A Semantic Annotation Framework for Patient-Friendly Electronic Discharge Summaries

Abstract

Discharge summaries are intended to include information necessary to communicate the post-discharge framework of care to care providers as well as patients and their families. An important aspect is the availability of easily understandable discharge information to empower patients as partners in their post-discharge care. However, these summaries are found to impose comprehension barriers for consumers. We explore semantic annotation as an approach to improve discharge summaries by assigning links of various semantic types to entities in the text. Our approach is grounded in automated text analysis and panel assessment of a corpus of 200 Electronic Discharge Summaries (EDSs) to identify the barriers to patient use of these summaries. These analyses identified the presence of advanced clinical vocabulary, abbreviations and inadequate patient advice as major obstacles. In response to the findings from corpus analyses, we implemented two components, SemLink and SemAssist. Both of these components use the Unified Medical Language System (UMLS) and the Open Access Collaboratives' Consumer Health Vocabulary (CHV) as biomedical vocabularies and the General Architecture for Text Engineering (GATE) as a natural language processing framework. SemLink is designed to provide readability support for EDS text by adding hyperlinks to the most appropriate and readable consumer-based web resource for difficult terms and phrases. SemLink was developed iteratively and can embed its results in portable document format (PDF). In a preliminary automated evaluation, SemLink achieved 95% precision in hyperlinking topically relevant Web resources in which 83% of hyperlinks could be restricted to resources of reading grade-level 8 or less. In the final evaluation by expert feedback, SemLink generated 65% topically relevant hyperlinks as agreed by the majority of the experts. SemAssist is designed as an interactive ontology-based Clinical Decision Support System to assist EDS authors in providing optimal medication advice for patients. The system offers a pre-formulated auto text and an alert critique about the inclusion of advice on side effects, required patient actions and follow-up related to postdischarge care for a set of high risk medications. Together, SemLink and SemAssist illustrate the application of a semantic annotation framework to support consumers in getting the most from their EDSs by exploiting both dynamic hyperlinking and authoring support. Our approach may have a wider range of applications to support other health-related document types and clinical users.