds that it is unlikely we will have
aggressively suppressed the virus by mid-September. Victoria had a fortnightly average of 115
new cases on Sept-4 (1,624 over 14-days)
Based on current levels of social distancing, the 14-day case average is
likely to be around 60 cases by mid-September. By contrast the worst
fortnight that NSW has experienced outside of Stage 3 restrictions was
UniMelb projects that Victoria will typically
13 domestic cases per day on average.
have a fortnightly average of around 63 new
If restrictions are eased while the virus is still circulating widely in the cases on Sept-17 (882 over 14-days)
community, there is a real risk that infections will rebound – causing 95% interval

restrictions to be reimposed and last much longer.

NSW’s worst fortnight without Stage 3
restrictions was 13 domestic cases
(179 over 14-days)

Note: The UniMelb DPM model begins on the 3rd of September, and 14-day averages are only available from the 17th.
Aggressive suppression is our best bet for avoiding a yo-yo effect
A yo-yo effect is where lack of control is achieved, causing restrictions to be continuously lifted and reimposed. The University of Melbourne model suggests
that if we ease restrictions when there is a fortnightly daily case average of 25, there is a 6 in 10 chance of having to lock down again before Christmas.

Ultimately, a wide range of different scenarios could play out over the coming
months in Victoria. Our exact path will depend on policy decisions, how well
Victorians can follow public health advice – and luck.
There are strong elements of randomness in how SARS-CoV-2 spreads
throughout a community. One person who is infected with the virus might be very
infectious to others, for a long time, and have lots of contacts before they are told
to isolate. Another might have few contacts or be less infectious.
Running a large number of model simulations tell us what is most likely to occur.
In 640 out of 1,000 model simulations, reopening too early (at 25 cases per
day over the fortnight, on average) causes a yo-yo effect.
The below graph shows just 1 these 640 scenarios. The cases fall and
restrictions are eased slightly, then significantly from when the fortnightly
average case numbers hit 25 cases per day (350 cases total). Cases soon
start to rise, and restrictions need to be tightened again before Christmas to
avoid a large third wave that overwhelms the health system.

Indicative simulation of easing at a fortnightly average of 25 cases a day

14-day new case average
Restrictions eased partially in Restrictions tightened Restrictions eased
lead up to 25 case average, at 25 cases
then significantly at 25 cases
Appendix: the model and its authors
Authors
Melbourne University’s Dynamic Policy Model (DPM) is the result of an
extensive international collaboration among a multi-disciplinary team from
Australia and New Zealand over many months.
Primary contributors are:
 Dr Jason Thompson from Melbourne University’s Transport, Health and
Urban Design Research Lab,
 Professor Mark Stevenson from Melbourne University’s Transport, Health
and Urban Design Research Lab,
 Professor Tony Blakely from the Population Interventions Unit at Melbourne
University’s School of Population and Global Health.
 Professor Rod McClure from University of New England’s Faculty of
Medicine and Health.
Contributors in the project have generously given their time pro-bono to DHHS
to provide outputs from their model.
1 Blakely et al (2020) The probability of the 6-week lockdown in Victoria (commencing 9 July 2020) achieving elimination
of community transmission of SARS-CoV-2. Med J Aust 2020;In press. https://www.mja.com.au/journal/2020/probability-
6-week-lockdown-victoria-commencing-9-july-2020-achieving-elimination
The model
The University of Melbourne’s agent-based dynamic policy model (DPM) for COVID-19
imagines a simplified world where people (agents) move around like pieces on a chess
board. Each person has their own characteristics. Some are old, some are young,
some go to work and some go to school. Some are very infectious when they get
COVID-19, and some are not.
If a person moves into the same square as another person who has COVID-19, they
may catch the virus. People can reduce their risk by avoiding other people, keeping
1.5m distance or wearing a mask. If a person becomes infected and is traced by the
health system, they are isolated and are less likely to infect others. As greater (or
lesser) restrictions are imposed by the DPM, people change the way they move around
the chess board – Following restrictions, some may stay at home more, or deliberately
try to avoid interacting with others. If case numbers decline and restrictions are
loosened, agents' mobility and interactions increase.
Agent based models are used throughout academia to model phenomena as diverse
as economics, transport, and infectious diseases.
The DPM has parameters that are based on the disease mechanics of COVID-19, and
have been validated against Australia and New Zealand’s first and second wave of
infections. This means that the model is helpful in predicting more likely outcomes of
changes in social and health policies related to social contact and therefore, disease
transmission.
This model has been peer-reviewed and published in the Medical Journal of Australia.1
A detailed set of parameters is available online here. It continues to be updated and
enhanced as the pandemic progresses.
Appendix: Detailed parameters of the model
The model uses assumptions about how coronavirus spreads based on the academic literature and experience in Victoria. These assumptions will necessarily change over
time as we learn more about the virus This model has been peer-reviewed and published in the Medical Journal of Australia. A detailed set of parameters is available online
here.
Model parameters
Physical distancing – decay over time
Mask wearing (% of population compliant)
Mask efficacy
Incubation period distribution (days)
Illness period distribution (days)
Infectiousness
Mean adherence with isolation of infected
cases (% beta distribution (28,2))
Asymptomatic cases (% of all cases)
Infectiousness of asymptomatic cases vs
symptomatic cases
Target peak effective reproduction number
(Rt) across model runs
How often are positive cases told to isolate?
Public compliance with isolation orders
How infectious are children?
Model estimates
Social distancing decays by 1% per day at the beginning of each easing phase, with a floor of ~15% below the initial level for each phase (based on
observed behaviour found in traffic statistics in Victoria)
90% of people comply out of home when compulsory
50% of people comply out of home when recommended but not enforced in law (assumed parameter based on expert opinion)
~20-25% reduction in transmission per close out of home close contact (based on confidential epidemiological advice from external epidemiologists from
their upcoming paper)
Log-normal distribution with mean 5.1 and sd = 1.5 (Lauer et al., 2020)
Log-normal distribution with mean = 20.8 and sd = 2 (Bi et at., 2020)
An individual’s infectiousness is drawn from a uniform distribution between 0 (not infectious) to 100 (very infectious). Their infectiousness increases during
their incubation period then decreases to zero during their illness period (Chang et al., 2020)
Mean = 0.93, sd = 0.05 (assumed parameter based on expert opinion in conjunctions with available public data sources)
Mean = 33%, sd = 3% (He et al., 2020; Pollán et al., 2020).
33% (He et al., 2020)
The average infected person in the model infects about 3 (2.6 – 2.7) other people by the last day of their infection.
There is a 25% chance per day that any infected person will be traced (75% they won’t) by the tracking and tracing system - 80% of people will be traced by
day 5/6.
93%, sd = 0.05 (beta distribution (28,2)) (assumed parameter based on expert opinion in conjunctions with available public data sources)
Children under 10 years are 50% less infectious than adults. Children over 10 years have the same infectiousness as adults. (assumed parameter based on
expert opinion in conjunction with available public data sources. Park et al., 2020; Pollán et al., 2020)
Appendix: Models should be used as a guide
Modelling a pandemic is challenging and there is
Modelling a pandemic is challenging. The further out estimates are made, the more uncertainty there is about the outcome. With more time and more data to inform the forecast,
uncertainty about the outcome
assumptions can be either confirmed or altered to create a more accurate short-term forecast. While every effort has been taken to reflect the societal, epidemiological and policy
settings in Victoria, as with all modelling there are limitations. Significantly this model can measure the number of new daily cases – but not the number of unknown source cases.

What the model is not currently specified to tell us

• The number of unknown source cases (community transmission or ‘mystery cases’) which should be core to the decision of whether to ease restrictions.
• The differences in infection rates in geographic areas – including high-risk LGAs and low-risk regional areas.
• The relative risk of catching coronavirus from going to work in certain industries (e.g. abattoir and healthcare workers).
• How the weather might affect transmission risk.
• When a vaccine will be available.
• What the risk of new cases arriving from overseas is.
• Fine details about the testing and tracing system.
• Details about differences in demographic risk outside of students and essential workers.