Temporal Evolution of Diagnostic Code-Narrative Discrepancy in Pancreatic Imaging Surveillance: A Longitudinal...

Area of Science:

• Medical Imaging
• Natural Language Processing
• Gastroenterology

Background:

• Radiology reports for pancreatic cystic lesions often include uncertainty expressions (e.g., rule-out, differential diagnosis) alongside structured codes.
• Previous studies noted the prevalence of hedging language, but the temporal relationship between codes and narrative uncertainty during long-term surveillance is underexplored.

Purpose of the Study:

• To analyze the temporal evolution of diagnostic language in pancreatic imaging reports.
• To integrate structured codes with natural language processing (NLP) for analyzing narrative uncertainty.
• To quantify code-narrative discrepancy patterns over extended follow-up periods.

Main Methods:

• Retrospective analysis of 1791 pancreatic imaging reports from 399 patients (March 2020-March 2024).
• Developed a regular expression-based NLP algorithm to extract uncertainty expressions.
• Standardized structured codes into five categories (Cyst, IPMN, Other, Tumor, R/O Malignancy) and analyzed patient trajectories.

Main Results:

• 59.4% of patients had stable diagnostic codes, yet 32.7% showed persistent narrative uncertainty.
• Uncertainty expression rates did not significantly decline over time (early 52.1% vs. late 62.1%).
• 33.9% of patients with stable 'Cyst' codes received IPMN-related hedging language; diagnostic codes changed in 40.6% of patients.

Conclusions:

• Structured diagnostic codes and narrative uncertainty in pancreatic imaging surveillance follow divergent temporal paths.
• Persistent hedging language suggests reporting practices and risk communication, not necessarily evolving diagnostic certainty.
• Improved alignment between structured codes and narrative communication is needed for clearer patient management.