Machine learning for predicting treatment plans in patients with small non-functioning pancreatic neuroendocrine tumors

Background: This study aimed to investigate the impact of surgical or conservative treatment on the prognosis of small non-functioning pancreatic neuroendocrine tumors (NF-PNETs) and to predict appropriate treatment strategies by constructing machine learning (ML) models. Methods Using clinical data from patients with small NF-PNETs in the Surveillance, Epidemiology, and End Results (SEER) database from 2000 to 2020, the Kaplan-Meier survival curve(KM) was used to assess the impact of different treatment approaches on survival. Cox analysis was performed to validate the prognostic value of treatment approaches. Six ML models (Support Vector Machine, Logistic Regression, Random Forest, AdaBoost, K-Nearest Neighbor, and Multilayer Perceptron) were applied to predict the most appropriate treatment approach for conservative patients. Performance evaluation was performed using metrics such as area under the curve (AUC), F-score, predictive accuracy, sensitivity, and specificity to compare the performance of the models. Results A total of 2602 patients with small NF-PNETs were included in this study, of whom 94.2% received surgical treatment. Patients who underwent surgery had more prolonged overall survival, and treatment modality was identified as an essential factor associated with improved survival (HR: 0.092, 95% CI: 0.053–0.158). In a Cox multivariate analysis, surgical treatment was significantly associated with a reduced risk of death (HR: 0.188, 95% CI: 0.126–0.279). Evaluation of the random forest algorithm showed that some patients with conservative treatment require a more aggressive treatment plan and have a relatively good discriminatory ability (AUC, 0.915, 95%CI: 0.872–0.959). Conclusion Surgical treatment may be a promising strategy for prolonging survival in patients with small NF-PNETs. Further individualized decision-making is needed to ensure patients have the best possible outcome.