Using the natural language processing technology that underpins IBM Watson, Carilion Clinic was able to identify 8,500 patients who are at risk of developing congestive heart failure within one year. The pilot project, which began in October, took only six weeks to wrap up. “We look at this as the tip of the iceberg–the power of predictive modeling and natural language processing,” says Steve Morgan, chief medical information officer at Carilion, which operates eight hospitals in southwest Virginia.
Hospitals spent millions implementing electronic health records, and are now under pressure to harness data to improve patient outcomes and lower costs, as reimbursement shifts from fee for service to payment based on performance. Carilion is part of an accountable care organization, where it stands to reap financial rewards by intervening to keep patients out of the hospital.
As a result, the business of predictive analytics in health care has exploded. Hospitals that will benefit most are those with rich and comprehensive patient data.
Carilion, which uses IBM’s data warehouse, and Epic electronic health records, collaborated with both companies to catch patients before they develop heart failure, a common cause of hospitalization. Despite the electronic data available at a doctor’s fingertips, such as diagnoses, medications, and lab tests, details are often buried in clinical notes. “It’s a challenge for physicians to aggregate all that data,” says Morgan.
IBM gathered three years worth of data belonging to 350,000 patients. In addition to more than 200 factors such as blood pressure, beta blocker prescriptions, and weight, it combed through more than 20 million notes, uncovering nuggets of information that are not entered in a medical record’s fields. They include the number of cigarette packs a patient smokes, the pattern of prescriptions, and how well the heart is pumping. Additional details that might have escaped a doctor’s eye, include a patient’s social history, depression, and living arrangements.
Predictive algorithms uncovered 8,500 patients at risk of having heart failure within a year; 3,500 were ferreted out because of natural language technology.
Epic helped extract the data, and is working with Carilion and IBM to display the information, which scores a patient’s risk, to health care providers.
The next step is to track those patients. “We’re going to see if intervention improves outcomes,” says Morgan.
http://www.forbes.com/sites/zinamoukheiber/2014/02/19/ibm-and-epic-apply-predictive-analytics-to-electronic-health-records/
Comments