Toward an Optimized Staging System for Pancreatic Ductal Adenocarcinoma: A Clinically Interpretable, Artificial...

Area of Science:

• Oncology
• Surgical Oncology
• Cancer Staging

Background:

• The American Joint Committee on Cancer (AJCC) staging system for pancreatic ductal adenocarcinoma currently uses positive lymph nodes (PLNs) to define nodal status.
• This approach treats tumor size (T stage) and nodal status (N stage) as independent factors.
• Emerging data suggest that lymph node ratio (LNR) may be a more accurate predictor of outcomes.

Purpose of the Study:

• To compare the predictive accuracy of LNR versus PLN for pancreatic ductal adenocarcinoma.
• To investigate the interactions between tumor size and nodal status in predicting patient outcomes.
• To develop and evaluate artificial intelligence-based models for improved cancer staging.

Main Methods:

• Utilized Shapley Additive Explanations (SHAP) analysis to compare LNR and PLN in a cohort of patients who underwent pancreatic ductal adenocarcinoma resection.
• Trained optimal classification trees (OCTs) using tumor size and nodal status (defined by the best predictor) to predict 1-year and 3-year mortality.
• Compared OCT performance against the AJCC schema and XGBoost models, exploring variable interactions via SHAP.

Main Results:

• SHAP analysis identified LNR as a superior predictor of outcomes compared to PLN.
• OCT models demonstrated improved predictive performance over the AJCC schema in both derivation and validation cohorts.
• Significant interactions were observed between LNR and tumor size, indicating complex prognostic relationships.

Conclusions:

• Lymph node ratio (LNR) is a more effective prognostic indicator than positive lymph nodes (PLNs) in pancreatic ductal adenocarcinoma.
• Interactions between tumor size and nodal status are crucial for accurate staging.
• Optimal classification trees offer a powerful, interpretable method for integrating these factors into future staging systems.