Talk to the Veterans Crisis Line now
An official website of the United States government
Author List:
Foulis PR (James A. Haley VAMC)
Tremblay M (Uninvsersity of South Florida)
Luther SL (VISN 8 Patient Safety Research Center)
Berndt DJ (Uninvsersity of South Florida)
French D (VISN 8 Measurement Evaluation Team)
Powell-Cope G (VISN 8 Patient Safety Research Center)
Objectives:
Knowledge Discovery is defined as the non-trivial, potentially useful, and ultimately understandable patterns in data. Data mining is a systematic process that can be applied to health care to discover knowledge. This study focuses on exploring the computerized medical record using data mining applied to identify patient fall-related injuries (FRI). FRI is a high cost, high volume adverse event in the VA, but is difficult to identify from VA administrative databases. Objectives:
1. Evaluate text-mining tools and other knowledge discovery techniques for use with VA electronic medical records.
2. Use these tools to predict the presence of a FRI in data extracted from the Tampa VA electronic medical records.
Methods:
A literature review was completed to identify the state of art in data and text mining software for application to health care. A list of all services provided to patients with injuries at the Tampa VA (FY01–03) was identified by searching the Ambulatory Events Database. The electronic medical records for the unique patients identified were extracted, organized into a database and de-identified for analysis. We conducted key phrase extraction from progress notes, radiology reports, medication records, surgical reports, and discharge summaries. Based on these results we employed data mining tools to build models of key phrases to predict the presence of a FRI.
Results:
A description of available tools that are appropriate for knowledge discovery using electronic medical records will be provided. Results of the text mining will be organized by main concepts, relative frequency of each concept, frequency of each concept co-occurrence, centrality of each concept, and similarity in contexts in which each concept occurs. Information about these phrases will be used to form clusters for subsequent data mining efforts. Machine learning and statistical models to predict FRI will be presented with key phrases and patient attributes as independent variables.
Implications:
Preliminary results suggest that knowledge discovery techniques may be useful in identifying FRIs which typically are under-coded in administrative data.
Impacts:
Results of this study provide new information about FRI in the VA system. More importantly however, knowledge discovery techniques employed here represent an example of how the tremendous resource of the VA electronic medical record can be better utilized. These techniques have potential for broad application in both research and management.
