[The analysis of CIRSmedical.de using Natural Language Processing]

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Natural Language Processing

Background:

• CIRSmedical.de is a German cross-institutional reporting and learning system established in 2005.
• It hosts over 6,000 event reports, with past analyses focusing on specific topics rather than a comprehensive evaluation.
• Natural Language Processing (NLP) offers a novel strategy for analyzing these text-based reports.

Purpose of the Study:

• To systematically evaluate all available case reports from CIRSmedical.de using NLP.
• To describe the characteristics of event reports and expert commentaries within the system.
• To identify patterns in text length, reporting behavior, sentiment, and keywords.

Main Methods:

• Analysis of 6,480 case reports from CIRSmedical.de (as of December 10, 2019).
• Utilized free text fields and expert commentaries (feedback from the CIRS team).
• Employed Python with NLTK and SpaCy libraries to analyze text length, sentiment, and keywords.

Main Results:

• Heterogeneous findings in report length and word count across subject groups; anesthesiology reports are numerous but vary greatly in length, while psychotherapy reports are few and short.
• Professional groups (nurses, doctors, other staff) produce reports of similar length.
• Sentiment analysis revealed commentaries are more negative than reports, likely due to their length; keywords were identifiable but highly varied.

Conclusions:

• NLP enables a systematic description of text properties in event reports and commentaries, offering insights into reporter intention, focus, and mood.
• Sentiment and text length analyses highlight potential issues: short reports may lack sufficient information, while long commentaries risk not being read.
• NLP facilitates rethinking input methods and forms, serving as a foundational step for automated text classification and improved user interaction.