With intense demand expected for a vaccine for the COVID-19 virus, health experts will need evidence-informed strategies to decide who should get it first.
Aiming to provide these insights, University of Guelph Prof. Daniel Ashlock is helping build an artificial intelligence (AI) model that can analyze numerous factors to provide guidance on how to most effectively deploy vaccines and other disease mitigation strategies.
“If this model works, it could have significant implications for how public officials distribute vaccines across Canada,” said Ashlock, a professor in the Department of Mathematics and Statistics. “This will be an open source software, so anyone can use it. Knowing who, how and where to vaccinate first is critically important to mitigating the spread of the virus.”
The AI model will take a collection of vaccine or diagnostic test deployment strategies and use simulations to select the most effective strategy based on current information, such as the number of tests available.
It uses “hyperheuristics” and is highly adaptable to incorporate new information, strategies and recommendations as they arise. A key innovation about this model is that it can even consider asymptomatic carriers, who are sometimes excluded from epidemic and pandemic modelling.
A heuristic is a common-sense technique or rule of thumb such as “test people known to have come into contact with an infected person.” A hyperheuristic method combines many heuristics and chooses which one to apply based on current conditions; it can also pull in other relevant factors such as vulnerability, recovered individuals and test availability.
“Think of it as a ‘good advice generator’ that can swallow a lot of data to generate that advice,” said Ashlock. “Each of the hyperheuristics gives decision support to health units deciding who to test or who to vaccinate. The work we do on our own test networks creates general principles in the form of advice: for example, ‘vaccinating grocery store workers has more impact than vaccinating members of vulnerable populations.’”
This tool will ultimately be housed on a cloud server called Github, and be made freely available to decision-makers such as public health managers. Those decision-makers can then act on the advice provided by the hyperheuristics when considering access to tests and vaccines.
The project team involves researchers at three institutions, including Prof. James Hughes at St. Francis Xavier University in Antigonish, N.S., and Prof. Sheridan Houghten at Brock University in St. Catharines, Ont. Ashlock brings the mathematics perspective and “catches errors and notices bad assumptions” in the model, as well as finding new information and developing new heuristics that will help the AI model be faster and more effective.
“In early July, we will have a much better idea about what is working, and once we’re there, we will be able to begin to make the technology available for potential end users by late summer or early fall.”
The initiative is funded by the Nova Scotia COVID-19 Health Research Coalition.
Contact:
Prof. Daniel Ashlock
dashlock@uoguelph.ca