Carolinas centralizes data analytics to reduce readmissions and redesign care

Reducing the number of patients who return to the hospital after discharge has long been a focus for executives and clinicians at Carolinas HealthCare System.

Carolinas clinicians had compiled nearly 600 variables, including disease history and social factors, for predicting which patients are likeliest to return for unplanned readmissions within 30 days and figuring out the best ways to avoid those events. The predictive task was harder because Carolinas and its affiliates make up a large, geographically sprawling organization that logs more than 10 million patient encounters a year at more than 40 hospitals in three states.

The publicly run health system based in Charlotte, N.C., made better identification of patients at risk for readmissions one of the first projects for its new centralized, in-house analytics department, which uses a 10-terabyte enterprise data warehouse to spot emerging trends. “This is where the ability to analyze data sets becomes very valuable,” said Dr. Allen Naidoo, vice president of advanced analytics for Carolinas. “We are able to risk-stratify patients. It allows clinicians to focus on those patients who are at the highest risk of coming back to the hospital.”

After database experts analyzed 18 months of patient encounters, they concluded that of the 600 or so readmissions variables, 40 were highly predictive of future readmissions. Factors that rated high included language barriers, sodium level, end-stage renal disease and a history of emergency department visits.

That information has allowed the in-house data division, known as Dickson Advanced Analytics, to develop a system that predicts readmission risk with 79% accuracy, using statistics that are updated hourly to reflect the patient record. The system has been used to assess more than 20,000 patients since July and to administer more than 30,000 care interventions.

The result is that care managers can now prioritize their work to focus on the patients at highest risk of readmissions. The system uses personalized interventions rather than one-size-fits-all solutions. “Those are things that everyone can and should be doing,” said Dr. Lyle Berkowitz, associate chief medical officer of innovation for Northwestern Memorial Hospital in Chicago and co-editor of the book Innovation with Information Technologies in Healthcare.

According to a May report by the Institute for Health Technology Transformation, high-level analytics is becoming a critical competency for health systems because economic and political forces are shifting financial risk for managing patient health onto providers. The biggest challenge is the meaningful integration of information from a wide spectrum of sources.

That means crunching the terabytes streaming in from patients' hospital records and integrating that with data from outpatient and clinic visits. Then provider systems need to integrate the clinical data with claims data from payers to facilitate analyses on total costs of care and the social determinants of health.

Carolinas has done all of that, but getting there was not easy. The first step was deciding to keep the analytics functions in-house rather than outsourcing the job. The system also decided to centralize its data analytics staff into a single department, Dickson Advanced Analytics.

Centralization meant departments had to give up staff members—not always an easy sell. “I still remember the day all the high-level executives of the system said we need to give for this, it's important for the future of the organization,” said Dr. Michael Dulin, chief clinical officer for analytics and outcomes research. “It was a bold decision.”

Carolinas spent an initial $5 million to build a data warehouse, Naidoo said, and other costs have mounted along the way.

Today, Dickson performs a wide range of data analyses. Each project is designed for a specific department, such as an initiative to reduce hospitalizations for asthma patients through better primary-care management, and a project to develop population-health models for geographic regions to prevent nonemergency hospital visits.

“We should think about not just readmissions, but all admissions to the hospital as things that we could avoid if we had access to appropriate analytics,” Dulin said.

Follow Joe Carlson on Twitter: @MHJCarlson