Prediction of morality in Retinoblastoma using machine learning based on SEER database

Background: Retinoblastoma (RB) is the most common intraocular malignancy tumor in children. However, few studies have explored the mortality of it and selected the optimal predictive model. In this study, we aim to choose a model to identify the morality of RB. Patients and Methods: A total of 780 patients with RB obtained from SEER database were enrolled in this study. All patients were completely randomized in a 7:3 ratio into a training set (n = 546) and a test set (n = 234). The construction and visualization of all machine learning models were conducted by Python 3.8.0 and R 4.3.0. Various metrics including Area under the receiver operating characteristic curve (AUC), decision curve, receiver operating characteristic curve (ROC), accuracy, sensitivity (recall rate), specificity, and F1 score were utilized to assess and compare the predictive performance of the five models. Results: Among these patients, over half of the children fall within the 1-4 years age group (51.8%). The CatBoost model performed best among the five models with an average AUC of 0.969, followed by the XGBoost model (AUC=0.968). Interestingly, the CatBoost model also achieves the highest score of accuracy, sensitivity (recall rate), specificity, and F1 score were 0.921, 0.882, 0.882, and 0.925 respectively. Conclusion: CatBoost can be utilized as the optimal model for identifying the prognosis of Retinoblastoma (RB). This finding contributes a novel and significant contribution to research within the RB patient population.