Analytics IDs Patients at Risk for Heart FailureBy Maggie O'Neill | Posted 2014-04-07 Email Print
WEBINAR: On-demand webcast
Next-Generation Applications Require the Power and Performance of Next-Generation Workstations REGISTER >
Carilion Clinic applies predictive analytics and natural-language processing to electronic medical records to identify patients at risk for heart failure.
A pilot project using predictive analytics and applied natural-language processing identified 8,500 patients of the Carilion Clinic who are at risk for congestive heart failure.
The analyzed data is being entered into the clinics' electronic medical records (EMRs) system, but it should enable physicians to provide flagged patients with earlier preventative care. That process is taking longer than anticipated, but is expected to reach completion within two months, according to Steve Morgan, MD and chief medical information officer at Carilion Clinic. The clinic, with numerous hospitals, outpatient clinics and urgent care facilities in Virginia, is also doing more data validation during this time.
To extract and analyze the data, Carilion partnered with Epic, whose EMR system they use, and IBM, which helped Carilion build an enterprise data warehouse two years ago. Records for about half of the clinic's one million patients were pulled and eventually parsed down to 350,000 for the pilot.
IBM analyzed both structured and unstructured data in medical records to identify patients at risk. Discrete data points, such as weight and medications, can be found in structured EMR fields. Unstructured data includes physicians' notes that are typed or read into a patient's EMR or discharge papers. Non-discrete data can include details on a patient's ejection fractures, a measure of their heart's functioning that was not captured 100 percent discretely, or even the mention of key terms related to congestive heart failure.
The natural-language processing technology searched for key words or phrases within the unstructured data. In all, 20 million documents were analyzed.
Because approximately half of all patients who develop heart failure die within five years, according to the Centers for Disease Control and Prevention, early identification is essential. About 3,500 of the 8,500 patients Carilion identified as at-risk would not have been found if the project had analyzed only the structured data, according to Morgan.
"I was a bit surprised at the total number," Morgan recalls, "and I was definitely surprised at how many patients we would have missed if we had used only discrete data."
Ever since IBM built the data warehouse for Carilion, the companies had been looking for ways to collaborate on data analysis related to chronic disease. The pilot suggests that applied technology can be used to improve health care outcomes for patients through earlier interventions, but it also can lead to cost savings for treatment and care, according to Morgan. The CDC reports that heart failure has a U.S. price tag of approximately $32 billion each year.
Morgan said the clinic is focusing on an "overall transformation of care" and that he is "definitely most interested in" having more conversations about the applications of predictive analytics and natural-language processing in the future. Within a couple of years, he anticipates that EMRs will improve real-time functioning and be able to interpret and provide feedback to physicians at the time they type in or dictate their notes.