Tool-supported Interactive Correction and Semantic Annotation of Narrative Clinical Reports

Area of Science:

• Medical Informatics
• Natural Language Processing
• Clinical Data Management

Background:

• Narrative clinical reports in electronic health records require structured processing for secondary use.
• Physicians and automated systems benefit from easier, less erroneous data interpretation.
• Challenges exist in extracting structured information from narrative text, especially in less-resourced languages.

Purpose of the Study:

• To design a method and software for interactive correction and semantic annotation of narrative clinical reports.
• To facilitate easier and less erroneous processing of clinical reports by unfamiliar physicians and decision-support systems.
• To gain insights into clinical report creation and annotation processes, and to provide a ground-truth dataset for automated transformation training.

Main Methods:

• A three-phase preprocessing method: tokenization, normalization, and enrichment with extracted structured information.
• Development of software tools for interactive correction, expansion, and semantic annotation of narrative clinical reports.
• Validation of the method in the cardiology domain using Czech narrative clinical reports.

Main Results:

• The three-phase preprocessing method was validated on 49 Czech narrative clinical reports in cardiology.
• Two cardiologists annotated 1500 clinical terms, linking them to classification systems (ICD 10, SNOMED CT, LOINC, LEKY).
• A database of correct clinical terms and codebook terms was established.

Conclusions:

• Structured information was successfully extracted from Czech narrative clinical reports using the proposed method.
• The developed software, tailored for Czech, can serve as a model for adapting the approach to other less-resourced languages.
• Extracted structured information supports medical decision-making, quality assurance, and further research.