Abstract
NLP technologies can play a significant role in health-care where a predominant segment of the clinical documentation is in text form. In a graduate course focused on understanding semantic web services at West Virginia University, a class project was designed with the purpose of exploring potential use for NLP-based abstraction of clinical documentation. The role of NLP-technology was simulated using human abstractors and various workflows were investigated using public domain workflow and semantic web service technologies. This poster explores the potential use of NLP and the role of workflow and semantic web technologies in developing healthcare IT environments.
1 Workflow Summary
The workflow explored in this project was the following:
Clinical documents that are marked up Clinical Document Architecture (CDA) release 2 format are submitted to a workflow engine.
The workflow engine routes the document to be archived in a repository
The document is then routed to an abstraction web service to allow data abstraction to happen. The abstracted elements are archived as CDA document as well.
Availability of abstracted documents triggers three different web services. A coding web service attempts to use the abstracted elements to code the document. Another web service evaluates the abstraction for core measures defined by JCAHO. A third reviews the medication abstraction to run drug-drug interactions. The following short description provides additional details about each web service.
2 Summary of each web service
2.1 Archival web service
This supported the notion of storing CDA documents and allowed for retrieving them.
2.2 Abstractor Web Service
The goal of this service is to simulate NLP technology but using a human abstractor, to take a look at narrative text and fill out a structured template. The abstractor module is notified with the availability of a new document.
2.3 Coding Web Service
The goal of this service is to use the abstraction module to drive the rules for exercising the selection of particular ICD-9 codes using the Protégé ontology tool.
2.4 Core Measures Web Service
The goal of this service is to use an encoding of the core measures rules using Protégé and to apply them to the abstracted document.
2.5 Drug-Drug Interactions
The goal of this web service is meant to take a look at the medications that a person is taking and determine if they can cause any potential adverse reactions.
3 Technology Choices
The class project utilized a variety of open-source and publicly available java-based technologies: Taverna was the workflow engine used, Apache and Tomcat provided the web service infrastructure, and Protégé provided the knowledge tools for the web services built.
4 Conclusion
The 18 students were divided into groups of three where three of them will develop a set of coordinated web service focused on specific areas such as Acute myocardial infraction (AMI), Heart Failure (HF) etc. The prototypes underscored the fact that the abstraction process (and one that uses NLP technology as well) needs to be matched to the use of the information down stream. A simple attribute-value abstraction will not work and the abstraction actually must mirror an object model for the data abstracted.
Acknowledgments
I am grateful to the discussions on this front with both Dr. Shawn Murphy and Vipul Kashyup from Partners Healthcare. Jim Dierwechter, a student funded by Med-Quist, helped in supporting the execution of this complex experiment in class. And, last but not the least I am indebted to my students who suffered through a project definition that evolved and kept changing through out the Fall of 2005.