Overview - Clinical Decision Support 2016

Proceedings | Data | Results | Runs | Participants

In handling challenging cases, clinicians often seek out information to make better decisions in patient care. Typically, these information sources combine clinical experience with scientific medical research in a process known as evidence-based medicine (EBM). Information relevant to a physician can be related to a variety of clinical tasks, such as (i) determining a patient's most likely diagnosis given a list of symptoms, (ii) determining if a particular test is indicated for a given situation, and (iii) deciding on the most effective treatment plan for a patient having a known condition. Finding the most relevant and recent research, however, can be quite challenging due to the volume of scientific literature and the pace at which new research is published. As such, the time-consuming nature of information seeking means that most clinician questions go unanswered. In order to better enable access to the scientific literature in the clinical setting, research is necessary to evaluate information retrieval methods that connect clinical notes with the published literature. The TREC Clinical Decision Support (CDS) track simulates the information retrieval requirements of such systems to encourage the creation of tools and resources necessary for their implementation. In 2014 and 2015, the CDS tracks used simulated patient cases presented as if they were typical case reports used in medical education. However, in an actual electronic health record (EHR), patient notes are written in a much different manner, notably with terse language and heavy use of abbreviations and clinical jargon. To address the challenge specific to EHR notes, the 2016 CDS track used de-identified notes for actual patients. This enabled participants to experiment with a realistic topic/query and develop methods to handle the challenging nature of clinical text. For a given EHR note, participants were challenged with retrieving full-text biomedical articles relevant for answering questions related to one of three generic clinical information needs: Diagnosis (i.e., “What is this patient's diagnosis?”), Test (“What diagnostic test is appropriate for this patient?”), and Treatment (“What treatment is appropriate for this patient?”). Retrieved articles were judged relevant if they provided information of the specified type useful for the given case. The assessment was performed by physicians with training in biomedical informatics using a 3-point scale: relevant, partially relevant, not relevant.

Track coordinator(s):

• Kirk Roberts, The University of Texas Health Science Center
• Dina Demner-Fushman, U.S. National Library of Medicine
• Ellen M. Voorhees, National Institute of Standards and Technology (NIST)
• William R. Hersh, Oregon Health and Science University

Track Web Page: https://www.trec-cds.org/

---