Agent-based modelling: lessons for this and future pandemics

WE have navigated the coronavirus disease 2019 (COVID-19) policymaking for three years now – no one needs to be told that making decisions in a pandemic is complex. As our available options for responding to severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) outbreaks increase, so too do the public health questions requiring clear answers. Which age groups should be targeted for ongoing regular COVID-19 vaccination? How quickly should we bring back restrictions if infections in the community are increasing? What should our mask guidelines look like in Australia moving forward?

Addressing these questions without knowing what is around the corner is a difficult, and inherently uncertain, task. This is where modelling frameworks help – they allow us to quickly weigh up the advantages and disadvantages of a large number of choices in parallel.

We recently published the results of a comprehensive agent-based model in The Lancet Regional Health – Western Pacific (here), which estimates the health and economic consequences of over 100 potential policy packages implemented in Victoria in the 12 months from October 2022. These policies packages included combinations of:

• higher or lower stringency of lockdown restrictions to SARS-CoV-2 morbidity in the community (ie, a more or less strict approach to responding to infection waves);
• the provision of N95 masks by the government during large outbreaks to replace existing cloth or surgical masks;
• health promotion campaigns to increase overall mask wearing compliance during large outbreaks; and,
• 13 hypothetical vaccination schedules, targeting two different age groups with two future COVID-19 vaccines.

The model also considered the possible emergence of new SARS-CoV-2 variants, each with varying levels of immune escape and ability to cause severe disease. Modelled combinations of potential future variants and policy options totaled over 1000.

Our analysis uncovered several key insights that will be useful to Australian health professionals in discussing measures the public can take to reduce their risk and what to expect in the coming months.

More stringent public health and social measures may continue to be necessary, especially if a new variant is highly virulent

In our model, public health and social measures, which determined how many people were staying home, wearing masks, avoiding contact with others and so on, were dynamic. This means that the imposed level of restrictions increased as a function of health system pressure, with two levels of stringency. Over 12 months, the stricter of the two options (ie, the option to move to higher levels of restrictions more quickly during an outbreak) reduced infections and deaths by a quarter, on average, but predictably increased economic costs to society. Despite this trade-off, however, we consistently found that stricter policies for escalating restrictions performed better once we considered costs (including gross domestic product loss from lockdowns), illness, pressure on hospitals and deaths. The model had five stages of public health and social measures – if we consider stages 4 and 5 to be what we knew as lockdowns, we are talking about four to 25 days per year in lockdown if a highly virulent variant emerged (e.g. like Delta) under stricter policies with ongoing optimal levels of vaccination compared to zero to seven days per year under the more relaxed policy setting with ongoing optimal levels of vaccination, and substantially less for a less virulent variant. So, nothing as bad as 2020 and 2021 – with high vaccine-derived and hybrid immunity, now much of the benefit of stricter public health and social measures can be realised at the lower end of the scale (eg, density limits and encouraging people to work from home during surges).

Ongoing vaccination with high levels of uptake is key

We already know that it is important for people to stay up to date with their COVID-19 vaccinations. Our modelling supports this. To fight waning immunity following vaccination, we found that the best approaches generally involved vaccination of people older than 30 years with good coverage (75% for people >60 years old and 50% for people aged 30–59 years) and two doses within a year. In saying this, the marginal benefits associated with vaccinating 30–59-year-olds as opposed to people aged >60 alone were relatively minor. Because vaccination has additional benefits such as less time under restrictions, the vaccination schedules we modelled tended to be cost-saving from an economic perspective, so there is economic benefit to be gained here too.

Waiting until large outbreaks have started before wearing masks leaves it too late

We included two mask policies in our model: government provision of N95 masks to use instead of cloth or surgical masks, and health promotion campaigns to increase general levels of mask wearing. Both policies were activated in our model once COVID-19-related hospitalisations within the population exceeded critical levels. In brief, we found that these policies made very little difference to cumulative health outcomes over one year. Mask-wearing is known to be effective at reducing virus transmission, so we think the lack of impact we see in the modelling is due to other restrictions being put in place during waves (eg, a free N95 mask cannot do much when people are at home not wearing it). What we are seeing in additional, yet to be published, modelling is that increasing overall mask compliance consistently at all times (and among all age groups) is a more effective way to reduce infections and deaths than waiting until things are getting out of hand, although whether society considers this proportionate is a topic for another discussion.

Where to from here?

Despite recent rhetoric asserting that the prevention of SARS-CoV-2 transmission is now an individual-level responsibility, the results of our modelling highlight the benefit of ongoing, collective, society-wide responses to the pandemic in the medium term. We are now working to extend our analyses to capture alternative mask scenarios, new approaches to restrictions, and additional vaccination schedules. Modelling frameworks such as those explored in our recent work can guide policymakers to work through the costs and benefits of their choices in a way that is both rigorous and transparent. In this capacity, simulation modelling frameworks are essential to future pandemic response planning in 2023 and beyond.

Dr Joshua Szanyi is a medical doctor with additional qualifications in public health and epidemiology. He has worked in basic laboratory research, adult and paediatric clinical medicine, public health research and global health roles, and is currently an advanced trainee in public health medicine at the Melbourne School of Population and Global Health’s Population Interventions Unit.

Hassan Andrabi is a PhD student in decision neuroscience at the Centre for Brain, Mind and Markets, at the University of Melbourne. He previously obtained his Bachelors degree in Accounting and Finance (2019), and Honors in Finance (2020), both at the University of Melbourne. He currently works as a software engineer at the Population Interventions Unit, in the Melbourne School of Population and Global Health.

Professor Tony Blakely is an epidemiologist and public health medicine specialist at the Melbourne School of Population and Global Health, University of Melbourne. He leads the Scalable Health Intervention Evaluation (SHINE) program that evaluates the impact of interventions on health and costs. He has been a prominent commentator on COVID-19, and is now Chair of the New Zealand Royal Commission on COVID-19.

The statements or opinions expressed in this article reflect the views of the authors and do not necessarily represent the official policy of the AMA, the MJA or InSight+ unless so stated.

Subscribe to the free InSight+ weekly newsletter here. It is available to all readers, not just registered medical practitioners.

If you would like to submit an article for consideration, send a Word version to mjainsight-editor@ampco.com.au.