Application of machine learning techniques for survival prediction in pediatric malignant non-seminomatous germ cell testicular tumors: a SEER database study

Authors : İncinur Genişol, Batuhan Bakırarar

Pages : 358-367

Doi:10.31362/patd.1528303

View : 57 | Download : 102

Publication Date : 2025-04-01

Article Type : Research Paper

Abstract :Purpose: Childhood testicular cancers constitute 1-2% of all childhood tumors. According to the Surveillance, Epidemiology, and End Results (SEER) database, based on data from 2013 to 2019, the 5-year survival rate is 95.2%. The second most common type of testicular tumor is malignant non-seminomatous germ cell tumor. In recent years, various statistical techniques and extensive databases have been used to obtain information on disease prognosis and survival. In this study, we aimed to develop software using artificial intelligence and machine learning techniques to accurately predict the overall survival of patients with malignant nonseminomatous germ cell testicular tumors. Materials and methods: Our study included data from 788 patients aged 0-18 diagnosed with malignant nonseminomatous germ cell testicular cancer between January 1975 and December 2019. The main hypothesis of the study was to provide overall survival (OS) in years from the date of diagnosis to the date of death or the last follow-up date for surviving patients. In addition to survival analysis, we also analyzed patient age at diagnosis, race, laterality, year of diagnosis, tumor histological type, T stage, N stage, M stage, tumor size, mortality, and follow-up duration. Results: The OS was found to be 41.29±0.43 years. The median survival time was 43.21±0.62 years for patients <15 and 40.34±0.52 years for patients aged ≥15. We developed software that enabled the provision of patient-specific survival in addition to OS for all patients. Conclusion: Recently, artificial intelligence techniques such as machine learning, have shown remarkable advancements compared to other statistical methods. As a result, in this study, we found that the survival rate in pediatric NSCGT was higher if the tumor was diagnosed after 2000, was less than 2 cm in size, and had a T1M0N0 stage yolk sac tumor. We created a 10-year survival prediction model with the results and thought that this model would contribute to the advancement of artificial intelligence studies in prognosis, recurrence and survival analysis.
Keywords : Non-seminomatöz germ hücreli tümör, testis tümörü, SEER, machine learning, sağkalım