Development and internal validation of an XGBoost-based prognostic prediction model and risk stratification system for overall survival in pediatric neuroblastoma: a SEER-based study.
Using SEER data from 2,427 pediatric neuroblastoma patients, the study developed and internally validated an XGBoost model that predicted 1-, 3-, and 5-year overall survival and separated patients into three prognostic risk groups.
Open original publication →What the AI sees
Using SEER data from 2,427 pediatric neuroblastoma patients, the study developed and internally validated an XGBoost model that predicted 1-, 3-, and 5-year overall survival and separated patients into three prognostic risk groups.
Research significance
The evidence supports prognostic discrimination using routinely available clinical variables; it remains an inference that this model could improve treatment selection, surveillance, or outcomes, because no model-guided intervention, comparison with full contemporary neuroblastoma risk systems, or external validation is reported.
Source abstract
BACKGROUND: Neuroblastoma (NB) is a highly heterogeneous pediatric malignancy with markedly variable clinical outcomes. Although established risk classification systems provide important prognostic information, individualized survival prediction remains challenging. This study aimed to develop and internally validate machine learning-based prognostic models for overall survival (OS) in pediatric NB patients and to establish a risk stratification system based on the best-performing model. METHODS: Data from 2,427 pediatric patients diagnosed with NB between 2000 and 2021 were extracted from the Surveillance, Epidemiology, and End Results (SEER) database. Eligible patients were randomly assigned to a training cohort (80%) and an internal validation cohort (20%). Least absolute shrinkage and selection operator (LASSO) regression identified six prognostic factors, including age, primary tumor site, tumor size, distant metastasis, SEER stage, and chemotherapy status. Six machine learning algorithms, including extreme gradient boosting (XGBoost), logistic regression, support vector machine, random forest, k-nearest neighbors, and decision tree, were developed to predict 1-, 3-, and 5-year OS. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), concordance index (C-index), and Brier score. RESULTS: Among the six models evaluated, XGBoost demonstrated the best predictive performance in both the training and internal validation cohorts. In the validation cohort, the AUCs for predicting 1-, 3-, and 5-year OS were 0.767, 0.796, and 0.807, respectively. The C-index of the XGBoost model was 0.784 [95% confidence interval (CI): 0.746-0.821]. Metastatic status was identified as the most influential predictor. Based on XGBoost-derived risk scores, patients were successfully stratified into low-, intermediate-, and high-risk groups with significantly different survival outcomes (all P<0.05). CONCLUSIONS: An XGBoost-based prognostic model demonstrated favorable performance for individualized survival prediction in pediatric NB and enabled effective risk stratification using routinely available clinical variables. This model may complement existing prognostic assessment approaches; however, further external validation is required before clinical implementation.