With artificial intelligence- and machine learning-based solutions becoming more common in healthcare, the next big step will involve folding AI and ML algorithms into wide-scale business and clinical workflows.

That was an overarching conclusion of State of the Industry: Machine Learning and Artificial Intelligence in Healthcare, a wide-ranging panel at last month’s HIMSS19 conference in Orlando, Florida. (Coming as no surprise, AI was cited as the most impactful innovation by HIMSS19 attendees.)

It’s an exciting time in healthcare because “analytics is finally catching up to the kinds of problems healthcare has,” said Stephen Blackwelder, chief analytics officer at Duke University Health Systems. These are what he called “meat and potatoes” problems, both administrative and clinical.

But capitalizing on that wider, broader application of AI renders a number of hurdles and bottlenecks, Blackwelder continued. One pervasive issue: the lack of data governance at many institutions that prevents aligning promising technologies, such as AI and ML, with business goals.

“At some point, you’ve got to evolve or mature into a mode where you’re able to align these really expensive and preciously rare resources with your corporate strategy. We’re going through that process right now at Duke,” he said, adding, “It’s really never too early to make your AI efforts, your data science efforts, align to the organization in the same way that you’ve done with other technologies.”

A related problem, said Benjamin Mansalis, MD, chief medical information officer at INTEGRIS Health, is that facilities and clinics have their own set of processes and leadership structures, which translate to different healthcare processes. “And that gets embedded into the data in a way that a data scientist might have a hard time parsing to leverage to create insights,” he said.

Meanwhile, some institutions have realized benefits from point-solution analytics, looking at things such as length of stay and readmissions. Evergreenhealth, for example, has reduced care variation and clinical variation by studying length of stay across its 30-plus hospitals, reported Jeffrey Tomlin, the hospital’s senior vice president and chief medical and quality officer. This resulted in both a clinical payback and a financial payback, he said.

For David Vawdrey, MD, vice president of the Value Institute at New York-Presbyterian Hospital and an assistant professor in Columbia University’s Department of Biomedical Informatics, analytics has helped create an “early warning system” for patients at risk of cardiac events. The 3,000-bed system was seeing about 60 unanticipated cardiac events per month, and by studying EHR and other data inputs, it is now able to identify deteriorating patients. “The thing we’ve found challenging is what I call the ‘last mile’ problem, or how to get the right insight to the right people at the right time,” he said. There’s also a trade-off between over-alerting, causing physician fatigue, and an actual benefit, he added.

Workflow

Embedding IT into the clinical workflow means more than EMR use, according to Vawdrey, noting that while healthcare has largely adopted the EMR, it does not train staff on “the optimization of using these tools.”

A related issue is that the technology is moving much quicker than the culture, which leaves some physicians wary of these cutting-edge tools in their practice. The panelists agreed that better information and less manual effort – using the EMR to present insights from a variety of AI-assisted systems and devices – is the key to obtaining physician buy-in.

Conversely, “if the machine is recommending things that aren’t helpful or waste time, you’re going to lose [physicians],” Tomlin said.

Data quality is an issue across the board, observed Vawdrey, stating, “It’ll take years to see the kind of payoff we’re talking about here today.”

The panel agreed that case management and other administrative tasks will be the first target of AI and ML approaches, followed by tools aimed at clinicians.

In his recent article “How will the convergence of artificial intelligence and analytics impact healthcare and life sciences in 2019 and beyond?” Andy Dé, senior director, Healthcare and Life Sciences at Tableau Software, likewise suggested that AI and analytics integrated with EMRs “will deliver actionable insights for superior care delivery and personalized care at a lower cost for healthcare providers.”

But like the HIMSS19 panel, the Tableau executive was clear-eyed about the obstacles. “The most significant challenge to delivering on the innovations above is the somewhat siloed approach to innovation deployment and delivery,” he wrote. “The notion of AI partnering with analytics to deliver actionable insights that will enable prediction and prescription for rapid decision-making, as well as automation of manually repetitive tasks, is still a work in progress.”