A clinically applicable machine learning model for personalized survival prediction in metastatic pancreatic neuroendocrine tumors

Abstract

Introduction

Metastatic pancreatic neuroendocrine tumors (pNETs) carry a poor prognosis. Currently, no validated model exists to accurately predict survival in this population, highlighting the need for effective prognostic tools.

Materials and methods

In this study, we developed and validated a machine learning-based survival prediction model using data from the Surveillance, Epidemiology, and End Results (SEER) database. The model incorporated ten key prognostic factors, including AJCC TNM stage (T, N, M), tumor grade, primary surgery, non-primary site surgery, chemotherapy, primary site, age, and sex. The eXtreme Gradient Boosting (XGBoost) algorithm was applied to construct the model.

Results

A total of 1430 patients were included in the study. The XGBoost model showed strong predictive performance, with area under the receiver operating characteristic curve (AUROC) values of 0.781, 0.747, and 0.741 for 1-, 3-, and 5-year survival, respectively. The model was implemented in a web-based application that delivers real-time, individualized survival estimates to support clinical decision-making and personalized treatment planning.

Conclusion

By capturing complex interactions among prognostic variables, the model provides accurate survival predictions and supports patient-centered care. It addresses a critical gap in prognostic tools for metastatic pNETs.