Long-term survival and second malignant tumor prediction in pediatric, adolescent, and young adult cancer survivors using Random Survival Forests: a SEER analysis

Creating an interactive database for nasopharyngeal carcinoma management: applying machine learning to evaluate metastasis and survival

Objective

Nasopharyngeal carcinoma (NPC) patients frequently present with distant metastasis (DM), which is typically associated with poor prognosis. This study aims to develop and apply machine learning models to predict DM, overall survival (OS), and cancer-specific survival (CSS) in NPC patients to provide optimal tools for improved predictive accuracy and performance.

Methods

We retrieved over 8,000 NPC patient samples with associated clinical information from the Surveillance, Epidemiology, and End Results (SEER) database. Utilizing two methods for handling missing values-imputation or deletion-we created various cohorts: DM-all, DM-slim, OS-all, OS-slim, CSS-all, and CSS-slim. Five machine learning models were deployed for the binary classification task of DM, and their performance was evaluated using the area under the curve (AUC). For the survival prediction tasks of OS and CSS, we constructed 45 combinations using nine survival machine learning algorithms. The Concordance Index (C-index), 5-year AUC, and Brier score assessed model accuracy. Patients were stratified into two risk groups for survival analysis, and the survival curves were presented.

Results

This study examines the relationships between clinical factors and survival in NPC patients. The analysis, visualized through forest plots, indicates that demographic and clinical variables like gender, marital status, tumor grade, and stage significantly affect metastatic risks and survival. Specifically, factors such as advanced stages increase metastasis and survival risks, while enhanced treatments improve survival rates. In the cohort for DM prediction, results revealed that the random forest model was the most effective, with an AUC of 0.687. In contrast, when predicting overall survival (OS), the random survival forest (RSF) model consistently showed superior performance with the highest mean C-index of 0.802, a 5-year AUC of 0.857, and a Brier score of 0.167. Similarly, for cancer-specific survival (CSS) prediction, the RSF model demonstrated a mean C-index of 0.822, a 5-year AUC of 0.884, and a Brier score of 0.165. An online Shiny server was developed to allow the models to be used freely and efficiently via http://npcml.shinyapps.io/NPCpre.

Conclusion

This study successfully established an online tool by machine learning models for NPC metastasis and survival prediction, providing valuable references for clinicians.

Union With Recursive Feature Elimination: A Feature Selection Framework to Improve the Classification Performance of Multicategory Causes of Death in Colorectal Cancer

Despite the use of machine learning tools, it is challenging to properly model cause-specific deaths in colorectal cancer (CRC) patients and choose appropriate treatments. Here, we propose an interesting feature selection framework, namely union with recursive feature elimination (U-RFE), to select the union feature sets that are crucial in CRC progression-specific mortality using The Cancer Genome Atlas (TCGA) dataset. Based on the union feature sets, we compared the performance of 5 classification algorithms, including logistic regression (LR), support vector machines (SVM), random forest (RF), eXtreme gradient boosting (XGBoost), and Stacking, to identify the best model for classifying 4-category deaths. In the first stage of U-RFE, LR, SVM, and RF were used as base estimators to obtain subsets containing the same number of features but not exactly the same specific features. Union analysis of the subsets was then performed to determine the final union feature set, effectively combining the advantages of different algorithms. We found that the U-RFE framework could improve various models' performance. Stacking outperformed LR, SVM, RF, and XGBoost in most scenarios. When the target feature number of the RFE was set to 50 and the union feature set contained 298 deterministic features, the Stacking model achieved F1_weighted, Recall_weighted, Precision_weighted, Accuracy, and Matthews correlation coefficient of 0.851, 0.864, 0.854, 0.864, and 0.717, respectively. The performance of the minority categories was also significantly improved. Therefore, this recursive feature elimination-based approach of feature selection improves performances of classifying CRC deaths using clinical and omics data or those using other data with high feature redundancy and imbalance.