Mandatory check-ins via QR code have become a common sight. What are the behind-the scenes workings of a system that is crucial to containing outbreaks of COVID-19?
By Gary Anders
Multiple outbreaks of COVID-19 across Australia over the last 18 months have heightened public awareness around the importance of contact tracing systems.
Tracing the primary sources of outbreaks, their immediate contacts, and identifying exposure sites as quickly as possible, has been critical in helping to contain the spread of the virus.
Australian state and territories continue to rely heavily on systems that use complex mathematical modelling to map out the potential severity of each occurrence and to determine what population control settings may need to apply.
Working with the Institute for Disease Monitoring in the United States, the Melbourne-based Burnet Institute, an internationally renowned medical research institute, has developed COVASIM, a unique individual-based model that can assess the impacts and risks associated with COVID-19 and includes interventions such as contact tracing.
The COVASIM model has been calibrated to specific locations and, as well as being widely used in Victoria, has been applied across parts of the US, the United Kingdom and countries across Africa and Asia.
The importance of tracing models
The application of probability scenarios is at the heart of contact tracing models.
“The mathematics comes in when we calculate the probability that an outbreak will become out of control,” says Dr Nick Scott, an econometrician at the Burnet Institute.
“What we mean by that is that we’ll calculate the probability that further public health responses will be needed – or that restrictions will need to be increased.
“These probabilities come from the fact that each person in the model has different numbers and types of contacts, and so there’s a lot of chance involved at the early stages of an outbreak.”
Through COVASIM, which is an online open access platform, it is possible for users to simulate potential scenarios whenever COVID outbreaks occur using real-time data.
“We’ve worked very hard to try and get the contact tracing as realistic as possible, because it’s important to get the model dynamics right when the cases are low and when we’re considering things like outbreak risks and early responses, particularly things like the snap lockdowns that have been going on,” Scott says.
“We’ve been using an agent-based model. What that means is we simulate individual people in the population, and in the model we can give each person slightly different characteristics.”
This can include varying the numbers and types of contacts that individuals have, their social habits and networks, and the use of community infrastructure including public transport.
“It’s really important for us to treat these types of contacts differently, because we know that there are different transmission risks associated with the different settings and different kinds of contacts,” Scott explains.
“We also know that contact tracing works differently depending on where the people interact with one another.
“Sometimes when you simulate an infection, nothing will happen. It will just fizzle out.
“And then other times when you introduce an infection, you might get unlucky, and it might have multiple generations of infections and lead to an outbreak. What we do is calculate the probability of different outcomes.”
Why data is key to contract tracing
Professor Peter Taylor, chair of operations research, mathematics and statistics at the University of Melbourne, says that the use of contact tracing data is what informs mathematical models of the COVID-19 pandemic.
Taylor has been part of a COVID simulation project known as Safe Blues, which has also included experts from the University of Queensland and US universities Cornell and MIT, to track how social mobility affects the spread of epidemics.
“The idea is to spread an ‘artificial epidemic’ on people’s mobile devices. We don’t measure contacts directly but, by simulating an artificial epidemic on the real contact process, we can use machine learning to predict how a real epidemic will spread given the same contact process.”
Taylor says a typical epidemic model needs to take into account the way the epidemic spreads from one individual to the other.
“There’s two aspects to it. One is how close people come to each other, which is a social thing, and also depends on regulatory measures.
“The second thing is, given contact, what’s the probability of transmission. So, if I’ve not got COVID and I’m standing next to someone who has, then what’s the probability that they give it to me?”
Taylor says that may be affected by things like mask wearing, hand sanitisation, the biology of the disease and variants, and even the biology of individuals.
“Most epidemic models aren’t very good at separating those. They just say, what’s the probability that person A, who’s got the disease, passes it on to person B, who hasn’t?
“In the statistical sense, they’re factors that are not identifiable. What contact tracing tries to do is at least get information about the first bit of it.”
Predicting disease outcomes
Of course, contact tracing models aren’t just confined to COVID-19.
“The model could be adapted to another disease if there was an outbreak of a new pathogen,” Scott says.
Irrespective of what is being traced, the success of tracing models comes down to the input and analysis of data to predict how a disease outbreak could spread across the community.
“The thing about mathematical modelling is, you can sit down and think about the process of the epidemic spreading, but you want to know the parameters – rates of spreading, rates of meeting people, probabilities of transmission,” Taylor adds.
“They’re the things that need to be filled in for the model to be useful.
“Contact tracing helps inform the mathematical models, and then the mathematical models help inform policy and decision-making.”