Healthcare is a ripe target for machine learning to both optimize processes and greatly improve care delivery.
Take Mercy Health, for instance. When the health system was considering ways to improve care delivery, hospital executives looked at one of the most successful initiatives it had undertaken in the last decade: supply chain management.
"We have a lot of experience with operational efficiency," said Todd Stewart, MD, vice president of clinical integrated solutions at Mercy. Using the operative suite as an example, he noted that all the supplies that go into and through it are very expensive. And there were many base concerns that needed to be addressed, including issues that seem obvious but are not, such as block time for a surgical case, including how you define start and stop times.
Likewise with Mercy’s care pathways project. The hospital had a lot of data related to patient care that it can use to determine the best ways to manage patients.
That’s where machine learning comes into play. Stewart, who is both an informaticist and a practicing physician, said the tool Mercy uses, from Ayasdi, can pinpoint variations to determine optimal care as measured by quality, mortality, morbidity and length of stay.
"When you see a lot of variation in process, there’s opportunity for standardization and efficiencies," Stewart said. "That is a classic problem for machine learning — taking a large complex diverse data set and using the machine learning tool across those broad metrics allows us to get insight in ways we couldn’t with any other technological manner."
Mercy began by initially focusing on common high-volume procedures such as joint replacement, that are also expensive because of the hardware to replace a hip, for instance.
"That’s a straightforward place to start. Most of those are extensible across a large health system," Stewart said. "They’re expensive and amenable to efficiencies."
It’s also easier to standardize a knee replacement than, say, pneumonia or congestive heart failure because if you take any two patients with pneumonia, one might be better in three hospital days while the other needs eight. But even still Stewart said that Mercy is very close to standardizing care for those conditions as well.
Mercy, in fact, has already applied the machine learning tool to some 30 clinical pathways and in fiscal 2016 saved $14 million. Approximately halfway through fiscal 2017, Stewart said the organization has already saved $9 million and he expects it will surpass $14 million by fiscal year's end.
"We’re keen on developing predictive models for readmissions, things like sepsis, to who the really expensive patients are and manage those people better," Stewart said. "Healthcare data is a beast — it’s tougher than other industries."
HIMSS17 runs from Feb. 19 to 23, 2017, at the Orange County Convention Center in Orlando. Stewart will share lessons learned from his experience in a session titled, "Using Machine Intelligence to Reduce Clinical Variation," on Feb. 21, 2017 at 1-2 p.m. in Room 208C.
This article is part of our ongoing coverage of HIMSS17. Visit Destination HIMSS17 for previews, reporting live from the show floor and after the conference.