Hospitals, Artificial Intelligence, Health Tech

AI in healthcare means grafting ‘Star Trek’ onto ‘The Flintstones’

At the MedCity INVEST Digital Health conference Tuesday, experts discussed some of the challenges to implementing artificial intelligence in healthcare.

L-R: Jodi Hubler, Matthew Versaggi, Steven Sigmond, Mark Foley

One of the biggest challenges to implementing artificial intelligence in the context of healthcare is that the data that exist to do it are “dirty” – which is another way of saying they’re often inconsistent or simply inaccurate.

The challenge that such messy data pose to implementing AI in healthcare was among the topics of a panel discussion at the MedCity INVEST Digital Health Conference, which took place Tuesday in Minneapolis. Lemhi Ventures managing director Jodi Hubler moderated the panel, which included UnitedHealth Group – Optum Technology senior director of AI and cognitive technology Matthew Versaggi; Carrot Health CFO Steve Sigmond; and Mayo Clinic director for translational informatics Mark Foley.

In contrast with popular visions of AI becoming a sentient entity – like the machines in “The Matrix” or Skynet from the “Terminator” franchise – panelists explained that it’s really more about data and the analysis thereof.

“It’s very simple: powerful statistical predictive calculations on high-powered computers,” Foley said, to provide a succinct definition of AI. “It’s really probability, and we just have fast enough computers now that can tell you an answer.”

But one of the hard things about bringing that into healthcare is that the sector has been behind when it comes to technology in general.

Hubler remarked that on the one hand, a “Star Trek” vision of technology is frequently touted. “But healthcare operates on a ‘Flintstones’ chassis.”

To give one example, Foley recalled a TED Talk he heard where few people in the audience reported having heard of fax machines being used in the last three years, but noted that they are still widely used in healthcare for communication between hospitals and nursing homes.

Meanwhile, Sigmond pointed out that in countries like India that never fully developed landline telephony, cell phone technology was adopted wholesale later on.

“It’s kind of like we’re building a 2020 cell phone network in an emerging market, and they never had the chance to get invested in this legacy technology,” he said, referring to healthcare’s integration of AI. He said that in healthcare, he has seen people still using Windows 98 and relying on other outdated technology, but this also creates the opportunity to skip attempts to develop old systems and move straight into the latest cutting-edge technology.

“Chicago is a good metaphor,” Versaggi added, referring to how the city was rebuilt “the way it should be” – on a grid plan, rather than like the European-like streets of Boston – after it had the “good fortune” of burning down in 1871.

But “dirty” data are another challenge.

Foley said his weight has fluctuated within three kilograms over the last 25 years, but on one visit he found he had gained 100 kilograms because his information was entered incorrectly. A lot of data entry in healthcare, he said, is copy-paste, and incorrect data makes it challenging to implement AI. “You can’t build algorithms when you have bad data,” he said.

Obtaining those data can be difficult as well. On the topic of data ownership and security, Versaggi said he heard at quantum computing conferences he attended that companies sometimes go overseas to obtain black market data for their models because they can’t get it in the U.S., where laws make obtaining it domestically difficult. A recent incident illustrating the importance of data ownership was the lawsuit filed against the University of Chicago and Google, alleging that the alliance they formed two years ago to capture electronic health record information for analysis was shared without being properly deidentified.

Photo: Stephanie Baum, MedCity News

Shares1
Shares1