Measuring the Quality of CDA to FHIR Transformations Using Text Mining Techniques.

Area of Science:

• Healthcare Informatics
• Data Science
• Natural Language Processing

Background:

• Data interoperability is a major challenge in healthcare informatics.
• Fragmented standards and complexity hinder secondary use of health data.
• Transformations between health data formats (e.g., HL7 CDA to FHIR) are complex and hard to validate.

Purpose of the Study:

• Define common transformation errors: incorrect logical mappings, missing data points, and faulty classifier translations.
• Present an algorithm for detecting these transformation errors.
• Assess transformation quality using text mining and similarity measures.

Main Methods:

• Developed an algorithm analyzing free-text similarity between input and output data.
• Utilized text mining techniques for data assessment.
• Generated statistical indicators: schema semantic similarity, equivalent values ratio, and value token count ratio.

Main Results:

• A prototype algorithm was implemented and validated against a structured test suite.
• The algorithm reliably detected the defined transformation errors.
• Identified that semantic similarity indicator sensitivity could be enhanced with advanced embedding models.

Conclusions:

• The developed algorithm effectively detects common errors in health data transformations.
• The text mining approach offers a viable method for assessing transformation quality.
• Further research can refine the algorithm and expand its application to diverse health data models and similarity assessment tasks.