AI on the COVID-19 frontline and beyond
Professor Megan Coffee’s colleagues at a New York hospital became doctors to save lives. Now — besieged with a fresh onslaught of ambulance arrivals — they are learning ‘on the fly’ how to predict which COVID-19 patients will need mechanical ventilators.
There is limited data at hand to help them make this decision.
“Age, gender and comorbidities alone are not enough to predict who will develop severe symptoms,” said Professor Coffee.
“For clinicians this has been a real challenge, throughout the pandemic, in terms of deciding the right treatment options.
“Pilots need to know where they are flying; and doctors directing care do, too. Yet doctors have had to plan patient care — learning as they go along, who is most at risk of deteriorating in their clinical condition.”
They say adversity is often the catalyst for change and COVID-19 has been no exception.
At the start of the coronavirus outbreak in Wuhan, Professor Coffee and her colleague, Anasse Bari, Professor of Computer Science at New York University, set out to create an AI-powered decision support tool to help clinicians in this context.
Learning from historical clinical data about former coronavirus patients, the AI tool can predict which COVID-19 patients will develop severe respiratory illness, before any major symptoms present.
“By flagging severe cases early, it can lower emergency visits, help hospitals decide which cases to monitor and, in turn, save human lives,” Professor Bari said.
This predictive analytical technology is one of many ways AI has proven useful in the trenches of COVID-19 health care.
An AI platform has helped to track the propagation of the virus, by algorithmically analysing alternative data sources.1
A computer vision algorithm, coupled with predictive analytics, has predicted which geographical areas would be most vulnerable in a disease outbreak.2
In drug development, AI has come up with potential vaccine candidates, far quicker than humans have been able to.3
Meanwhile, AI-powered drones have been used to deliver medical supplies and food throughout the pandemic.4
But although AI has formed a major part of the COVID-19 response, experts aren’t sure if this crisis will be the tipping point for widespread AI usage. Before that can occur, cultural attitudes and norms within the healthcare industry may need to change, they say.
“At present, AI is mainly just used in visual detection. For example, cancer on mammogram films, photos of skin lesions, tuberculosis chest X-Rays or signs of diabetes on retinal exams,” Professor Coffee said.
“This is a shame because AI could play the role of master diagnostician, help doctors identify possible causes of an illness or predict more severe patient outcomes. Not just in pandemics, but in everyday clinical cases.”
Professor Coffee said there are a number of possible reasons why clinicians and medical procurers may be hesitating over AI.
“Doctors still carry beepers. We aren’t always the earliest adopters of technology,” she said.
“One important reason for this is that we need to be extra sure that the technology works.”
Uncertainty about how a data-fed AI machine comes up with its prediction — also known as the ‘blackbox problem’ — is of particular concern, highlights Google Health Clinician Scientist Dr Martin Seneviratne.
“Health care is an industry in which people are used to asking ‘why?’ and understanding the exact mechanisms which underpin any given area of practice,” he said.
“But then again, there are many medications (take lithium) where we might not understand the full mechanism of action. But it has been shown to be safe and effective for many patients.
“We should have the same approach for digital therapeutics — study the logic behind them for sure, but also build empirical evidence to show they are safe and effective.”
Professor Bari said clarity in terms of AI’s role in health care may also be needed before widespread uptake occurs.
“My general theory about AI is that it exists to serve as an extension to human intelligence, not as a replacement,” he said.
“We have reached a point in science where, as AI experts, we can admit (with empirical evidence) that the way a computer thinks is fundamentally different to how humans think.
“This is why I believe human judgment and creativity is needed more than ever in the applicability of specialised AI.
“An AI system will advise and make suggestions. The human will have to make the final decision — either following the AI or, at times, correcting it, so the AI becomes more intelligent.
“An AI system is in a constant learning process, and humans make it possible to advance this learning, by providing feedback and new training data.
“That is exactly the vision we have for AI in health care.”
Whether or not COVID-19 will prompt a revolution of AI in health care is still unknown, but one thing we can be sure of is that it has fostered its evolution.
The pandemic has exposed the healthcare industry across the Asia–Pacific region, which is...
The pandemic led to a significant increase in workloads at hospitals across Australia, forcing...
Beyond the COVID-19 pandemic, how do we maintain the momentum in virtual care that is...