Three-Tier Prognostic Stratification of Lung Carcinoids (NET G1-G2-G3) by Multivariable, Data-Driven Integration of Ki-67 and Mitotic Count

Pulmonary neuroendocrine tumors (NETs) include typical and atypical carcinoids, corresponding to low- and intermediate-grade malignancies. Integration of necrosis, mitotic count per 2 mm² (MC), and Ki-67 index, a desirable criterion in the current WHO scheme, may enable a classification framework beyond conventional histology. We used a clustering method for mixed-type data, including necrosis, MC and Ki-67 index, on a single-institution cohort of 358 typical and 125 atypical carcinoids with long-term follow-up. An external validation series of 259 additional carcinoids was used. A three-cluster solution was selected, yielding a tiered prognostic stratification model, that included NETG1 (383 cases; mean MC: 0.7; necrosis: 8.1% of cases; mean Ki-67 index: 2.5%), NETG2 (77 cases; mean MC 2.3; necrosis 27.3%; mean Ki-67: 11.4%), and NETG3 (23 cases; mean MC: 5.5; necrosis 60.9%; mean Ki-67: 25.1%). The model was independent of histological classification. Ki-67 index was the strongest discriminator, distinguishing NET G1 tumors with a Ki-67 index < 6% from NET G3 tumors (Ki-67 > 16%), while MC contributed to identifying a subset of NET G2 tumors. The selected cut-offs were derived through a statistical approach evaluating hundreds of decision-tree configurations, following a sequential algorithm from Ki-67 to MC, with 6–16% identified as the most informative Ki-67 range for intermediate stratification. Necrosis added no discriminative value, as it was closely correlated with the Ki-67 index and MC. This classification corresponded to a progressive worsening of overall survival and, more significantly, relapse-free survival, and it was transferable to the external cohort. In conclusion, based on routinely available pathological parameters, this model provides a practical and reproducible classification of resected pulmonary NETs, overcoming the traditional typical–atypical carcinoid dichotomy and is potentially applicable to biopsy material.