ROCHESTER – Artificial intelligence can now act as an artist or a writer. After years of hype, does that mean AI is ready to play doctor?
Many institutions like Mayo Clinic, Google and venture capital firms believe that AI’s potential in medicine is ready to mature into truly useful tools to help patients and doctors.
The Mayo Clinic Platform is working with Google and other technology giants as a number of startup firms to combine AI and Mayo’s more than a century of medical records and data. In early December, the Mayo Clinic Platform hosted its inaugural Health FWD conference to spotlight AI and life science entrepreneurship.
Dr. John Halamka, president of the Mayo Clinic Platform initiative, pointed out in his keynote address at the event that health care needs improvement.
“Despite many best efforts, care is still disconnected, reactive, late stage, confusing, expensive,” he said, referencing his mother’s recent experiences.
Halamka said that AI has the potential to improve the patient experience in many ways.
That’s the same message that many were expressing 12 years ago when IBM’s Watson supercomputer competed against and defeated human contestants on the game show “Jeopardy!” In the years that followed, hundreds of headlines described how AI would transform every aspect of medical care.
However, those breathless predictions faded away as many of the early applications failed to live up to those expectations.
Frank Jaskulke, vice president of intelligence at Minnesota’s Medical Alley organization, believes that AI is on the upswing following the traditional trajectory of technology.
Using AI is less expensive than in its early days, he said, and that is one factor spurring the acceleration of the number of AI-related startups appearing in the health care industry.
Medical Alley, which works closely with entrepreneurs in health care, has an estimated 100 to 150 members who are using AI in some fashion, according to Jaskulke. And that number is growing.
Jaskulke cited Gartner’s Hype Cycle, which includes five phases: Technology Trigger, Peak of Inflated Expectations, Trough of Disillusionment, Slope of Enlightenment and Plateau of Productivity.
“Every technology goes through this where there’s a huge amount of hype and people say that it’s gonna change the world forever. And then there’s the Trough of Disillusionment when it doesn’t quite work out that way. But then the real applications develop and it becomes a real thing,” he said. “AI, I think, is in that phase. It’s never universal, but more and more companies are using real AI technology to solve real clinical problems. It’s definitely accelerating.”
Jeff Clement, a doctoral candidate at the University of Minnesota who researches how clinicians and patients evaluate AI recommendations, also believes that AI is starting to make a real impact in health care.
“I think that we’re at the stage where everybody sees that this could do some good and a lot of people are using AI in their personal lives. They’re using AI to help them with traffic directions…They’re talking to Alexa at home, but they’re still very wary,” he said. “So we’re still very early in actually deploying and developing algorithms that would make a big impact on problems in patient care. And we’re still seeing huge advancements in AI technology.”
While AI in health care hasn’t had a big breakout application or company yet, artificial intelligence has seen some success in the medical realm in the past 10 years.
In Rochester, Mayo Clinic helped launch Ambient Clinical Analytics in 2013 under the leadership of well-known Rochester entrepreneur Al Berning.
Ambient provides clinical decision tools designed for use at a patient’s bedside or a nurse’s station. It was approved for medical use by the U.S. Food and Drug Administration in 2015. Ambient has found success in helping doctors and nurses detect and treat sepsis with its AWARE Sepsis DART system.
Many AI systems in health care use algorithms to compare a patient’s test results to large databases of patient records.
Halamka explained at the Health FWD conference that Mayo Clinic’s patient records are not enough and more are needed to include more patients of different demographics.
“The challenge, of course, is that we’re going to try to make these models generalizable to the globe using Mayo Clinic. Ten million patient records aren’t sufficient,” he said. “You know that machine learning is math, not magic. And if you have the wrong clinical context to the wrong data, inputs and outputs, you can generate some pretty compelling and utterly worthless algorithms.”
In 2022, Mayo Clinic signed a 10-year contract with a southern hospital system called Mercy. That agreement adds 15 million patient records from Arkansas, Missouri and Oklahoma to the mix. Halamka added that Mayo Clinic also is working with centers in Canada to add more variety to the patient database.
“They want to make sure that when they build models they are appropriately representative… This is the thing that I very much appreciate about Mayo Clinic’s involvement. They’re starting not from the tech side. They’re starting from the needs of the patient and asking ‘How do we build tech tools that achieve that goal,” said Jaskulke.
One challenge for developing AI algorithms is trying to keep racial or gender bias out of the results.
Referencing the old technology adage of “garbage in, garbage out,” Jaskulke pointed out that “You have to have really good data on the front end, to get an effective tool on the back end.”
Clement said bias is a major issue that developers need to address when working with AI in health care.
“If we train an algorithm just on the way we’ve always been doing things, we may be just replicating human bias. Human biases show up in the data as it exists now. So if we train an algorithm on that data, a lot of those biases will be captured in the algorithm,” he said. “That’s when you work with the developer and basically fine tune it on your patient data. So rather than a sort of a general solution, you’re able to customize it on your patient data.”
Referencing the old technology adage of “garbage in, garbage out,” Jaskulke pointed out that, “You have to have really good data on the front end, to get an effective tool on the back end.”
While artificial intelligence appliances are being developed in every aspect of health care from pharmaceuticals to diagnosis to monitoring, AI is already a standard tool being used in areas like radiology or electrocardiography
“Image processing is sort of an easier task for AI,” said Clement. “If you’re doing something with radiology, there’s a good chance that your radiologist has access to what’s called a CAD tool, Computer Aided Diagnosis.”
When it comes to businesses and economic development, is AI in health care on the upswing to the Plateau of Productivity?
“It depends on the sector. There are parts of health care health technology that are doing well. There are parts that are struggling. There are some that are in between. I feel good about the ecosystem. Rochester is growing. We’re seeing innovations in St. Cloud,” said Jaskulke. “We’ve had a very strong five years in the startup community, including 2022. Those innovations take five to 10 years for stuff to develop. I think those things we are seeing in the early stages now will pay off in 2024 or 2025 in major job creation, major patient impact and major wealth creation.”
While AI has potential to improve health care, experts said that doesn’t mean a computer program will be diagnosing and treating patients on its own. AI in health care is about augmenting the knowledge of a doctor or nurse.
“How many patient cases can a doctor see over their entire career? Many thousand, but not 100 million,” said Clement. “I’ve heard this phrase that by reading books you gain a 1,000-year-old mind, because you gain the knowledge. This is the same idea. There are rare diseases and conditions that might only come up a handful of times in a physician’s whole career, even at a place like Mayo Clinic.”