A New Way of Representing Clinical Reports for Rapid Phenotyping

With the introduction of comprehensive electronic medical records (EMRs), all aspects of patient care can now be captured in both structured and free-text format. The existence of such data provides the opportunity to improve patient care and to facilitate clinical and translational studies of large cohorts of patients. Currently, the data are accessible only using simple keyword-based methods, while identifying phenotypes requires manual chart review, a time- and resource-intensive process. In this abstract, we present an alternative indexing approach for clinical text in EMRs based on UMLS concepts augmented with the section information (e.g., history of present illness, past medical history) as well as their assertion values (e.g., present, absent, conditional) to represent the context.