Development and Validation of a Deep Learning-Based Survival Prediction Model for Pediatric Glioma Patients: A Retrospective Study Using the SEER Database and Chinese Data

Glioma

Accurately predicting the survival of pediatric glioma patients is crucial for informed clinical decision-making and selecting appropriate treatment strategies. However, there is a lack of prognostic models specifically tailored for pediatric glioma patients. This study aimed to address this gap by developing a time-dependent deep learning model to aid physicians in making more accurate prognostic assessments and treatment decisions.

This retrospective study focuses on survival prediction in pediatric glioma patients using a population-based approach. The model was trained using the Surveillance, Epidemiology, and End Results (SEER) Registry database. To identify specific tumor types, the International Classification of Diseases for Oncology, 3rd Edition codes (ICD-O-3) were used, including codes 9450, 9394, 9421, 9384, 9383, 9424, 9400, 9420, 9410, 9411, 9380, 9382, 9391, 9393, 9390, 9401, 9381, 9451, 9440, 9441, 9442, 9430, and 9380, covering astrocytic tumors, oligodendroglia tumors, oligoastrocytic tumors, ependymal tumors, and other gliomas. Inclusion criteria comprised all primary brain tumors (C71.0-C71.9, C72.3, C72.8, C75.3) diagnosed between 2000 and 2018, among patients under 21 years old, and meeting the third edition of the ICD-O-3 classification. Only patients with available survival time were included, and those with unknown or missing clinical features were excluded. This cohort consisted of 258 pediatric glioma patients diagnosed at Tangdu Hospital in Xi'an, China, between January 2010 and December 2018. These patients had complete clinical data and comprehensive follow-up records.

No related papers found yet.

China