A data-constrained survival stratification model for surgically managed glioblastoma across public and real-world...

Glioblastoma (GBM) has highly heterogeneous postoperative survival. Although contemporary prognostic assessment ideally requires molecular, functional, and surgical-detail variables, these data are often unavailable in population-based and multicenter real-world datasets.

Adults with histologically recorded GBM and tumor-directed surgery (resection or biopsy) in SEER formed the derivation cohort. A multivariable Cox model estimated 6-, 12-, and 18-month overall survival and was externally evaluated in CGGA and an independent institutional cohort without refitting. Because extent of resection, biopsy details, performance status, MGMT promoter methylation, IDH status, and treatment-regimen details were unavailable or non-harmonizable, the model was explicitly intended for registry-based survival stratification when contemporary core prognostic variables are absent.

The study included 3, 842 patients in the SEER derivation cohort, 512 in the CGGA validation cohort, and 96 in the institutional validation cohort. Age, marital status, year of diagnosis, tumor location, tumor size, radiotherapy, and chemotherapy were retained in the final model. The model showed acceptable discrimination and calibration in the derivation cohort and reproducible risk stratification in both validation cohorts. In the independent non-public institutional validation cohort, the 6-, 12-, and 18-month AUCs were 0.748, 0.721, and 0.694, respectively, with a Harrell's C-index of 0.708, supporting reproducible risk stratification in a real-world clinical patient cohort.

This multicohort study developed a registry-compatible survival stratification model for surgically managed glioblastoma for GBM using routinely available variables. The model may support survival estimation under data-constrained conditions, but should not be interpreted as a comprehensive neuro-oncology prognostic system or treatment decision-support tool.