EPID-13. DATA-DRIVEN MACHINE LEARNING MODELS AS PROGNOSTIC TOOL FOR PREDICTING SURVIVAL OUTCOMES IN PEDIATRIC PATIENTS WITH GLIOMA OF THE OPTIC APPARATUS

Abstract

Abstract Optic pathway gliomas (OPG) are low-grade lesions arising from the astrocytes of the visual pathway and diencephalon. For most patients, observation appears to be the treatment of choice, but it remains unclear how frequently serial imaging should be obtained and at what point lesions can be deemed to have become quiescent. There is therefore an ongoing need for prognostication tools that can accurately predict the behavior of OPG to help identify those lesions meriting therapeutic intervention. In the present study, we leverage population-level data from the Surveillance, Epidemiology, and End. RESULTS: program (SEER) database to develop a predictive calculator for overall survival (OS) in pediatric OPG. Accordingly, we extracted all available patient-level data from the SEER database. Training and testing datasets were established. Cox proportional hazards regressions were then used to identify independent predictors of survival, and model validation was performed. A total of 1,052 pediatric patients met inclusion criteria. Most were between 5 to 9 years old (45.7%), males (51.6%), white (85.6%), non-Hispanic (82.1%), and 232 patients (22.1%) came from houses with a median income less than $54,999, while 111 patients (10.6%) lived in a rural county. Nearly 90% of patients did not undergo surgical debulking. Twenty-three patients (2.2%) received radiotherapy and 294 (27.9%) received chemotherapy. Additionally, univariate and multivariate analyses were performed to identify variables associated with OS. Median household income and subtotal resection were found to be predictors of OS on univariate analysis. Furthermore, on multivariate analysis, median household income greater than $55,000 and subtotal resection were predictive of higher OS. Age at diagnosis, sex, rural versus urban setting, benign versus malignant behavior, and use of chemotherapy or radiotherapy did not appear to affect OS. Finally, we created a calculator (https://spine.shinyapps.io/optic_glioma/) using our model to estimate 5-,10-, and 1-year OS probability and other survival outcomes.