Philip Payne, PhD, has seen health informatics come full circle, he told Becker's. He would know. He directs the Institute for Informatics at Washington University School of Medicine in St. Louis and has been in the field for decades.
"When I was a graduate student … we were very focused on, how would we digitize the delivery of care? How would we help providers and patients make better decisions through the use of artificial intelligence? How can we accelerate research using better data management tools? How can we connect the dots to either biomolecular measurements or population-level measurements?" he said.
Dr. Payne, who is also chief data scientist for the medical school, said the field diffused for a bit — there were about 20 different subdomains of informatics — before returning to its core message.
"That's likely because AI has become more mature," he said. "We have widespread adoption of electronic health records. People are recognizing that the information overload in the lab, in the clinic and at the population level is deleterious and all of a sudden we're back to, 'How does computation and systems thinking and human decision-making come together to improve outcomes?'"
Dr. Payne originally planned to study medicine at the University of San Diego before he started playing around with databases there. He realized that rather than serving one patient at a time, with this work he could affect hundreds of thousands in a fell swoop. He was a professor and chair of biomedical informatics at Columbus-based Ohio State University before founding the Institute for Informatics at Washington University in 2016.
Showing the usefulness of AI
Dr. Payne and his peers have already put AI to some cool uses at Washington University. Researchers there helped develop a tool to predict which patients will develop sepsis. It has been able to say, with more than 80 percent accuracy, who would become septic within eight to 10 hours.
"We did that using raw, unfiltered data from the EHR — not highly curated data, not administrative data, not other specialized data collected for research — just the raw, messy, incomplete, sometimes inconsistent data that comes out of human beings interacting with an electronic health record," he said.
He and his colleagues have also employed AI to predict which children with neurofibromatosis, a rare genetic disease, were likely to develop comorbidities.
He said when it comes to leveraging big data, partnerships with industry provide greater "scale and agility" than traditional research funding, which can take years to yield results. To that end, Washington University and its affiliated health system, St. Louis-based BJC HealthCare, recently began collaborating with analytics startup CuriMeta to share healthcare data sets with life science companies. Washington University and BJC led a $6 million funding round for the tech firm.
Dr. Payne said that rather than waiting for potential partners to come to them, the collaborative will identify ones that will benefit all three parties.
"The biggest innovation from my perspective is changing the business model from one that's reactive to proactive, creating partnerships that are mutually beneficial and making sure every one of those projects turns into a new diagnostic or therapeutic that we can deliver to the patients we serve," he said.
How AI will become ubiquitous in healthcare
Dr. Payne cautioned there's been a lot of "breathless abandon about the power of AI to change healthcare."
But just like the newest test or drug or therapy, AI must endure scientific rigor. It has to go beyond an "engineering exercise where something might work in a lab" to an application that's effective in the "messy real-world environment that is human life and healthcare," Dr. Payne said.
"So do I think AI will fundamentally change healthcare? Absolutely," he said. "Because the power of well-formulated, well-demonstrated AI is that we can see the future for the patients that are in front of us today."
Dr. Payne said he sees AI at the point where EHR interoperability was 20 years ago, when it "was largely seen as an unsolvable problem."
"We couldn't figure out how to make different systems talk to each other and share syntax and semantics, in the absence of human beings hand-encoding every piece of data and mapping between them, which simply didn't scale," he said. "Yet today, we accept that we can have these automated tools that can arbitrate between different systems using standards and shared data models and terminologies and ontologies, and to some extent AI, to automate these processes.
"AI is going through the same evolution. It's seen as a highly specialized tool where there are more questions than answers, and we need to demonstrate its utility. Today, nobody goes to a computer and says, 'Well, I sure hope this interoperability solution works and I can see this data from another EHR.' Success in the AI field will be when we don't talk about AI as being special anymore. It will just be baked into the electronic health record and other tools."
Dr. Payne said one day, providers may be able to see, at the click of a button in the EHR, treatments and outcomes from hundreds or thousands of patients to inform decisions about their own patients.
"And that will be a massive AI lift, and they will think nothing of it," he said. "It's not a whole lot different than when you ask your voice assistant at home, 'What the weather's going to be?' There's a tremendous amount of AI behind that. But we're totally willing to accept that that's just something the voice assistant does."