We've heard Dr Ashley Bloomfield mention modelling results at daily press conferences. When we hear that we can expect around 100 cases from this outbreak, what does that mean? How do we get to that number? Te Pūnaha Matatini modeller Dr Rachelle Binny demystifies the process for science reporter Jamie Morton.
First off, what actually is modelling? And what value does it offer decision-making in outbreaks?
In a model, processes that occur in real systems are encoded in mathematical expressions, which are then used to predict how that system will behave.
Models are very useful for understanding epidemics - mathematical rules describe how contacts between infected and susceptible individuals allow a virus to spread through a population and predict what might happen under different policy scenarios.
This allows decision-makers to plan ahead and to weigh the possible outcomes of different strategies for managing the disease.
They keep us on our toes with their questions as they work to understand the current situation and how certain interventions will affect it.
Can you tell us about the models you're using in this outbreak? What are they and how do they work?
Te Pūnaha Matatini are mainly using two types of model to predict spread of Covid-19.
One is called a stochastic branching process and the other is a contagion network model.
Both models predict the spread of Covid-19 through New Zealand's population, but they each make some different assumptions.
They can tell us how large an outbreak might get and how likely we would be to eliminate community transmission in various scenarios, such as different durations spent under alert level 4.
However, because real outbreaks are governed by random events – such as whether a superspreading event occurs at a bar, school or church, or whether someone who is experiencing symptoms gets tested – models can only tell us what range of outcomes might be likely and what can be expected on average.
How far has modelling come since the pandemic began? And why are network models more sophisticated than traditional stochastic models, for instance?
The network approach simulates the connections between every New Zealander and accounts for interactions occurring between people in different settings, like the workplace, in schools or at home.
This makes it particularly useful for predicting how transmission occurs in a specific community.
The contagion network team were able to set the network model running for Devonport as soon as we received word of the location of the first detected case in this outbreak.
The branching process model instead assumes that interactions are the same across these different settings, and is useful for predicting transmission when all individuals have the same chance of being in contact with anyone else, while accounting for differences between age groups.
In this current outbreak, both models are providing important information about the possible outbreak size and will be useful for predicting how case numbers might compare under different scenarios.
We've seen the modelling change slightly as we learn more about this outbreak. Why is this?
Over the course of the pandemic, the models have been regularly refined and adapted to capture changes in how the virus spreads - like the emergence of new variants, more transmissible than the original strain - and changes in the way we protect ourselves from the virus, like more people getting the vaccine, getting tested when they are sick, or using the contact tracing app.
Because New Zealand went hard and early last year, we've been able to eliminate past outbreaks and then learn from these experiences.
We've drawn on information from New Zealand case data and from other countries experiencing much larger outbreaks, to improve our models and update parameters.
The latest modelling results suggest that we can still likely expect around 100 cases from this outbreak. Why is this?
Identifying the likely time the first case arrived in New Zealand is useful, because it gives us greater confidence that the virus has been circulating in the community for a relatively short timeframe.
However, there are still several unknowns, for example how many large superspreading events might have occurred in that timeframe or how many cases may have spread outside of Auckland.
These chance events can significantly alter outbreak trajectories, and given the high number of locations of interest where large gatherings occurred, there were several opportunities for transmission to occur.
Ahead of last year's main wave, modelling gave us information around how many people might be hospitalised with Covid-19, or might die. Do the models we're using now go as far as predicting this information?
The models tell us about the spread of Covid-19 and predict the number of people who are infected over time.
If there are good estimates of how likely an infected person is to experience severe disease and require hospitalisation, or how likely they are to die from the disease - accounting for factors like age, ethnicity and whether someone is vaccinated - then the model can also be used to predict the number of hospitalisations and fatalities.
However, these estimates are different for some newer variants like Delta so we're updating the models to reflect these differences as new information becomes available.
If there's anything you think people need to keep in the back of their minds when modelling results are mentioned on the news, what is it?
The more data we have to work with, the better estimates our models can produce.
This early on in an outbreak, while we're still awaiting results from contact tracing, testing and genome sequencing, it's hard to get an accurate estimate of outbreak size because we don't yet know how many of the current community cases may be associated with superspreading events.
However, over the next couple of weeks, as more case data comes through, we'll be able to calibrate the models and get a better idea of how quickly the Delta variant has been spreading, and how the trajectory of the outbreak might change under alternative scenarios - for example, different durations spent in alert level 4.
We update and run the models every day as we get new information about this outbreak, so that our decision-makers can make informed choices about how to proceed.