How AI is helping to spot the next pandemic

Forecasting future outbreaks is becoming increasingly complex. But as infectious disease specialist Kamran Khan explains, this is an area where algorithms can help.

The Asian tiger mosquito lives only a few weeks and flies less than a kilometre in its lifetime. Yet it has become one of the world’s most dangerous disease transmitters. Since the 1960s, it has exploded out of southeast Asia, colonizing every continent except Antarctica. In the Americas, it has spread as far north as New Hampshire and could reach Toronto within a decade.

As it conquers ever larger swaths of territory, more of the world’s population is at risk from the diseases it carries, including dengue, Zika and chikungunya. “We’re seeing these vector-borne diseases expand their ranges,” says Kamran Khan, an infectious disease physician at St. Michael’s hospital. “We’re seeing outbreaks in parts of the Mediterranean and the United States. These aren’t outbreaks where someone has returned from another part of the world. They’ve acquired the infection locally.”

The mosquito’s dramatic expansion is being driven by human activity. It reached North America on imported tires from Asia in the 1980s and has ridden ships and planes to Europe, Africa and Australia. And the warming climate has enabled this tropical creature to establish itself in more temperate regions that would previously have been off-limits.

Worrying as it is, the tiger mosquito’s globetrotting is merely a symptom of a broader problem. Skyrocketing global travel, rising temperatures and human encroachment into wilderness areas are changing the patterns of numerous diseases and creating new opportunities for pathogens to jump from animals to humans.

That’s complicating an already difficult task: spotting where the next pandemic might come from.

The X factor

Though most of us are trying to forget the pandemic (quite literally, it seems), it remains very much on the minds of public health officials. According to one estimate, a child born today has a 38 per cent chance of living through a pandemic like COVID-19. That figure is based on historical data that show increasingly frequent outbreaks over the past 400 years. But identifying the most likely risks is complicated.

The World Health Organization has a list of a dozen priority diseases that are the greatest concerns for public health, including Ebola, Marburg, SARS and Crimean-Congo haemorrhagic fever. But it cautions “this is not an exhaustive list, nor does it indicate the most likely causes of the next epidemic.” Tellingly, it also includes Disease X, representing an illness that’s not yet on our radar.

With so much uncertainty, “we’re going to have to get better at anticipating what is coming next,” says Khan.

Open-access online services, such as the International Society for Infectious Diseases’ ProMED system, have enabled the rapid dissemination of information about emerging threats. It has scored numerous wins, including being the first network to flag SARS as a concern. But the volume of information to crunch through is enormous — a similar system, Canada’s Global Public Health Intelligence Network, processes 3,000 news reports each day.

To help spot emerging variants or novel diseases, scientists are increasingly looking to artificial intelligence. “Today, this enormous diversity of global data is starting to become available, and we can use things like machine learning to look for patterns,” Khan says. Since the start of the pandemic, large amounts of time and money have been put into developing these AI systems. While some basically scour the internet for infectious disease news and report back on what they find, others provide more advanced analysis.

BlueDot, a Toronto-based company founded by Khan, uses AI to ingest information on disease incidence in multiple languages from such sources as public health agencies and global news services. It then layers this data on to detailed information about population densities, travel patterns, weather and other factors to model if and how an outbreak might spread.

AI systems have also been created that triage COVID-19 patients and estimate their likely length of stay in hospital, which could be useful in helping health systems deploy scarce resources where they’re most needed during future outbreaks. In a study earlier this year, researchers in the Biosecurity Program at the University of New South Wales, concluded that AI tools could be “revolutionary and highly sustainable” for public health surveillance, especially in parts of the world with weak healthcare systems.

From insight to action

Though the potential for AI is enormous, it also faces some hurdles. Developers train AIs to perform a particular function by showing them reams of relevant data. If you want an AI that identifies cats, you’d show it pictures of felines until it learns the difference between a tabby and a tiger. But major disease outbreaks are rare events, limiting the pool of potential training data. And there’s a further wrinkle: “When we’re talking about infectious diseases and climate change, we’re dealing with things that humanity has never experienced before.” That’s not just the rapidly shifting areas where diseases are found. Heatwaves and food insecurity caused by climate change could play havoc with people’s immune systems, hastening the spread of disease. There’s also the possibility of unknown bacteria and viruses being released by melting permafrost, with unpredictable consequences. It’s not yet clear how these kinds of events might impact the reliability of AI predictions.

Eventually, many of AI’s current limitations will be overcome and structures will emerge to enable data to be shared more quickly and easily. But in Khan’s view, artificial and human intelligences will always have to work together to spot emerging threats — machines will sift data for needles in a haystack; people will weave together the story. That’s how BlueDot became one of the first organizations to sound the alarm on COVID-19, back in December 2019. “We were picking up that there was this unusual respiratory illness in Hubei province in China. The machine basically presented that to our team and said: look, this is something you should pay attention to.”

But in the end, it’s the interaction between these AI systems and human policymakers that will determine whether we’re finally able to get out ahead of outbreaks. For crucial months at the start of 2020, repeated warnings from public health experts went unheeded by many governments that struggled to comprehend the scale of the threat.

“In the response to pandemics, behavioural science is really important,” says Khan. “What motivates people to take an insight and act on it? Ultimately, these insights have to turn into action quickly because we’ve learned in the last 20 years or so that when we’re dealing with infectious disease outbreaks, they move really fast. That means we have to move faster.”

 
To hear more insights on how the spread of disease is changing, listen to the latest episode of the MaRS podcast Solve for X: Innovations to Change the World.

 
Photo courtesy of MaRS Discovery District