Automated Interpretation of Medical Prescription Text, 10-R9642

Printer Friendly Version

Principal Investigator
William L. Arensman

John D. Madrid
Nathan A. Price
Keith S. Pickens
Jeffrey A. Wilkinson

Inclusive Dates:  07/01/06 – 07/02/07

Background - Computerized physician order entry (CPOE) systems have been shown to reduce the number of serious medication errors when implemented and used properly. Nevertheless, many physicians have been reluctant to adopt CPOE, citing a steep learning curve and a lack of efficiency. This perceived lack of efficiency is often attributed to the user interface, which invariably provides for entry via a series of blanks on form-like screens.

Approach - This project explored an alternative process in which the physician enters traditional prescription text into a user interface that is analogous to a prescription pad. The text is then automatically interpreted into a standardized electronic format, resulting in a prescribing process that is more similar to past prescribing practices than CPOE systems currently available. Prescription text for analysis was gathered from physicians who were asked to write samples of typical prescriptions to treat hypothetical diagnoses. Proof-of-concept software was developed that:

  • Actively assists the physician, providing immediate feedback that the computer is interpreting the prescription properly with each keystroke.
  • Provides the most relevant prescribing data at a glance.
  • Rapidly classifies partial terms so that the user experiences no perceivable delay.

Accomplishments - The project met or exceeded each of its objectives. The parsing technology exceeded the project success goal by parsing 91 percent of the sample prescriptions without errors. An additional 8 percent of the prescriptions were parsed with only one or two errors, and 1 percent had three or more errors. The project met its other success criteria by incorporating several techniques that allowed it to "learn" or improve with use. The technology:

  • Offers more frequently used terms higher in the list of auto-completion choices.
  • Improves its auto-completion choices by tracking frequently used prescription patterns.
  • Incorporates new patterns into its parser on a per-user basis to improve with use.

This project has shown that this new user interface concept is viable and provides an avenue to enhanced CPOE adoption. As a direct result of this project, SwRI now has the capability to demonstrate the technology and pursue external projects in the areas of CPOE, natural language processing, and physician interface research.

Figure 1. Physician order Natural Language Processing Concept

Figure 2. Proof-of-Concept

2007 Program Home