AI's potential runs up against lingering data issues

By Jonah Comstock
Share

Artificial intelligence has huge potential to transform care, but the healthcare system needs to walk before it can run, according to a panel of speakers from IBM Watson, the Mayo Clinic, and the American Medical Association at an after hours dinner at HIMSS18 Monday.

“I think this AI stuff is absolutely real, but at the same time we haven’t finished the first job, which is creating systems that are usable by clinicians,” Mayo Clinic Chief Information Officer Cris Ross said.

The panel talked a lot about EHRs which are still the number one pain point for physicians, who are spending twice as much time entering data as seeing paper. That documentation, for the benefit of the payer, reflects a set of priorities that need to be changed according to AMA President Dr. James Madara.

 

“We have to flip this model, starting at the point where we think the truth of the healthcare system is,” he said. “Because what we’re doing today would be the equivalent of General Motors making cars and paying attention to the dealers but not caring at all about the drivers or the mechanics. No other field would do that and yet that’s where we’ve ended up.”

For machine learning to meet its full potential, it needs access to as much data as it can get, which makes data silos a problem for the future of AI as well.

“I think it’s a lot about trust,” IBM Chief Health Officer Kyu Rhee said. “Fundamentally health and healthcare is about trust. Data is such a natural resource and sharing data across multiple stakeholders requires trust. It requires trust from the patient, requires trust from the provider and the payers, and all these different stakeholders that need to connect this data together to bring these insights out.”
So for organizations like the Mayo Clinic to put machine learning to work on things like clinical trial matching and clinical decision support, they have to make their own data resources.

“All enterprise systems want to own the data,” Ross said. “It’s part of how they compete, it’s part of their raison d’être, they just do it that way. And the challenge is that transactional systems are wonderful for transactions but they’re not great for insights. Our example, and I know other people who do the same thing, is we’re essentially replicating all of our data, out of all of our transactional systems, and putting them into a big data environment in which we’re driving all kinds of analytics.”

Mayo is also moving that data environment to the cloud, Ross said, because of omnipresent security concerns.

“We’re getting out of the data center business, moving as much as we can to Azure and Google Cloud,” Ross said. “I think soon onsite data centers will be quaint, like onsite electric plants were circa 1920. But the threats of cyber intrusion are just not going to go away, and once healthcare institutions get buttoned up I’m afraid to say they’re going to go right for the consumer and patient. And we’ll have a moral responsibility to help them protect their data and, in a non-patronizing way, help them to make smart choices about how they steward their data and what risks they may or may not be exposed to.”

Rhee suggested that in addition to keeping data safe and secure, hospitals should make it easier for patients to voluntarily share data toward big data machine learning efforts.

“We have to complement the data privacy and data security thinking with a different concept called data philanthropy,” he said. “In the same way I get to choose to donate my organs when I die, maybe I get to choose to donate my health data when I’m alive.”

Ultimately, AI will succeed when it has enthusiastic buy-in from physicians and patients. Madara thinks that’s a strong possibility if systems are designed right.

“When we look at the history of technology, what we learn is — whether it’s new approaches to therapy or robotic surgery — new technologies that allow physicians to take better care of patients, in an unambiguous way, are rapidly adopted,” he said. “What’s not rapidly adopted are things that don’t work very well, are highly ambiguous, or create more effort.”

монтаж автополива

https://www.progressive.ua/

www.pillsbank.net/en/products/kamagra-effervescent