COVASIM – an individual-based model assessing the impact of easing COVID-19 restrictions.

When COVID-19 first reached Australia, Federal and State Governments implemented a series of behavioural control measures, including physical distancing and isolation/quarantine to reduce virus transmission. This approach has been highly successful.

As control measures are relaxed across Australia, care and vigilance is needed to limit the real risk that COVID-19 cases could rapidly rise again.

Hence it is vital that governments have high quality precise information about the likely impact of relaxing various control measures, and the time required to monitor the impact of relaxing these measures. It is also important to understand the likely impact of interventions on reducing transmission in the community.

More about COVASIM

Modelling the Victorian Roadmap | 19 September 2021

Burnet’s COVID-19 mathematical modelling was commissioned by the Victorian Government to inform the Victorian Roadmap.

The Roadmap has been developed based on expert modelling from the Burnet Institute and is set against COVID-19 thresholds including hospitalisation rates, and the vaccination targets already set out in the National Plan to transition Australia’s National COVID-19 Response.

“The modelling helped the Victorian public health teams get a picture of what our hospitalisation rates could look like while cases are still rising and develop trigger points to indicate if the system is becoming overstretched – allowing time to implement further health measures and protect it from becoming overwhelmed,” Premier Andrews said.

“The Burnet modelling also shows that the key to opening up and reducing risk in Victoria will be making sure workers across the state are vaccinated.”

Modelling the Victorian roadmap

Since July 2021 Melbourne has experienced a resurgence in delta variant COVID-19 cases. Despite a lockdown being introduced on 5 August, cases continue to grow, and at 17 September daily diagnoses have reached a 7-day average of 454.

With Victoria’s COVID-19 strategy shifting away from COVID-zero, protecting the health of the population will require achieving high vaccination coverage as quickly as possible, maintaining control of the epidemic to protect the vulnerable, and ensuring that the health system has capacity to provide care to all who need it. An important question is: as vaccine coverage increases, how best can restrictions be eased that prevents health system capacity from being exceeded?

The COVASIM model was used to simulate options for easing of restrictions over the October-December period. Model inputs included data on demographics, contact networks, workforce composition, contact tracing systems and age-specific vaccination rates. As well as options for easing restrictions, additional policies around vaccine allocation and testing were examined to determine potential approaches to further reduce the epidemic peak.

Scenarios were run to estimate the number of COVID-19 infections, hospitalisations and ICU requirements in Melbourne:

• Maintained lockdown: A counterfactual scenario to set baseline estimates from which restrictions are eased.

• Roadmap: School and childcare returns throughout October; increased outdoor activities at 70% two-dose vaccine coverage (people 16+ years); retail and indoor activities with density limits commence at 80% adult vaccine coverage; and mandatory vaccination of authorised workers, teachers, childcare workers, parents of children in childcare, hospitality workers, hospitality patrons.

• Roadmap with additional testing: The roadmap scenario but assuming vaccinated people continue to seek symptomatic testing at the same rate as non-vaccinated people, even for mild symptoms.

• Roadmap with a 15% reduction in non-household transmission. The roadmap scenario, but with an assumption that a 15% reduction in non-household transmission could be achieved immediately and sustained.

Key findings

• Even without any easing of restrictions, there is a moderate risk of exceeding health system capacity

• Based on the current epidemic growth rate, a peak in 7-day average daily diagnoses of 1400-2900 is estimated to occur between 19-31 October

• Corresponding peaks in hospital and ICU demand were 1200-2500 and 260-550 respectively, with 24% of simulations resulting in hospital demand exceeding 2500 beds.

• In the roadmap scenario, the significant easing of restrictions at 80% vaccine coverage led to 63% of simulations exceeding 2500 hospital demand, and resulted in a second epidemic peak over mid-December.

• High rates of symptomatic testing among people who are vaccinated could reduce the impact on the health system In a scenario with vaccinated people testing at the same rate as unvaccinated people, the risk of >2500 hospital demand was reduced from 63% to 29%. However, this may be difficult to achieve in practice.

• If a 15% reduction in non-household risk could be achieved and sustained through a variety of additional targeted public health and testing interventions, the risk of >2500 hospital demand could be reduced to 18%.

• When 80% adult vaccine coverage is reached, the case numbers, hospital and ICU numbers can provide a guide as to the likelihood of the health system capacity being exceeded and whether restrictions can be safely eased consistent with the roadmap or whether a more staggered approach may be required.

• Due to uncertainty about whether the epidemic growth rate will be sustained, seasonal impacts and vaccine efficacy parameters against the delta strain, updated projections are required as more data becomes available.

Decisions to ease restrictions should be based on the latest epidemiological and health system information.

Image: Figure 4: Roadmap scenario. Includes schools returning to in person learning throughout October; childcare returning and mobility restrictions easing in October; limited outdoor gatherings at 70% two-dose vaccine coverage among people 16+ years; indoor gathering with density limits at 80% two-dose coverage among people 16+ years (Table 2 and Table 3); and mandatory vaccine requirements. Dashed vertical lines represent estimated dates of reaching 70% and 80% two-dose coverage among people 16+ years.

Model assumptions

Models make simplifying assumptions to approximate the real world, particularly where data are not available. Some of these assumptions may lead to the model projections being optimistic or pessimistic compared to what may actually occur. For example, compliance with vaccine mandates in Australian settings is as yet unknown; in the roadmap scenario 95% compliance has been assumed, but the roadmap may be slightly optimistic depending on how successfully it can be implemented. To best interpret the model outputs, it is useful to understand some of the main assumptions that may make these projections optimistic or pessimistic.

Optimistic assumptions

The results could be optimistic (meaning the real world will be worse than estimated) because we have assumed:

• Schools and childcare can achieve a 50% reduction in transmission risk through ventilation and other mechanisms
• No waning of vaccine immunity over time
• No quarantine or testing exemptions have been included for vaccinated people (i.e. vaccinated people continue to be required to quarantine for 14 days if they are identified as contacts)
• Compliance does not further reduce over time (33% of people are assumed to have had between household contacts in the current lockdown / model calibration period)
• 95% compliance with vaccine mandates
• Schools and childcare are able to conduct their own contact tracing
• Vaccines are delivered equally across all sub-population groups. It is possible that people who are more concerned about COVID-19 and are minimising their number of contacts to lower their COVID-19 risk may be getting vaccinated before people who and less concerned about COVID-19 and are at higher risk.

Pessimistic assumptions

Conversely, the results could be pessimistic (meaning the real world will be better than estimated) because we have assumed:

• No impact of seasonality, when it is possible that warmer weather may reduce transmission (but unquantified at the moment).
• The current epidemic growth rate will continue (with the exception of declines due to vaccine immunity), when it is possibly biased by recent infections being concentrated in communities with below average vaccine coverage.

Uncertain assumptions

In addition, the results could be either optimistic OR pessimistic because:

• Average duration of stay in hospital and ICU is unknown. If it were longer or shorter than we have estimated (e.g. average 11 days in ICU, see appendix) then this would increase or decrease peak demand.
• Vaccine efficacy assumptions may be better or worse than the parameters we are using (Table 1), but are based on best estimates at the time of analysis.

Limitations

The findings presented are derived from an individual-based model, which is an imperfect representation of the real world.

• Results are based on model inputs up to 17 September 2021. As the outbreak evolves and more data becomes available, the uncertainty reduces and it becomes clearer which trajectory we are on.
• There is uncertainty in the average length of stay in hospital and ICU, and this would impact estimates of peak hospital and ICU demand.
• Results do not include seasonal effects, which are unknown.
• Results do not include reduced compliance with restrictions over time. In particular, towards the end of our projections, we have assumed that testing, contact tracing and quarantine continues despite high vaccination coverage, which may overestimate the effectiveness of this system if people are less compliant with QR sign in and other
• This model currently only attributes basic properties to individuals, specifically age, household structure and participation in different contact networks. The model does not account for any other demographic and health characteristics such as socioeconomic status, comorbidities (e.g. non-communicable diseases) and risk factors (e.g. smoking) and so cannot account for differences in transmission risks, testing, quarantine adherence or disease outcomes for different population subgroups.
• The model does not include a geospatial component and so cannot capture geographic clustering of vaccination or infection within some communities.
• The model simulates symptomatic testing by having a parameter for the per day probability of being tested if symptoms are present. This means that the distribution of time from symptom development to testing is binomial, which may differ from the true distribution of time from symptom onset to testing.

Model parameters are based on best-available data at the time of writing. Results from new studies could change estimates of social mixing, contact networks, adherence to policies, quarantine advice, and disease characteristics (e.g. asymptomatic cases), and these could change these results.

Download the Burnet Institute VIC Roadmap Modelling

COVID-19 Mathematical Modelling of resurgence risk: | 26 Sept 2020

Estimating risks associated with early reopening in Victoria

Authors: Dr Romesh Abeysuriya, Dominic Delport, Professor Margaret Hellard AM, Dr Nick Scott. Funding: Commissioned by the Victorian Department of Health and Human Services.

Following the introduction of Stage 4 restrictions in Melbourne, daily new detected cases of COVID-19 have been declining. Accordingly, a roadmap detailing possible sequences of policy relaxations has been proposed to return to a “COVID normal”, together with criteria for triggering each step. Due to the high social and economic impact of the restrictions currently in place, it is important that restrictions are relaxed as quickly as possible. However, relaxing too quickly increases the risk of a resurgence in infections, which may then require a reintroduction of restrictions to contain.

In this study, we use COVASIM to estimate the risk of Victoria experiencing a third COVID-19 epidemic wave if Stage 4 restrictions were eased on the 14th September 2020 or two weeks later on the 28th September.

In both scenarios, restrictions were eased to a level of restrictions similar to Victoria in early June (pre-Stage 3), approximately the “final step” in the Victorian government roadmap or NSW in September. Specifically we modelled:

• Schools, childcare and workplaces reopen
• Cafes, restaurants, pubs, bars, entertainment venues, and places of worship all open with a four square metre distancing rule
• Community sport and small social gatherings are allowed
• Test results take 24 hours to become available
• Contact tracing takes an additional 24 hours following test results, and includes use of the COVIDSafe app
• The number of tests per day is increased to maximum capacity observed in June upon easing
• Large events are banned and mandatory masks are maintained.

While there are a wide range of options for incremental relaxation, in this study we sought to specifically examine the impact of timing, to examine the relationship between the degree of containment prior to relaxation and resurgence risk.

Conclusions and recommendations

Overall, our results suggest that Victoria would not have been able to safely return to NSW-level restrictions on 14th September, and there would be a high risk associated with lifting all restrictions at once on the 28th September.

Download the COVASIM Modelling of resurgence risk.

Projected epidemic outcomes for COVID-19 strains against different vaccine rollouts | 11 June 2021

Estimating impacts of a COVID-19 outbreak without public health interventions, after vaccines have been administered

Authors: Dr Romesh Abeysuriya, Professor Margaret Hellard AM, Dr Nick Scott.

Click to view a larger version of the graph.

Model scenarios have been run, calibrated to Victoria, Australia, to help answer the question: What is the impact of different levels of vaccine coverage, if public health control measures were stopped and the virus was allowed to spread through the community?

Burnet Institute has developed an Excel-based tool that summarises thousands of simulations of different scenarios. In each scenario, new infections (one per day) begin to be introduced to the Victorian community at some point following the commencement of vaccine rollout. The vaccine rollout is assumed to continue at a fixed rate with increasing coverage every week.

The tool can compare outcomes when different COVID-19 strains are introduced, and vaccine efficacy assumptions are varied. Before using the tool or interpreting outcomes it is critical that the following key points and examples are read and understood. For additional information, or advice in interpretations, please contact the authors.

Critical points for understanding these projections:

• The scenarios assume a user-defined vaccine rollout speed of either 150,000 or 250,000 doses per week in Victoria (75,000 or 125,000 vaccinated people per week, due to second doses). The results are different if the rate of vaccine rollout is different.
• The scenarios do not currently include any major public health response to gain control of outbreaks. On detection of the first case, the model assumes symptomatic testing increases (isolation of positive cases continues), masks become recommended but not mandatory, and contact tracing continues but only up to 250 diagnoses per day. Hence the projections represent hypothetical near-worst-case scenarios.
• The results are based on a collection of model assumptions about the contacts of individuals and disease transmission dynamics . If these best-estimate assumptions are optimistic or pessimistic, then compared with these projections actual epidemic outcomes will be more optimistic or pessimistic respectively.

One scenario created by Burnet Institute Head of Modelling, Dr Nick Scott and colleagues assumed a 50 per cent vaccine efficacy in preventing infections and a 93 per cent efficacy at preventing deaths among people who did become infected; a virus which was 1.5 times as infectious as the one in Victoria in June-November 2020; and where 80 per cent of people aged over 60 and 70 per cent of people younger than 60 years of age were eventually vaccinated.

“We found that if the virus enters the community when 60 per cent vaccine coverage has been reached and is left unchecked, we could see 4,885 deaths in Victoria within a year if no public health responses are introduced,” Dr Scott said.

“If we get peak vaccination coverage up to 95 per cent, the number of deaths reduces to 1346.”

Conclusions and Recommendations

• Vaccine hesitancy and the emergence of new COVID-19 variants mean Australia is unlikely to achieve herd immunity
• Public health initiatives remain vital in controlling COVID-19, even in vaccinated populations. Without public health measures, thousands of Victorians would be hospitalised and die if an initially small outbreak was left to spread through the community unchecked
• Australia requires higher vaccine coverage to return to normal life.

WHAT IS COVASIM?

The Burnet Institute and the Institute for Disease Modelling in the USA has developed a unique individual-based COVID-19 model (COVASIM) that can assess the impact and risk associated with relaxing various physical distancing policies on the resurgence of COVID-19.

It has already been applied to a number of high, middle and low-income settings, including a number of states in the USA and countries across Africa. The individual-based simulation model can be applied to all Australian jurisdictions.

It provides governments with more specific and precise data to inform their COVID-19 responses.

EARLIER PUBLISHED FINDINGS FROM COVASIM

N Scott, A Palmer, D Delport, R Abeysuriya, R Stuart, C Kerr, D Mistry, D Klein, R Sacks-Davis, K Heath, S Hainsworth, A Pedrana, M Stoove, D Wilson, M Hellard.
Modelling the impact of reducing control measures on the COVID-19 pandemic in a low transmission setting (In press – MJA) Accepted September 2020. Online 2 Sept.

The COVASIM model assessed the impact and risk associated with relaxing various physical distancing policies in Victoria, Australia at the end of the first COVID-19 wave. A key finding of that work was that relaxing restrictions too quickly could lead a considerable resurgence of COVID-19 in the community if there was failure to detect early clusters of infection.

Visit the Know-C19 Hub for more policy briefs and reports from the Know-C19 team.