New AI technology foretells COVID-19 outbreaks

Researchers at New York University have come up with a novel way to help foretell COVID-19 surges before they materialise — an approach that could cost-effectively restrain viral spread until vaccination programs are complete.

Analysing online searches of mobile and isolated activities, their preliminary AI tool can predict where and when an outbreak might take place. Equipped with this information, authorities could stage urgent responses — like that recently seen in Victoria — nipping the virus in the bud before it replicates to uncontainable levels, and reducing the impact on local businesses.

Professor Megan Coffee and Professor Anasse Bari from New York University say the technology is based on “alternative data” — a concept previously used in finance to generate data-driven investments, such as predicting business earnings from satellite images of parked cars.

“What we wanted to do here was develop a barometer of behaviour,” Professor Coffee said to Hospital + Healthcare. “When we model infectious disease outbreaks, the curves are always nice and smooth, but in the reality of any epidemic I’ve worked in, there are ups and downs, sometimes even a roller-coaster of surges and drops in cases. These changes can be due to many factors, but human behaviour is a crucial cause.”

“As COVID-19 spreads, when someone googles the closing time of a bar or searches for directions to a gym, they give insight into what future risks they may undertake,” Professor Bari added.

“On a population scale, we can use these searches to create a data analytics tool that will be a barometer of behaviour. We can extend the data analytics tool to learn from an ensemble of data sources to track indoor activities that increase transmission such as open-table, to restaurant volumes, biking and taxi data, and real estate data to measure mobility and mixing.”

The tool is still in preliminary stages and still needs to be fine-tuned to better account for different activity preferences, as they vary by geographic regions, age and backgrounds — and the phase of the epidemic.

However, it would not be the first time the researchers have successfully used AI to improve the national pandemic response. Last year, they made headlines with a tool that could predict which COVID-19 patients would develop severe illness — a solution that could help clinicians work out who should be prioritised for acute care.

Next up, the researchers are working on extending their technology to other data sources that could hopefully capture more signals and predictors to the pandemic.

“This pandemic needs a holistic approach to fight it, from predicting clinical severity using AI, to building early signal tools using alternative data to help predict spikes and falloffs of the pandemic curve,” Professor Bari concluded.

Image credit: ©stock.adobe.com/au/Monster Ztudio