In first study of its kind in New Zealand, data scientists have shown how artificial intelligence models similar to the popular ChatGPT can help hospitals predict surges in severe respiratory cases weeks before they hit. Photo / Janna Dixon
In first study of its kind in New Zealand, data scientists have shown how artificial intelligence models similar to the popular ChatGPT can help hospitals predict surges in severe respiratory cases weeks before they hit. Photo / Janna Dixon
In a first-of-its-kind study, scientists have found data-crunching algorithms can predict waves of severe respiratory cases in hospitals weeks in advance
The best-performing models draw on similar technology to that under-pinning clever AI like OpenAI’s ChatGPT and Facebook’s Llama chatbot.
Health New Zealand says it’s also looking at these types of models to help manage demand in our stretched emergency departments
Now, in the first study of its kind, scientists have shown how artificial intelligence (AI) models similar to the popularChatGPT can help hospitals predict these surges weeks before they hit.
The past few years have brought some of New Zealand’s busiest winter seasons: in July, officials issued a plea to the public after Middlemore Hospital’s emergency department (ED) was swamped with record patient numbers.
Yet the complex dynamics of seasonal nasties like flu have long made it extremely tough for planners to predict what demand hospitals might face season-to-season – let alone week-to-week.
Too often, that uncertainty meant surgeries being cancelled at short notice after intensive care beds suddenly filled with patients suffering from severe respiratory infections.
While the health system already uses forecasting models to manage pressure, data scientists have found the very latest AI might make a big difference in easing the burden.
The University of Auckland’s Dr Steffen Albrecht said these machine-learning tools could crunch entire years of variable hospital data at once to reveal useful statistical patterns.
Middlemore Hospital's emergency department saw a record number of patients in a single day amid this year's busy flu season. Photo / Mike Scott
In a just-published study, Albrecht and colleagues found two algorithms in particular – each developed for time-series forecasting – showed remarkable accuracy at picking admission numbers.
One was able to give reliable upper and low bounds of hospitalisation rates weeks ahead of time, while another produced slightly more accurate forecasts, although struggled somewhat with providing reliable confidence intervals.
As the models were tested only on datasets from the five years before the pandemic, he saw a need to trial them further on those from the post-Covid era, in which pressure has only increased.
“For the data we got from the Auckland hospitals, we are trying to find out if a single forecasting model will work to predict severe respiratory infection admissions – or if a separate model for admissions positive for [coronavirus] needs to be used,” Albrecht said.
Health New Zealand Te Whatu Ora’s group manager of emerging health technology and innovation, John Herries, said the agency was also looking at future use of these types of models.
Herries added there were “rigorous processes in place to minimise risks and ensure we can safely gain the most benefits for the health system and our people”.
The Ministry of Health’s chief science adviser, Dr Ian Town, said AI was being used in other ways across the sector, such as for speeding up administrative tasks and supporting clinical decisions.
Jamie Morton is a specialist in science and environmental reporting. He joined the Herald in 2011 and writes about everything from conservation and climate change to natural hazards and new technology.
