Grady Health System in Atlanta, Georgia was able to save $4 million in prevented readmissions by using a combination of AI-driven predictive analytics and patient-level community interventions.

At the HIMSS Big Data and Healthcare Analytics Forum in Boston today, Robin Frady, executive director of business and clinical intelligence at Grady, described her team’s approach to AI when they started their project three years ago.

“We wanted to know not just that you’re at risk for readmissions, but we wanted to know why,” she said. “A lot of people will give you predictive models, but you don’t know what’s driving it. So we wanted to know why, based on your clinical factors, but also your socio-economic factors. Is it because you’re in a food desert? Is it because you’re too far away from the hospital to make it to your appointment? What is driving your high risk, and more importantly, what are the interventions that will move you — you specifically as a patient?”

That was the first lesson Grady’s team learned — to find a platform that would tell them not just who is high risk, but why. The second was to make sure that information was presented to the right people.

“First we integrated into Epic and we were showing it to people, but no one was really using it,” she said. “But then we realized something. We had this mobile integrated health service. These are our EMS folks. They respond to about 100,000 911 calls. This is a group that is busy, but they still have downtime. So we said, let’s go out and meet our patients where they are. … We realized, we have these patients at high risk for readmissions, what if we went and visited those patients?”

The pilot, which involved visiting only about 2,000 patients over two years, lowered readmission rates 31 percent from what was expected. They avoided 382 readmits, a total cost savings of just over $4 million.

“The best feedback we received was from our EMS folks who said ‘We felt we were getting to patients who were in trouble,’” Frady said. “They were not taking their meds. They were not eating. They were not doing what they needed to do, and they were going to be readmitted.”

Since then, Grady has started using similar analytics for following up with patients post-discharge and managing chronic care patients. The hospital system is just starting to look at senior populations and behavioral health populations as well.

The number one lesson, according to Frady? Put patients first.

“I really truly believe this: Other industries always focus on the customer, so we have to focus on the patient,” she said. “So however cool AI can be, it’s not as cool as making sure the focus is on the patient.”