The establishment of machine learning prognostic prediction models for pineal region tumors based on SEER-A multicenter real-world study.
AI interpretation is pending for this paper.
Open original publication →What the AI sees
Not AI summarized yet.
Research significance
Pending deeper interpretation.
Source abstract
BACKGROUND: Pineal region tumors (PRT) are rare intracranial neoplasms with diverse pathological types and growth characteristics, leading to varied clinical manifestations. This study aims to develop machine learning (ML) models for survival prediction, offering valuable insights for medical practice in the management of PRTs. METHODS: Clinical information on PRTs was extracted from the Surveillance, Epidemiology, and End Results (SEER) database. The Kaplan-Meier (K-M) analysis was used to analyze the survival of PRT patients. Univariate and multivariate Cox regression analyses were conducted to identify risk factors for the survival of PRT patients. Then, nomograms were constructed. Seven ML models including Decision Tree, Logistic Regression, LightGBM, Random Forest, XGBoost, K-Nearest Neighbor Algorithm (KNN), and Support Vector Machine (SVM), were developed to predict the prognosis of PRT patients. The predictive value of ML models was evaluated by the area under the receiver's operating characteristic curve (AUC-ROC), tenfold cross verification, calibration curve, and decision curve analysis (DCA). RESULTS: Univariate and multivariate Cox regression revealed that age, histopathology, radiotherapy, and tumor size were independent risk factors for overall survival (OS). Histopathology, surgery, radiotherapy, and tumor size were risk factors for cancer-specific survival (CSS). K-M survival analysis revealed that age, histopathology, marital status, radiotherapy, sex, and surgery significantly impacted OS, while age, histopathology, marital status, race, radiotherapy, sex, and surgery significantly influenced CSS. In the prediction of OS, the ML models with the best clinical utility were RF, Logistic Regression, and XGBoost. For CSS, the most effective models were RF, LightGBM, and RF. CONCLUSION: ML models demonstrate significant potential and high predictive efficacy in forecasting long-term postoperative survival in PRT patients, providing substantial clinical value.