Using Machine Learning to Revise the AJCC Staging System for Neuroendocrine Tumors of the Pancreas

Abstract

Background: Staging systems are essential for guiding treatment and predicting outcomes in cancer patients. For pancreatic neuroendocrine tumors, the American Joint Committee on Cancer (AJCC) Tumor, Lymph Node, and Metastasis (TNM) system is the current standard. However, its predictive accuracy is limited, as survival curves often overlap, particularly between Stage I and Stage II. Improved methods of patient stratification are therefore needed. Methods: We applied the Ensemble Algorithm for Clustering Cancer Data (EACCD) that involves calculating dissimilarities, ensemble learning, and hierarchical clustering. Data were obtained from the Surveillance, Epidemiology, and End Results (SEER) program of the National Cancer Institute. Models were developed with AJCC TNM variables (T, N, M) and expanded by including patient age. Results: The AJCC TNM system achieved a C-index of 0.6656 (95% CI: 0.6473–0.6839), with survival curves showing poor separation. In contrast, the EACCD model using TNM variables produced four prognostic groups with refined and clear separation, yielding a comparable C-index of 0.6685 (95% CI: 0.6518–0.6852). When age was added, EACCD identified five groups with even stronger stratification and a higher C-index of 0.7015 (95% CI: 0.6852–0.7178). Conclusions: EACCD provides a refined prognostic framework for pancreatic neuroendocrine tumors, outperforming the AJCC TNM system by offering clearer survival stratification, comparable or higher C-index values, and integration of additional clinical factors.