Machine learning based predictive model and genetic mutation landscape for high-grade colorectal neuroendocrine carcinoma: a SEER database analysis with external validation.

Background: High-grade colorectal neuroendocrine carcinoma (HCNEC) is a rare but aggressive subset of neuroendocrine tumors. This study was designed to construct a risk model based on comprehensive clinical and mutational genomics data to facilitate clinical decision making. Methods: A retrospective analysis was conducted using data from the Surveillance, Epidemiology, and End Results (SEER) database, spanning 2000 to 2019. The external validation cohort was sourced from two tertiary hospitals in Southwest China. Independent factors influencing both overall survival (OS) and cancer-specific survival (CSS) were identified using LASSO, Random Forest, and XGBoost regression techniques. Molecular data with the most common mutations in CNEC were extracted from the Catalogue of Somatic Mutations in Cancer (COSMIC) database. Results: In this prognostic analysis, the data from 714 participants with HCNEC were evaluated. The median OS for the cohort was 10 months, whereas CSS was 11 months. Six variables (M stage, LODDS, Nodes positive, Surgery, Radiotherapy, and Chemotherapy) were screened as key prognostic indicators. The machine learning model showed reliable performance across multiple evaluation dimensions. The most common mutations of CNEC identified in the COSMIC database were TP53, KRAS, and APC. Conclusions: In this study, a refined machine learning predictive model was developed to assess the prognosis of HCNEC accurately and we briefly analyzed its genomic features, which might offer a valuable tool to address existing clinical challenges.