Predictive value of current nodal staging systems and development of machine learning nomogram for resectable
Wen-Bo Zou1,2, Xiu-Ping Zhang1, Yu-Yao Song1
1Faculty of Hepato-Biliary-Pancreatic Surgery, The First Medical Center of Chinese People's Liberation Army (PLA) General Hospital, Beijing, China.
Background:
Given the growing interest in the influence of lymph node metastasis on the prognosis of patients diagnosed with pancreatic head cancer (PHC). This study aims to evaluate the ability of current four nodal staging systems predicting long-term outcomes and develop a machine learning model for predicting the prognosis of patients with resectable PHC.
Materials And Methods:
Participants with PHC were sourced from the Surveillance, Epidemiology, and End Results (SEER) database and allocated at random in a 7:3 ratio to training and internal validation cohort. External validation in a large-sample, multicenter cohort collected from three Chinese institutions was performed to verified the robustness of the optimal nodal staging system and predictive model. The concordance index (C-index), Akaike information criterion (AIC) and area under the curve (AUC) were calculated to evaluate the predictive capability and discrimination of different nodal staging systems. The machine learning procedures based procedure and Cox regression analysis were implemented for identification of the prognostic factors and construction of predictive model. The calibration curves, net reclassification improvement (NRI) and integrated discrimination improvement (IDI) and decision curve analysis (DCA) were using to assess predictive accuracy and clinical benefits of the predictive model.
Results:
All four nodal staging systems were independent prognostic factors for overall survival (OS). The log odds of lymph node ratio (LODDS) were verified as the optimal nodal staging system with highest C-index and AUCs, and lowest AICs compared to others, and has better predictive capability than others both in patients with < 12 and ≥ 12 retrieval lymph nodes (RLNs). Then, a predictive model including T stage, tumor differentiation, chemotherapy, and LODDS was developed and validated. This model had a higher C-index and AUCs than the AJCC staging system. The NRI, IDI, and DCA analysis also indicated that present model had good predictive capability and clinical utility.
Conclusion:
The nodal staging system LODDS is the optimal prognostic factor for OS in resectable PHC. It could effectively predict OS for resectable PHC patients without considering the numbers of RLN. The machine learning model could effectively predict OS for patients with resectable PHC.
