Candidate for the Lee B. Lusted Student Prize Competition
Purpose: For patients with psychiatric disorders, clinician notes are the principal source of information about the course and severity of their illness. A vocabulary framework is a structured representation of single words and short phrases used clinically when documenting patient encounters. Such a framework is an essential first step to developing automated tools for extracting clinical information from electronic medical records. We developed a framework that defines the vocabulary used by clinicians who provide clinical care to patients suffering from post traumatic stress disorder (PTSD). The purpose of this study was to learn what data sources are needed to create a comprehensive vocabulary framework.
Methods: A set of complementary data sources were used to construct a framework and classify terminology about the following clinical domains: change in symptoms, prescribed treatments, terms related to resiliency, and terms related to suicidality. Data from the following data sources were included: terms derived from published PTSD treatment guidelines, focus groups with mental health clinicians from five VA medical centers, automated text mining of 100 progress notes, manual annotation of terms from approximately 1,000 outpatient progress notes, and terms derived from SNOMED-CT.
Results: A total of 158 terms were derived from the PTSD treatment guidelines, 172 terms resulted from SNOMED-CT, 43 were identified in the focus groups, and another 20 resulted from text mining. However, 985 unique terms were identified from annotation of the progress notes. Of the unique terms, 397 terms were included in 4 symptom categories (re-experiencing, avoidance, arousal, and general) and 219 terms were included in 3 treatment categories (counseling, pharmacotherapy, and general).
Conclusions: Systematic annotation of outpatient progress notes was a rich source of clinical terms. The range of terms used by individual clinicians is broad, with multiple synonyms occurring in nearly all categories of symptoms and treatments. Existing structured vocabularies do not capture most terminology used by clinicians for care of patients with PTSD. In this clinical domain, development of a vocabulary framework from actual clinical text has great promise. Future studies will focus on incorporating this vocabulary into a PTSD term identification tool that can be used in chart abstraction, text mining, and clinical decision support development.
See more of: The 33rd Annual Meeting of the Society for Medical Decision Making