2020 has sparked countless stories of clever organisations adapting to the new normal, from home-delivered fine dining to nursing students learning to treat COVID-19 using augmented reality headsets. One Kiwi business used its expertise in tracking organisation memberships to develop a world-leading COVID forecasting model that saw them awarded a Microsoft AI for Good grant.
Since then, CloseAssociate’s algorithms, built using Microsoft artificial intelligence technology, have also helped the University of Illinois reintroduce students to campus more safely and predicted Victoria’s virus resurgence in mid-2020.
One of the most unsettling things about the COVID-19 pandemic has been the speed of change. Businesses considering online platforms or rolling out remote working in the next few years have been forced to adapt overnight. A single community case can lead to a nationwide lockdown within days.
For those of us – which is virtually all of us – without degrees in epidemiology, the spread of the virus can seem like a roll of the dice. Why are some areas worse affected than others? Why, when case numbers are falling, can there be sudden, terrifying flare-ups? The ability to stop these flare-ups occurring, and to allow citizens, governments and businesses to plan ahead, is crucial to rebuilding our national economies and keeping us all safe.
CloseAssociate, a Wellington-based company that specialises in helping membership organisations manage complex processes, seems an unlikely partner to help solve these issues. But in April 2020, co-founder Leon Grice, Chair of the NZUS Council, was invited to join a team of business experts to help the government solve urgent supply chain issues during the nationwide lockdown. In the team were New Zealand Defence Force logisticians working on scenarios to ensure New Zealand had the critical supplies it needed to respond to the emergency while the nation was in lockdown.
“All government agencies were thirsty for information and they were using the static model outputs from Te Punaha Matatini at the University of Auckland as the basis for scenario planning. They needed to run their own scenarios to fit their specific circumstances, but they needed to be done on a single model that didn’t vary from the work being done by the University of Auckland team, led by Professor Shaun Hendy,” Leon explains.
Leon immediately thought of his brother Stephen.
As one of New Zealand’s first licenced Microsoft engineers and with a PhD in physical chemistry, Stephen Grice had both serious digital know-how and a scientist’s passion for data and the power of applied mathematics.
Using Microsoft’s Azure cloud, web app Blazor and artificial intelligence technology, Stephen took Te Punaha Matatini’s Susceptible, Infected, Recovered (SIR) pandemic equations and developed a dynamic model that gave all of the logisticians and planners access to undertake their own “what-if” scenario models.
“At the beginning of April, Professor Shaun Hendy was modelling with COVID-19 data, but what we thought we could do was take the methods he developed and make them more accessible through a website that could be updated daily with new data,” says Stephen.
“The original results from Te Punaha Matatini were published as PDFs, which meant they only showed a snapshot in time. By creating an online model we wanted to make them dynamic and interactive to allow those “what if” scenarios to be tested.”
New maths for better predictions
To make it a true family affair, Stephen’s son Patrick, a new engineering graduate from the University of Auckland, was helping the team. While working on the SIR equations he had a eureka moment – he saw it was possible to use a numerical method to calculate a daily reproduction number. The reproduction number is a value that tells you, for a population, the rate that people are being infected. Also known as the R0 number, it gives you a reliable short-term prediction for the rate a disease is spreading. The more often-reported “daily cases” is by definition a reflection of the past.
The alternative to using a numerical method (which looks at differential equations) is statistical modelling on expensive supercomputers – and the results aren’t nearly as accurate.
Stephen realised Patrick was onto something and shared his original calculations with his old school friend, Professor Richard Laugesen at the University of Illinois. An expert in calculus, Richard told them Patrick had done something no one else had ever done before.
The trio formed a team to refine the method, creating a new approach based on several new techniques.
The first technique is based on the way high-definition video is transmitted over the internet, filtering out the “noise”, or unnecessary bits, as the file is compressed. The team treated case data the same way, filtering out the cases that didn’t seem to make sense – such as a sudden spike of cases on a Monday. Some countries weren’t inputting case data on weekends, skewing results, which CloseAssociate’s technique solved. This gave them much more accurate data to create scenarios with.
Another called “wild bootstrapping” identified the data people could be most certain of, so New decisionmakers could see how much confidence to place in particular scenarios. The team applied their methods to data from the European Centre for Disease Control and the data from Johns Hopkins University to create a real-time global COVID-19 database.
The result was a world-first numerical method – the Grices call it SIR+B – for calculating the rate of the virus’s reproduction, providing reliable short-term predictions.
“If you know the daily reproduction value with a high degree of accuracy you can reliably predict the direction of daily cases in the next 10 to 14 days. But the further out you go beyond that fortnight time period you are in the realm of human behaviour, which as we all know is impossible to predict,” Stephen says.
When Victoria in Australia was relaxing restrictions in early June due to a low number of daily cases, the CloseAssociate model revealed the reproductive number for the virus (the number of people who caught it from each infected person) was, Leon says, “going through the roof”.
“We contacted senior members of the CSIRO in Melbourne to provide our analysis and to report that they had a crisis brewing. You could map the outbreaks at their managed isolation facilities and the reproduction rates were skyrocketing,” says Stephen.
“We predicted there would be a spike in July, which unfortunately was the case and Victorian authorities progressively moved to lock down the state.”
Tracing the virus around the world
As a result of their clever innovation, the Grices have seen their free online COVID-19 modelling tool used by a range of New Zealand government departments, helping guide the national response to COVID-19. The model continues to be the gold standard for COVID-19 monitoring globally and has been used in the United States and in Australia.
The tool is also available for any international organisation or government to use and trace the likely spread of the pandemic in their own countries. From Norway to Nigeria to Nevada, CloseAssociate continues to run the algorithm on a daily basis for almost every nation around the world to provide an at-a-glance snapshot of how fast the virus is spreading.
It’s had some very practical results. In August 2020, the University of Illinois was looking to re-open its Urbana campus for 46,000 students at the start of the new academic year. However, CloseAssociate’s models indicated that students would bring infection onto the campus, which would result in a spread wide enough to close the campus.
University leaders realised they needed large scale, rapid, repeatable, non-invasive surveillance testing if they wanted to stay open. The university was already considering a new COVID-19 diagnostic test using saliva, and the model provided the evidence it needed to roll out the test to all students across its campus, alongside mask-wearing, rapid isolation of positive cases and online classes.
The non-invasive saliva test meant students could be frequently screened and cases could be caught before symptoms began to show. To date, the university has conducted almost one million tests on 46,000 people on campus and it has managed to keep operating despite surges of infection across the United States, with case numbers on campus much lower than the surrounding area.
In September 2020, CloseAssociate decided to licence the saliva screening technology from the University of Illinois. New business Rako Science is set to go live before Christmas, bringing the new testing technology to New Zealand.
Mapping the unknown
The outstanding work being done by CloseAssociate has now been recognised by a Microsoft AI for Good grant, awarded to organisations innovating with Microsoft technology in ways that help our planet and societies. Leon and Stephen say the funding will help them maintain their world map, developed on Microsoft Azure, which charts the progress of the virus.
“We have made it accurate down to province or county and we can even make it work down to postcode level,” Stephen enthuses.
He says using Microsoft technology just made sense.
“We’re native residents of the Azure platform and it’s quite clear that to do modern systems development you need to use the capabilities of the cloud. The Microsoft tools have enabled us to develop our model so much faster than we could have otherwise.”
Philanthropies Lead for Microsoft New Zealand, Elisa Willman, says social-minded innovators like CloseAssociate, who haven’t commercialised their platform, are proof you don’t have to be big to make a real difference in the world.
“The aim of the AI for Good programme is to empower clever thinkers everywhere to help solve the most pressing issues we face, and Leon and Stephen Grice are doing amazing things that governments and organisations across the globe are benefiting from. With our grant, we hope they’ll be able to continue helping countries plan ahead, manage the virus and save lives.”
“Long-term, tuberculosis is a bigger threat than COVID-19, and we’re also planning a framework to identify and monitor the disease in New Zealand,” Stephen says.
The brothers are excited about doing more original research and adding further value to the global dataset.
As Stephen says: “We’re from outside the medical field, which is devoted to statistical rather than mathematical modelling, so there was some scepticism of us at first. But we predict this methodology will become standard in every medical textbook within five years.”