891 metastatic PNETs patients (G1-typical carcinoid, 200; G2-atypical carcinoid, 68; G3-large-cell neuroendocrine carcinoma, 623) diagnosed between 2010 and 2016 were identified. Multivariate analysis was performed using a Cox regression model to identify prognostic factors associated with cancer-specific survival (CSS). The novel M component was established based on the hazard ratio of different metastatic organs. A disease-specific staging system was then proposed by using k-means cluster analysis.
For metastatic PNETs, involvement of bone, liver or brain and multiple metastatic organs were identified as independent prognostic factors in multivariate analysis. M categories was subdivided into three subcategories: M1a, lung involvement only or distant lymph node involvement only; M1b, bone involvement only or liver involvement only; M1c, brain involvement regardless of number of metastatic organs or multiple organs involvement except brain. Primary site surgery, chemotherapy and histologic subtypes were independently associated with CSS, but T component and N component were not. After regrouping histologic subtypes and novel M component, we proposed the following modified staging system: stage IVA (G1M1any, G2M1a-b), stage IVB (G2M1c, G3M1a-b) and stage IVC (G3M1c). The 2-year CSS were 77.9 %, 16.4 % and 5.3 %.
Subdivision of M component according to patterns of distant metastatic organs facilitates prognostic significance for PNETs. Brain metastases and multiple metastatic organs were associated with significantly inferior prognosis. Incorporating histologic subtypes and novel M categories create a disease-specific staging system showed good discriminatory capacity.