Must-haves for machine learning to thrive in healthcare

MIT professor John Guttag said that growing sets of aggregated data, federal rules mandating access to information, and existing tools make machine learning a reality today. Here’s what healthcare organizations need to know about the emerging technology.

By Mike Miliard

September 08, 2016

07:53 AM

When John Guttag keynotes the HIMSS and Healthcare IT News Big Data and Healthcare Analytics Forum in Boston on October 24, the MIT professor will describe the unique challenges of applying machine learning to healthcare – as well as the huge potential for efficiencies and quality improvements as these data techniques become more widespread across the industry.

Guttag, who heads the Data Driven Inference Group at the MIT's Computer Science and Artificial Intelligence Laboratory, and his MIT students are currently working closely with Mass General on integrating machine learning into clinical workflows, specifically with the aim of reducing healthcare-associated infections.

"I want to actually see things change in the system, not just write papers saying things could change," Guttag said. "The goal here is to have something good happen. I hope a year from now I'm able to say, 'Guess what, we've lowered the rate of nosocomial infections at MGH – and more importantly put together a description of how we've done it that is exportable to other organizations.’”

That’s no small feat but Guttag envisions big changes ahead for healthcare.

Healthcare organizations, for instance, now have more effective technologies for gathering information than ever before, and the federal government is mandating access to data sets that hospitals used to keep secret, such as infection rates.

"We're also seeing aggregated data," Guttag said. "Very few hospitals have enough data all by themselves to effectively deploy machine learning. But what we're seeing now is that, first of all, hospital systems are growing. The standalone hospital is rapidly becoming a thing of the past. And as we see these healthcare systems grow, there's aggregated data across systems.”

That critical mass is a must-have for machine learning to work. A small hospital, for instance, won't be able to do much with just its own EHR data.

Another must-have is the right expertise. "Machine learning in the healthcare context is not easy, and you either have to have expertise or hire expertise," said Guttag. "It's not a job for amateurs. It's not like you can just buy the technology, dump your data in and turn the crank. It doesn't work that way at the moment."

That also is changing and Guttag added that some private technologies can be very valuable to hospitals looking to implement machine learning.

"Most companies would tell you they have some secret sauce. And in many cases it's true. IBM has figured out how to make Watson work. The company I'm involved with, HEALTH[at]SCALE, we think we have some very valuable proprietary technology. Google has some very valuable technology – they've also released a lot of tools in the public domain which are pretty amazing,” Guttag explained. “The technology is in a pretty decent place for a lot of things. Is it going to be in a better place a year from now? Absolutely. The technology is improving by leaps and bounds."