Skip to main content

Teaching EHRs to understand natural language questions about patients

Wednesday, September 24, 2014 — Poster Session IV

10:00 a.m. –12:00 p.m.

FAES Academic Center




  • K Roberts
  • D Demner-Fushman


As electronic health records (EHR) continue to grow, efficient access to the large quantities of data they contain becomes more important. Instead of navigating through the EHR, doctors often utilize textual search functionality to quickly retrieve relevant patient records. However, this still requires analyzing the various search results, many of which may not be relevant to the doctor’s information needs. Instead, we investigate the feasibility of utilizing natural language questions, which can encode very precise information needs. Examples include, “What was the patient’s highest glucose reading?” and “Has the patient taken any sedatives in the past 24 hours?” To be able to process EHR questions, natural language processing (NLP) methods must be used to convert the unstructured question into a structured query. To this end, we are investigating the feasibility of representing natural language EHR questions in a structured query, specifically logical forms with lambda expressions. We have annotated a sub-set of EHR questions posed by doctors, collected in Li [2012], with logical forms, including medical codes for relevant medical concepts. The work demonstrates the structured representation of EHR questions is feasible, and lays the groundwork for automatic NLP methods capable of a deep understanding of EHR questions.

back to top