Divita Guy

There are a number of Natural Language Processing (NLP) annotation and Information Extraction (IE) systems and platforms that have been successfully used within the medical domain. Although these groups share components of their systems, there has not been a successful effort in the medical domain to codify and standardize either the syntax or semantics between systems to allow for interoperability between annotation tools, NLP tools, IE tools, corpus evaluation tools and encoded clinical documents. There are two components to a successful interoperability standard: an information and a semantic model.

Objective

The Consortium for Healthcare Informatics Research, a Department of Veterans Affairs (VA) Office of Research and Development is sponsoring the development of a standard ontology and information model for Natural Language Processing interoperability within the biomedical domain.

Medically unexplained syndromes (MUS) are conditions that are diagnosed on the basis of symptom constellations and are characterized by a lack of well-defined pathogenic pathways. The three most common MUS are chronic fatigue syndrome, irritable bowel syndrome, and fibromyalgia. Different types of persistent symptoms, originating from different organ systems, characterize these syndromes. Patients often meet the criteria for more than one MUS.

Objectives

We sought to develop a guideline and annotation schema that can be consistently applied to identify MUS found in VA clinical documents. These efforts will support building a reference standard used for training and evaluation of a Natural Language Processing system developed for automated symptom extraction. Our overarching goal is to characterize the occurrence of MUS in Operation Enduring Freedom/Operation Iraqi Freedom veterans.

Pro-WATCH (protecting war fighters using algorithms for text processing to capture health events), a syndromic surveillance project for veterans of operation enduring freedom (OEF)/operation Iraqi freedom (OIF), includes a task to identify medically unexplained symptoms (MUS). The v3NLP entity extraction tool is being customized to identify symptoms within VA clinical documents, and then refined to assign duration. The identification of medically unexplained symptoms and the aggregation of this information across documents by patient’s is not addressed here.

Objective

Pro-WATCH (protecting war fighters using algorithms for text processing to capture health events), a syndromic surveillance project, includes a task to identify medically unexplained symptoms. The v3NLP entity extraction tool is being customized to identify symptoms, then to assign duration assertions to address part of this project. The v3NLP tool was recently enhanced to find problems, treatments, and tests for the i2b2/VA challenge. The problem capability is being further refined to find symptoms. Machine learning models will be developed using an annotated corpus currently in development to find duration assertions.

The main stay of recording patient data is the free text of electronic medical records (EMR). While stating the chief complaint and history of presenting illness in the patients ‘own words’, the rest of the electronic note is written by the provider in their words. Providers often use boiler-plate templates from EMR pull-downs to document information on the patient in the form of checklists, check boxes, yes/no and free text responses to questions. When these templates are used for recording symptoms, demographic information or medical, social or travel history, they represent an important source of surveillance data [1]. There is a dearth of literature on the use of natural language processing in extracting data from templates in the EMR.

Objective:

To highlight the importance of templates in extracting surveillance data from the free text of electronic medical records using natural language processing (NLP) techniques.