IBM on Wednesday announced it had collaborated with Carilion Clinic, a healthcare system in Virginia, to identify patients at risk of developing heart failure in the next year.
In a pilot project, IBM’s software combed through electronic medical records searching for warning signs of heart failure. IBM’s natural language processing technology (also used in IBM’s computing system Watson, which won Jeopardy! in 2011) analyzed clinicians notes and discharge documents. These sources — often called “unstructured data” — aren’t often processed in predictive modeling based on health records.
Predictors of heart-failure risk included high blood pressure, use of certain prescription drugs, previous heart-related diagnoses, obesity and some lifestyle factors such as occupation, among others. The system had an 85 percent accuracy rate, according to the company, and identified about 8,500 people at risk for heart failure. IBM created and deployed the system within six weeks, and those identified could immediately be recommended for preventive care.
"We've learned that predictive analytics insights from both structured and unstructured data is imperative to meet our goal of improving patient care at lower costs," Steve Morgan, Carilion Clinic’s chief medical information officer, said in a statement.