Identifying Key Concepts from EHR Notes Using Domain Adaptation

Linking electronic health records (EHRs) to relevant education materials can provide patient-centered tailored education which can potentially improve patients’ medical knowledge, self-management and clinical outcome. It is shown that EHR query generation using key concept identification improves retrieval of education materials. In this study, we explored domain adaptation approaches to improve key concept identification. Our experiments show that a 20.7% improvement in the F1 measure can be achieved by leveraging data from Wikipedia. Queries generated from the best performing approach achieved a 20.6% and 27.8% improvement over the queries generated from the baseline approach.