Carolinas HealthCare uses analytics to prepare for future

BOSTON--Leadership at Carolinas Healthcare System looked at HIMSS Level 7 facilities in an effort to find out how they could get there as quickly as possible, said Michael Dulin, MD, PhD, chief clinical officer for analytics and outcomes research, speaking at the Big Data Healthcare Analytics Forum on Nov. 20.

The organization has 7,800 beds, more than 62,000 employees and about 1.4 pedabytes of data. In a “bold move,” they decided to take people from all across the organization and bring them into a single unit to build an analytic competency. That unit now has 110 people covering five different domains including client services and project management. It also includes research and development to evaluate projects. Making sure efforts are making a difference is important as is providing information back to providers so they know the work they’re doing is making a difference.

Carolinas has built a segmentation model to determine how best to use the data competencies, he said. They took three years of patient data—more than 2.2 million patients—with about 2,000 variables each and identified seven segments. “That flips the whole model of healthcare delivery on its head. In the past, we built the system and people came to us. Now, we see what the data shows and how we can rebuild the delivery system around these segments,” said Dulin. They built the warehouse and brought in all kinds of data including clinical, consumer and claims and created a data asset called Panorama.

The system learned that .6 percent of their population had the highest utilization and 24 percent of patients account for 76 percent of billed charges. That data help them understand when to start negotiating with payers and the impact of bringing a new practice on board.

All elements in Panorama are geocoded which helps Carolinas understand variations in costs across communities and how they can redeploy the care system to better manage risk.

The readmission risk model makes inpatients a population themselves, Dulin said. Case managers used to go room to room but now that they can identify those at highest risk, case managers can better tailor intervention and spend more time with those patients.

The enterprise data warehouse is “foundational,” said Dulin. “It’s important to have a conceptual architecture and surface it back.”

Using a standard, commercially available model helped Carolinas accelerate the process. Data scientists are there to use the data not provision data, he said. He also recommended data governance as an early investment. “This was huge for us. We didn’t people calling me and just making it my responsibility.” The team backs him up to analyze requests and make sure they align with organizational strategy.

Carolinas built an electronic asthma action tool to address higher rates of hospitalizations for exacerbations. They turned a 400-page guideline into a tool to use at the point-of-care in less than one minute. The tool also helps provide the care plan back to the patient. There are more than 6,000 different provisions of the asthma action plan. Despite that complexity, the system reduced hospitalizations by about 60 percent for children and any overall treatment for asthma exacerbation by 40 percent. Those figures were sustained at the 12-month period as well. With each hospitalization costing about $4,000, “there is a huge potential financial implication when considering more than 60,000 people with asthma.” The reduced rates also led to other benefits such as improved school attendance.

Carolinas has been working on predicting length of stay but that is “more of an operational issue,” Dulin said. They noticed, for example, that physical therapists conducted patient visits on a first-in-first-out basis. “We flipped that process and said, ‘here are the high-risk patients at risk of prolonged stay. Put them in the queue first.’”

Although the time already spent in the hospital is the biggest predictor of longer stays, Dulin said they had to think through the time component and take it out so it wasn’t an overwhelming predictive component. They also look at patients’ journeys between different segments and what predictive elements indicate someone moving from one to another and what interventions are appropriate and how they can prevent patients from developing higher cost conditions.