When Providence Health and Services began its shift into value-based care and improving patient outcomes, it turned to its data and assessed what other innovators were doing in Silicon Valley, according to its Executive Vice President and Chief Clinical Officer Amy Compton-Phillips, MD.
LAS VEGAS -- Teams looking to AI and machine learning as a solution to their system’s problems should be reminded that the technologies are not a magic bullet, Leonard D’Avolio, co-founder of Cyft and an assistant professor at Harvard Medical School and Brigham & Women’s Hospital, said at HIMSS18 Monday.
In fact, he said, the success of these technologies within the industry will depend on focused efforts from health IT teams toward changing these misconceptions among clinicians and executives.
“When you think of AI, you have been trained to believe that this is a black box that you dump data into, and then at some point in the not too distant future it cures cancer,” D’Avolio said at the closing keynote of the conference’s Machine Learning & AI event. “And then we got upset that said vendor didn’t cure cancer in two years, and we came with our torches — this was never going to cure cancer in two years! And yet, these goals are sexy, and they sell, and so that’s what we spend most of our time talking about.”
D’Avolio argued that these misconceptions are largely driven by the industry’s use of “counterintuitive,” “dangerous,” and “just completely wrong” metaphors. Describing AI as a “black box” or perpetuating discussions about how it will replace trained doctors actively prevents constructive discussions on what the technologies can really do, and how they should be applied.
“It’s a tool, not a sentient being,” he said. “If you can’t match the tool to the job, then you can’t get the job done. I have a hammer, [but] that board needs cutting so get out of here with that hammer, that tool’s no good to me.”
D’Avolio advised teams to have a clear problem and execution plan in mind before settling on AI as a solution. To see this tool flourish, they will need to involve multidisciplinary teams and forego abstract statistical measures in favor of clear-cut measures of implementation success — such as cost savings.
“Nobody cares about your statistical performance,” he said. “What the bosses really care about is whether or not you can save money, or make money, or keep people healthy. If you want to stay in business, do not flash your C stat, recall, precision. Show them the dollars and cents that you keep making or saving, or you’re going to get trumped by a higher priority.”
The ultimate goal, D’Avolio said, is for the industry to view AI and machine learning as just another analysis tool to be used alongside statistics and data queries. Reaching this point, however, will require IT teams to quell the hype and improve understanding within their own systems.
“I just want you to recognize, and to share, and to discuss [AI] with yourselves and with your clinicians," he said. "As soon as you hear something like ‘AI is a black box,’ ‘you need a ton of data, don’t you,’ and ‘oh, there’s a ton of bias in that,' you have to set them straight. You have to make them understand … that this is just a tool, like any quality improvement tool [or] traditional statistic tool you study.”
Full HIMSS18 Coverage
An inside look at the innovation, education, technology, networking and key events at the HIMSS18 global conference in Las Vegas.
Featured Decision Content
You say you’re data-driven? What was once bleeding-edge is becoming passé and giving way to a new turn of phrase: information-driven healthcare.
A couple of years ago, KureSmart Pain Management began researching healthcare technology platforms to help it manage the patient experience and its online reputation.
BOSTON – Though analytics technology is well-developed and widely deployed, putting those tools to work for better patient outcomes is still a big project as providers grapple with complex and sometimes competing priorities.
BOSTON – To successfully leverage healthcare analytics, it might be time to break big data down into smaller increments to better transform the information into knowledge.