Decoding the Clavien-Dindo Classification: Artificial Intelligence (AI) as a Novel Tool to Grade Postoperative
Sebastian Manuel Staubli1, Harriet Louise Walker2, Fuat Saner3
1Department of HPB Surgery and Liver Transplant, Royal Free Hospital, London, UK.
Annals of Surgery
|June 17, 2024
Summary
ChatGPT 4 demonstrates high accuracy in grading postoperative complications using the Clavien-Dindo classification (CDC). This artificial intelligence (AI) tool shows potential for efficient and precise analysis of clinical data.
Area of Science:
- Artificial Intelligence in Medicine
- Natural Language Processing Applications
- Surgical Complication Assessment
Background:
- The Clavien-Dindo classification (CDC) standardizes surgical complication grading.
- Consistent application of CDC in clinical practice presents challenges.
- Artificial Intelligence (AI) offers automated grading solutions.
Purpose of the Study:
- To evaluate ChatGPT's accuracy in applying the Clavien-Dindo classification (CDC) for postoperative complications.
- To assess AI's capability in grading surgical complications using Natural Language Processing (NLP).
Main Methods:
- Testing ChatGPT's ability to define CDC criteria.
- Evaluating ChatGPT's generation of clinical examples and grading of complications.
- Assessing ChatGPT's interpretation of complications from fictional and real clinical summaries.
Main Results:
- ChatGPT 4 accurately mirrored CDC criteria, outperforming version 3.5.
- ChatGPT 4 achieved 99% agreement in generating clinical examples and 97% accuracy in grading single complications.
- Near-perfect performance was observed when grading complications from real-world discharge summaries (κ=0.92).
Conclusions:
- ChatGPT 4 shows significant proficiency and accuracy in applying the Clavien-Dindo classification.
- AI tools like ChatGPT have the potential to become essential for analyzing Clavien-Dindo classification data.
- Future applications may involve automated extraction and analysis of complication data from clinical datasets.
