Metastatic patterns, prognostic factors, and deep learning model development in primary gastrointestinal melanoma: a retrospective cohort analysis

Background: Existing studies provide limited knowledge of the metastatic pattern, survival rate, and prognosis of primary gastrointestinal melanoma (PGM). This study aimed to investigate the metastatic patterns, prognostic factors, and conduct deep learning model of PGM. Methods: The Surveillance, Epidemiology, and End Results (SEER) database was analysed to determine survival time, survival rates, and metastatic patterns in PGM. Cox regression analysis identified prognostic factors associated with overall survival (OS) and cancer-specific survival (CSS). Patients were divided into discovery (80%) and validation cohorts (20%) to develop and validate deep learning-based models for predicting OS and CSS of PGMs. The area under the receiver operating characteristic curve (AUC) was used to evaluate model performance. Results: The median OS was 18 months [95% confidence interval (CI): 15-21] and 22 months (95% CI: 19-26) at CSS. OS rates were 60% (95% CI: 56-64%), 32% (95% CI: 28-36%), and 22% (95% CI: 18-26%) at 1, 3, and 5 years. The most common metastasis sites were the liver (19%), lungs (16%), bones (5%), and brain (4%). Older age, involvement of other sites, regional or distant stage disease, and two distant metastases were associated with worse OS or CSS, whereas systemic therapy was a protective factor. The deep learning models demonstrated performance in predicting OS (AUC: 0.7757-0.8366 at 1 year and 0.8046-0.8177 at 3 years) and CSS (0.7870-0.8169 AUC at 1 year and 0.7314-0.7720 at 3 years). Conclusions: The prognosis of PGM varies significantly among subtypes, and the models developed in this study provide accurate predictions of OS and CSS, offering potentials for clinical utility.