Talk to the Veterans Crisis Line now
U.S. flag
An official website of the United States government

Health Services Research & Development

Veterans Crisis Line Badge
Go to the ORD website
Go to the QUERI website

PPO 15-406 – HSR&D Study

New | Current | Completed | DRA | DRE | Portfolios/Projects | Centers | Career Development Projects

PPO 15-406
Towards Choosing Wisely: Assessing Reason for Exam in Carotid Ultrasound Reports
Wendy W. Chapman PhD
VA Salt Lake City Health Care System, Salt Lake City, UT
Salt Lake City, UT
Funding Period: July 2016 - June 2017

BACKGROUND/RATIONALE:
Carotid images are ordered in the VA for patients with asymptomatic carotid stenosis. Evidence supports very few indications for ordering images, yet over 200,000 carotid images are ordered. Low-value imaging can lead to risky interventions and divert resources to tests and procedures that may not improve veteran health. The Choosing Wisely program aims to decrease unnecessary procedures, tests, and medications ordered by physicians in the US health care system with carotid imaging being identified as a low-value test. Understanding the reason that physicians give for ordering the exam is the first step in decreasing low-value carotid imaging.

OBJECTIVE(S):
Aim 1a-c. Retrospectively catalogue and classify indications for ordering an ultrasound carotid exam.

Aim 2a-b. Evaluate our ability to automatically identify indications for exam using natural language processing (NLP).

METHODS:
Methods: We randomly sampled training and validation datasets for developing and evaluating the NLP tool. Using the training dataset, we developed a vocabulary of synonyms for each indication. We selected synonyms for indications from the Unified Medical Language System using Knowledge Author. We developed a series of rules to remove synonyms unlikely to occur in free text based on their characteristics e.g., semantic type "chest pain (finding)", NOS "Memory Loss NOS", etc. We applied the remaining synonyms to reason for exam fields using the pyConText algorithm and conducted an error analysis to expand the vocabulary based on synonyms, such as lay terms ("seeing stars" as blurred vision/change in vision), misspellings ("hollehurst plaque" as hollenhorst plaque), acronyms ("aion" as anterior ischemic optic neuropathy), etc. Using the validation dataset, we answered the research questions for each Aim.

FINDINGS/RESULTS:
(1a) How often are indications for exam provided? In the validation dataset, all exams were justified with one or more indications (n=600 patients; 771 indications total).
(1b) Where are indications for exam located in the patient record? Of these indications, about 65% occurred in the reason for exam field and about 35% occurred in the RAD/TIU notes.
(1c) What are the different indications for exam provided by physicians (most common)? Of the 24 indication types observed, the most frequent indications (count, percentage) included: carotid bruit: 225 (29%), stenosis/history of carotid disease: 155 (20%), hypertension with another risk factor: 97 (13%), dizziness/vertigo: 59 (8%), history of stroke/transient ischemic attack: 27 (3%), and blurred vision: 22 (3%).

(2a) How well can we identify indications for exam and map the text describing the indication to a set of standardized categories within the reason for exam field? pyConText identified these most common indications from the reason for exam field with reasonable to excellent performance (F1-score, recall, precision): carotid bruit: (92%, 91%, 94%), stenosis/history of carotid disease: (66%, 81%, 56%), hypertension with another risk factor: (67%, 64%, 70%), dizziness/vertigo: (85%, 85%, 85%), history of stroke/transient ischemic attack: (51%, 73%, 39%) and blurred vision: (61%, 44%, 100%).
(2b) Within reports? Our dataset contained imaging orders within a broad time interval, and we were not able to link the expert-generated reasons for exam to the reports. We would need to create a new reference standard dataset to complete this aim.

IMPACT:
Our long-term goal is to decrease low-value imaging in the VA by detecting the indication for exam and providing feedback through techniques such as facility-level report cards. This work supports this goal by 1) increasing our knowledge of indications provided by physicians for carotid ultrasound exam and 2) developing a baseline tool that could be implemented as a part of sustainable, long-term, scalable program for decreasing low-value imaging within the VA.

PUBLICATIONS:

Journal Articles

  1. Mowery D, Smith H, Cheney T, Stoddard G, Coppersmith G, Bryan C, Conway M. Understanding Depressive Symptoms and Psychosocial Stressors on Twitter: A Corpus-Based Study. Journal of medical Internet research. 2017 Feb 28; 19(2):e48.
  2. Shah RU, Rupp AB, Mowery D, Zhang M, Stoddard G, Deshmukh V, Bray BE, Hess R, Rondina MT. Changes in Oral Anticoagulant Treatment Rates in Atrial Fibrillation before and after the Introduction of Direct Oral Anticoagulants. Neuroepidemiology. 2017 Jan 31; 47(3-4):201-209.
  3. Chapman AB, Mowery DL, Swords DS, Chapman WW, Bucher BT. Detecting Evidence of Intra-abdominal Surgical Site Infections from Radiology Reports Using Natural Language Processing. AMIA ... Annual Symposium proceedings. AMIA Symposium. 2017 Jan 1; 2017:515-524.
Center Products

  1. Chapman WW, Mowery DL. Web-based Evaluation Workbench. 2017 Jul 1.


DRA: Cardiovascular Disease
DRE: Diagnosis, Treatment - Implementation, TRL - Applied/Translational
Keywords: Decision Support, Natural Language Processing, Technology Development
MeSH Terms: none

Questions about the HSR&D website? Email the Web Team.

Any health information on this website is strictly for informational purposes and is not intended as medical advice. It should not be used to diagnose or treat any condition.