Health IT

How long until machine learning tools go mainstream? (Updated)

Machine learning tools are certainly spawning some exciting applications such as using medical images to analyze tumors and offering a new way to improve clinical trial matching. But how long will it take for these tools to become part of mainstream medicine?

technology, tech, IT, health IT, information technology

Note: This post has been updated from an earlier version

A panel discussion about de-siloing data and machine learning applications sparked a wide-ranging conversation spanning data security, consumer-facing digital health tools and the shortcomings of many of them, and how Mayo Clinic is collaborating with IBM Watson Health, which just turned 3 years old.

Kyu Rhee, Chief Health Officer at IBM Watson Health, Dr. James Madara, CEO of the American Medical Association, and Cris Ross, Mayo Clinic Chief Information Officer took part in the talk on the eve of HIMSS at a dinner co-hosted by IBM and AMA. Sue Schade, principal with StarBridge Advisors, was the moderator.

Here were a few of the highlights:

Mayo Clinic has just over 130 significant AI and machine learning projects, according to Ross. Image analysis accounts for a majority of them, such as using retinal image analysis to determine the likelihood that a patient has hypertension or atrial fibrillation which could indicate a higher risk of stroke. He noted that the hospital is working with IBM Watson Health to improve clinical trial matching, to make it easier to identify and recruit patients for oncology studies. One particularly interesting project could point the way towards eliminating the need for biopsies of glioblastomas.

“We have radiologists now who are able to detect the genomic content of glioblastoma brain tumors for example by using MRI images that look at textures of those tumors, which is highly correlated with a particular form of mutation,” Ross said.

Still, he noted that he is much more a fan of what he called “little AI” for utilities such as natural language processing and data enrichment compared to the big AI that is talked about in the context of moonshots.

Ross contended that machine learning was mainstream now. It’s an interesting point since image analysis certainly seems to be one application widely embraced by healthcare organizations as a clinical decision support tool, although many institutions are still piloting this technology.

Cybersecurity 

“I hate and resent every minute I spend on information security because I am doing nothing other than preventing criminal activity and it really annoys the hell out of me,” Ross admitted. “We’re doing many things but one thing we are doing is moving more data to the cloud. But he pointed out that once hospitals become more difficult to penetrate, cybercriminals will turn their attention to a more vulnerable group — patients — which will trigger some challenging discussions.

“We will have, I think, a moral duty to help patients protect their data in a non-patronizing way to help them to make smart choices about how they steward their data and what risks they will and won’t be exposed to,” Ross said. “We cannot make this into a boogeyman that scares us from doing the right thing… We have to keep innovating with tech…”

Still plenty of snake oil for sale in digital health

In the question and answer session, Madara was asked how he regarded the digital health space now. He noted that plenty of digital health tools available to consumers fall short.

“We are very supportive of technology but we don’t like bad technology. If we look at mobile health apps and many of the devices that are wearables, often they are not evidence-based, often they were not validated, actionable, and still today almost everything is not connected…We are very focused on technology that can improve health. We are not so high on things that say they are medical technology but are nothing more than fancy games with little physiological meaning. I think there is a lot of junk out there.”

Madara pointed to the Xcertia program as an example of an initiative involving AMA, American Heart Association and other organizations to hammer out guidelines and best practices for mobile health apps.

For his part, Ross noted in a different part of the discussion that after doing some inventory, he and others were surprised to find 50 to 60 Mayo Clinic branded apps in Apple’s App Store. They soon changed that, he said with a smile.

“You can’t have evolution without mutation, you have to have variation and some of it won’t be reliable. So it’s a bit of a roller coaster ride…Generally, we are not even close to reliable, consumer-facing [health tech].”

Photo: coffeekai, Getty Images

Correction: An earlier version of this story incorrectly attributed the quotes in the Cybersecurity section to AMA CEO James Madara. We regret the error.

 

Shares0
Shares0