Topics
More on Accountable Care

AI can enable value-based care - if it's backed up by action

At Advocate Aurora Health, predictive analytics and telephonic coaching led to a 23% reduction in unnecessary utilization.

Can artificial intelligence improve value-based care? Tina Esposito, chief health information officer at Advocate Aurora Health System says the answer is yes -- and no.

"There is no return on analytics or AI unless someone does something with your findings, with your models," she said, speaking at the HIMSS Media Machine Learning and AI for Healthcare event in Boston on Thursday.

In a recent pilot of about 500 patients, Aurora Advocate, which has an ACO with a number of value-based care contracts, was able to reduce unnecessary utilization for heart failure patients by 23 percent using well-timed telephonic interventions triggered by predictive analytics.

But Esposito says the key to success wasn't merely the AI, but a seamless marriage between analytics and operations.

"Yes the data's important, the model's important, and we want to talk about some of that advanced analytic work and the resources that it takes and the large amounts of data we can ingest and work with. But those alone will not make the difference," she said. "You have to insure you're working very close to the business. … This tight-knit relationship with operations is important because more often than not the findings that we're getting from advanced analytics require some sort of change in how we approach our work, and it's really important to have the subject matter experts to help guide that."

For example, the original plan for the pilot was to use predictive analytics to select patients with seven different conditions for specific and prescriptive telephonic care management, as opposed to more general outreach that was happening previously. But once the data came in, it became clear that the timetable of that outreach was also important.

"We were able to predict patients showing up to our hospitals with any one of these seven conditions. So we were excited. What we were surprised at is that they were working so well that as soon as we identified these patients as at high risk for some level of utilization, they were actually showing up at our front door."

The predictions were coming true much faster than the originally planned interventions would have intervened.

"At the time we would wait up to three weeks to make contact with the patients," Esposito said. "So this was telling us that we needed to change the paradigm so that we were reaching out to patients immediately, because if we didn't they would show up in the hospital."

Ultimately, 55 percent of patients in the pilot fully completed the telephonic care management program and stayed with the program an average of 60 days.

Esposito said that another benefit of their highly specific predictive analytics, powered by comparing 400 variables in a data lake of 20 million patients, is that it allowed them to target interventions that would best help particular patients.

"We have a lot of programs, and we're very focused on insuring we have the right patients in each of those programs," she said. "It's very easy to come up with a solution and assume everyone is going to benefit from it, but we've seen over and over that you need to target that intervention to get the outcomes that you're after."

Ultimately, to reach value-based care, care has to be preventative and targeted. For Advocate Aurora, AI helped with both those goals.

"We have to work together and this notion of getting patients to self-manage their condition has to be part of the solution." Esposito said. "And we found this was an important way to be able to consistently deliver on that time and time again."