Patient characteristics derived from electronic medical records (EMR) can predict the risk of readmission after percutaneous coronary intervention (PCI), a study suggests.

The study, published online August 18 in Circulation Cardiovascular Quality and Outcomes, found that anticoagulation, emergency department visits, and anxiety were independently associated with 30-day readmission after PCI.

Early readmission after PCI accounts for an estimated $359 million per year in costs to Medicare alone. Preventing such readmissions could reduce health care costs, but existing risk prediction models only modestly discriminate between patients likely and unlikely to require readmission.

“Many factors associated with readmission are not modifiable per se—but they still can be useful to know about,” lead author of the study, Robert Yeh, MD, MBA, from Massachusetts General Hospital in Boston wrote in email to Reuters Health. “Models like these can help physicians identify patients who are at high risk for returning to the hospital and interventions to reduce this risk can be selectively tested and applied to this group.”

Dr. Yeh and colleagues conducted a case-control study of patients in the Partners PCI Readmission Project, to investigate the association between candidate prognostic variables and 30-day readmission after PCI.

The researchers used natural language processing (NLP) software to identify, extract, and analyze patient information stored in the EMR as unstructured data.

Multivariable logistic prediction modeling identified three independent variables that predict readmission: number of emergency department visits, use of anticoagulation, and history of being described as anxious.

“The anxiety finding in particular reminds us that to deliver truly excellent care for patients with coronary disease, we need to take care of the whole patient,” coauthor of the study, Jason Wasfy, MD, from Massachusetts General Hospital, wrote in email to Reuters Health. “Performing the angioplasty correctly and prescribing the proper medications is necessary but not sufficient.”

“Given the increasing focus on health IT, ‘big data,’ and using sources of unstructured clinical data, effective methods of collecting and analyzing this data is needed,” Thomas Maddox, MD, from the University of Colorado School of Medicine in Denver, who coauthored an editorial related to this report, told Reuters Health via email. “NLP is a potentially promising method to do this with EMRs, as demonstrated in this study.”

Funding for the study was provided by the Hassenfeld Scholars Program and a grant from the American Heart Association.