AI in Grading Postoperative Complications: Enhancing Surgical Outcomes

The Clavien-Dindo Classification (CDC) is a widely accepted method for categorizing the severity of postoperative complications. It ranges from Grade I (minor deviations from the expected postoperative course) to Grade V (death). Despite its clinical utility, the consistent application of CDC in real-world settings poses challenges. A recent study explores the potential of Artificial Intelligence (AI) in automating the grading process, promising improved accuracy and efficiency.

Methods
Researchers tested ChatGPT’s ability to define the CDC, create clinical examples, and grade complications from both fictional and real discharge summaries. ChatGPT versions 3.5 and 4 were evaluated. The study included comparison of AI-generated grades against human assessments to measure accuracy and consistency.

Results
ChatGPT 4 demonstrated high accuracy in defining the CDC and generating clinical examples, with 99% agreement. It achieved 97% accuracy in grading single complications, showing substantial improvement over version 3.5. In real-world discharge summaries, ChatGPT 4’s grading showed almost perfect agreement with human assessments (κ value of 0.92).

Discussion
The study underscores ChatGPT 4’s proficiency in automating the grading of postoperative complications. This capability can significantly reduce the workload on healthcare professionals and ensure consistent data quality across institutions. However, some inaccuracies and AI hallucinations highlight the need for ongoing human oversight.

Conclusion
AI, particularly ChatGPT 4, holds great potential in automating the CDC grading process, enhancing surgical outcome assessments, and standardizing data across clinical settings. While promising, the integration of AI into routine clinical practice must be approached with careful validation and continuous monitoring to ensure accuracy and reliability.

DOI: 10.1097/SLA.0000000000006399