A model to predict survival following pancreaticoduodenectomy for malignancy based on tumour site, stage and lymph node ratio

Background Site of tumour origin, lymph node metastases and lymph node ratio (LNR) are identified as important factors determining prognosis in patients undergoing pancreaticoduodenectomy (PD). This study hypothesised that a prognostic index to predict survival could be developed through statistical modelling based on these pathological variables. Methods Patients who underwent PD between 2004 and 2013 were included. Univariable and multivariable (Cox regression) analyses were performed to identify predictors of survival, and a prognostic index was derived. The prognostic index was then validated using an external patient cohort. Results A total of 567 patients who underwent PD were used as a derivation cohort. Tumour site (p < 0.001), tumour size (p = 0.002), T-stage (p < 0.001), vascular involvement (p = 0.002), number of positive nodes (p < 0.001) and LNR (p < 0.001) were significantly associated with survival in univariable analysis. LNR (p < 0.001), tumour site (p < 0.001), T-stage (p = 0.007) remained significant predictors of survival in multivariable analysis, and were combined to derive a prognostic index. The accuracy of the prognostic index was assessed both on the original cohort, and a validation set of 194 patients from another institutional prospective database. The AUROC scores for predicting the overall survival at 3 years were 0.77 in the derivation cohort and 0.74 in the validation cohort. Conclusion The Pancreaticoduodenectomy Prognostic Index is a validated clinico-pathological model based on tumour site, T-stage and LNR to predict long-term survival following PD.