Development and validation of a survival prediction model and risk stratification for pancreatic neuroendocrine neoplasms

摘要

Purpose We explored risk variables associated with cancer-specific survival (CSS) in patients with pancreatic neuroendocrine neoplasms (PNENs) and created a network dynamic nomogram model to predict patient survival time. Methods A total of 7750 patients with PNENs were included in this analysis, including 134 with functional PNENs and 7616 with nonfunctional PNENs. Clinical feature and prognosis differences between functional and nonfunctional PNENs were compared. Independent prognostic factors affecting CSS were analyzed by univariate and multifactorial Cox regression. Nomogram and web-based prognosis prediction of PNENs were developed and validated by C indices, decision curve analysis, and calibration plots. Results Patients with functional PNENs were younger at diagnosis than those with nonfunctional PNENs. Functional PNENs had better prognoses than nonfunctional PNENs (5-year survival rates: 78.55% and 71.10%, respectively). Univariate and multifactorial Cox regression analyses showed that tumor infiltration (T), nodal metastasis (N), metastasis (M), tumor site, differentiation grade, age, marital status, and surgical treatment were independent prognostic risk factors for CSS, which were included in the prognostic nomogram and web-based prognosis calculator. The calibration plots and decision curve analysis showed that the nomogram had excellent prediction and clinical practical ability. The C indices for CSS in the training and validation cohorts were 0.848 (95% CI 0.838-0.8578) and 0.823 (95% CI 0.807-0.839), respectively. We scored all patients according to the nomogram and divided patients into three different risk groups. The prognosis of the low-risk population was significantly better than those of the middle- and high-risk populations based on Kaplan-Meier survival curve. Conclusion We analyzed the clinical features of PNENs and developed a convenient and web dynamic nomogram to predict CSS.