AI Has an Image Problem in Healthcare, Expert Says

PHILADELPHIA — Artificial intelligence (AI) has an image problem when it comes to healthcare, but it actually represents a big opportunity to improve things, Tom Lawry said at the Population Health Colloquium here.

“When you think about everything that you’ve been reading, whether it’s lay journals or clinical journals, there’s a lot of talk about the [AI] threat, that we should go slower,” Lawry, managing director at Second Century Technology in Seattle, said. The speed of change is hard to keep up with and many things still need to be figured out, “and we will — we’ll find the guardrails,” he added. “But we should also be talking about what I call the ‘existential opportunity'” of AI.

Instead of pointing out AI’s “existential risk,” the healthcare community should instead be talking about the imminent healthcare risks faced by Americans, said Lawry, who was formerly national director for artificial intelligence in health and life sciences at Microsoft. “Americans are among the least healthy people in the rich world, and among the most likely to die early.”

“In America, maternal death rates of Black women are 243% higher” than their white counterparts, he said. “And our track record today [on that] is worse than when we started keeping records in 1918.”

Because of those issues and other challenges — for instance, the fact that the U.S. is the global leader in avoidable amputations related to diabetes — “I believe healthcare is a very noble cause,” said Lawry. “We have significant challenges that cannot be solved by the way we think and work today, which to me is where artificial intelligence comes in.”

He gave the example of Singapore, which has the worst incidence of diabetes of any Asian nation. A few years ago, “they literally declared war on diabetes, and they’re doing all kinds of things on the preventive side,” he said. “A U.S.-based company is making use of AI … They went into citizen data and then identified of the 5 million Singaporeans, which ones had the highest risk of being prediabetic moving to diabetic.”

From that group, “they picked up a subset of 100,000” to work with on adopting healthier habits. “And guess what? The early indications are, they’re slowing the growth from prediabetic to diabetic,” Lawry said, noting that the Singapore government saves $8,000 a year on average for every patient who doesn’t cross the line into diabetes. “So think about our ability to take that and be proactive, and actually manage diabetes rather than waiting until we can become the number-one leader in preventable amputations. That’s the power of AI.”

Like any technology, AI does have its limitations, he continued. For example, while AI may be better at pattern recognition, image analysis, automation, and information processing, humans are better at reasoning, judgment, imagination, creativity, and problem-solving.

“We haven’t figured out how to outsource any of that,” he said. “You hear a lot about — and it’s even in the journals — ‘well, maybe we should start training fewer radiologists because AI is going to replace them’ … Any time I hear that, I realize they don’t get this. One, they don’t understand what AI’s good at, and two, they probably really don’t understand what a radiologist actually does.”

“The idea is, whatever you do, the value proposition comes when we take anything in AI that’s better than humans, and bring them behind you — every healthcare knowledge worker — and say, ‘how do we make you better at something you care about?'” he said.

In an interview with MedPage Today after his talk, Lawry said it was true that AI could help radiologists by pre-sorting a big group of images into two smaller groups: those that didn’t appear to show a medical abnormality, and those that were more suspicious and likely needed the radiologist’s closer attention. However, he said, that did not mean that AI would eventually replace radiologists.

Instead, “it allows you to play them to a higher value,” he said. “They are the advisor to all of those other specialists — like the oncologists and the surgeons — so the ability to say, ‘I’m going to spend less time doing this routine stuff, which gives me more time to focus on those consults with the specialists'” would be a help to them.

“It’s a way of reducing that work burden, which causes burnout,” Lawry added. “If I can help restore what you do to be more aligned with that intrinsic reason you went into medical school, isn’t that a good thing?”

AI should be part of any healthcare organization’s human resources plan, he said during his talk. “There’s a huge upskilling initiative that has to take place with every knowledge worker in healthcare,” he noted. “Those who do that first, and do that well, are going to have a great competitive advantage, whether it’s being more efficient or [in retaining] workers. How do you become that talent magnet where people say, ‘that’s the place I want to be?'”

AI should also be included in diversity, equity, and inclusion (DEI) initiatives, he said. “So many of the things that are the biases and inequities in the real world, are starting to cross over into digital world into things like algorithms. We have the ability to solve for that much easier than eradicating biases and inequities that exist in humans.”

“In healthcare, AI is not about technology; it’s about empowerment,” Lawry concluded. “It’s about restoring clinicians and consumers to be better at what they care about.”