Predictive Analytics Help Christiana Care Identify At-Risk Patients

Terri Steinberg, M.D., chief health information officer and vice president of population health informatics at Wilmington, Delaware-based Christiana Care Health System, had a novel idea about three years ago: using predictive analytics to identify at-risk patients with ischemic heart disease.  

Steinberg told the opening session of Health Data Management’s Healthcare Analytics Symposium in Chicago on Monday that medicine unfortunately follows a “cookbook” approach to heart failure in which clinicians “do the same things for everybody.” What’s needed is to leverage population health management through the use of data algorithms to “identify individuals who need more care, who need less care, and to deliver it,” she added, which is particularly difficult when tracking large patient populations.

Also See: How Analytics Takes the Unknowns Out of ACOs

Towards that end, Christiana Care successfully applied for a grant from the Center for Medicare and Medicaid Innovation to help fund a population health project that risk stratifies patients in real time. As a result, instead of relying on static plans of care, patients at the health system are now actively managed by a multi-disciplinary team reacting to data-driven tasks and triggers based upon complex algorithms executed in an analytic database.

This requires a lot of data, she added. Christiana Care leverages its regional health information exchange, the statewide Delaware HIE, as well as admit, discharge, transfer (ADT) data, electronic health records data, community cardiology registries, and financial claims. Next year, the health system plans to add home biometric data as well as non-health data, which Steinberg calls a “game changer.”

“The goal is to identify the patterns and notify somebody who can do something about it, whether it’s a care coordinator or social worker or pharmacist,” Steinberg said. “Analytics will drive care coordination tasks.”

According to Steinberg, the health system’s technology infrastructure includes a data analytics system called Neuron from Coldlight Systems and a population health EHR from Medecision. For its part, Neuron rapidly monitors anonymous patient data to identify predictive patterns for patients most at risk for disease complications, hospital readmissions and high-cost events or treatments.

“Every time a piece of data hits the data analytics system it pings the population health EHR and essentially generates lists of patients who are at risk,” she said. “No news is good news. If you have a patient with ischemic heart disease—for example—and the system is not telling you that they need anything, it’s assumed that they don’t.”

Steinberg concluded that machine learning is better than cardiologists at predicting what patients are at risk. “The long and the short of it is that we have people in front of computers who are constantly being told who’s in trouble, or who’s likely to get in trouble, by the system,” she commented. “We think, as clinicians, we know how to take care of patients, but the analytics that we have running today do it much better.”