Shaan-Chirag Gandhi DPhil, Presenter: Nothing to Disclose

Roy Gordon Bryan MD, MBA, Abstract Co-Author: Nothing to Disclose

Sarita Nair MS, Abstract Co-Author: Nothing to Disclose

Abraham Lin BS, Abstract Co-Author: Nothing to Disclose

Arun Krishnaraj MD, MPH, Abstract Co-Author: Nothing to Disclose

Clinical practice increasingly relies upon imaging to provide rapid complementary data in the care of patients. While a brief clinical history often accompanies an imaging request, detailed knowledge of a patient’s past medical history and presenting symptoms often enhances study interpretation. Searching the electronic medical record (EMR) manually is time-consuming and may lead to lower-quality, less-efficient interpretations due to overlooking relevant history. To address this difficulty, we describe a process for developing customized search queries of the EMR built upon the Queriable Patient Inference Dossier (QPID) health record intelligence platform at the Massachusetts General Hospital.

Through literature reviews and interviews with referring providers, a list of relevant past medical history search parameters specific to three MRI exams (liver, prostate and rectal) was developed. Sixty-two patient records selected at random were searched across nine liver, prostate and rectal MRI search algorithms covering relevant past imaging, laboratory values, medications, and notes (see figure). Two independent reviewers compared the QPID-driven search results to their manual EMR review to assess positive and negative predictive values. In addition, a graphical user interface (GUI) incorporating interpretation guidelines was developed and presented to the end user to assist in image interpretation.  

The average search time per query was 3.4 ± 1.1 seconds and the inter-observer agreement between reviewers (Cohen’s κ) was 0.90. The pooled average positive predictive value (PPV) was 0.86 and negative predictive value (NPV) was 0.91 across all three exam types. For critical searches, such as medication lists or pathologic diagnoses, the PPVs and NPVs for individual exams approached unity.

This study demonstrates and validates the utility of constructing automated search queries of a patient EMR and displaying results within a GUI to optimize clinical data gathering for use in enhancing speed and quality of image interpretation. Future directions include a prospective demonstration of the impact of QPID-based searches on the efficiency and quality of imaging interpretation.

Gandhi, S, Bryan, R, Nair, S, Lin, A, Krishnaraj, A, Design and Validation of Automated, Customized Clinical History Searches for Imaging Interpretation.  Radiological Society of North America 2013 Scientific Assembly and Annual Meeting, December 1 - December 6, 2013 ,Chicago IL. http://archive.rsna.org/2013/13016235.html