Abstract
INTRODUCTION: Glioblastoma (WHO grade 4), defined by IDH-wildtype status and associated molecular features, carries poor prognosis, and real-world survival models incorporating molecular and immunologic variables remain limited. Severe radiation-induced lymphopenia (sRIL) is a proposed prognostic factor, but its independent effect in molecularly defined glioblastoma has not been established in the modern era. METHODS: We retrospectively identified 832 adults with glioblastoma, defined as WHO 2021 grade 4 IDH-wildtype diffuse glioma treated with maximal safe resection and adjuvant radiotherapy (RT) with or without chemotherapy between 2014 and 2024. A trial-eligible subgroup was defined using Stupp criteria. Clinical, molecular, and hematologic variables were analyzed. sRIL was defined as CTCAE grade ≥ 3 lymphopenia within 4 months of RT. Multivariable Cox models incorporated LASSO for variable selection and spline regression for non-linearity. RESULTS: Median overall survival (OS) was 13.0 months. On multivariable analysis, sRIL (HR 1.37, 95% CI 1.16–1.63) remained an independent predictor of worse OS after adjustment for MGMT status, age, resection extent, and treatment factors. MGMT methylation predicted benefit from concurrent TMZ (p(interaction)<0.001), and trial-eligible patients had longer OS across MGMT subgroups. Age and post-RT lymphocyte nadir were non-linearly associated with survival. Proton therapy was associated with a favorable but non-significant OS estimate (HR 0.87, p = 0.11), with no benefit in the trial-eligible cohort. CONCLUSIONS: In this large, real-world cohort, sRIL remained a strong independent prognostic factor. MGMT methylation predicted TMZ benefit, and trial eligibility conferred favorable outcomes. These findings underscore the prognostic relevance of post-treatment lymphopenia and support prospective evaluation of lymphocyte-sparing treatment strategies. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1007/s11060-026-05572-w.