A multivariate plasma biomarker model for glioblastoma exclusion: a retrospective case-control study

Background Glioblastoma (GBM) remains the most aggressive primary brain tumour, with limited diagnostic tools able to capture its molecular heterogeneity in real time. Current clinical practice relies on MRI and invasive biopsies, which are poorly suited for early or dynamic disease monitoring. Sensitive and accessible blood-based biomarkers could offer a minimally invasive alternative to aid GBM diagnosis and exclusion. Methods In this retrospective case-control study, plasma concentrations of seven neurological and inflammatory proteins (GFAP, UCH-L1, NfL, Tau, NSE, MPO, and PMN elastase) were quantified in 72 GBM patients and 102 healthy individuals (NAB-X biobank, n = 20; Clinique Saint-Luc Bouge, n = 82, age-matched) using ultrasensitive immunoassay platforms, complemented by conventional clinical laboratory assays and standard ELISA methods. Diagnostic accuracy was evaluated through ROC analysis and logistic regression modelling, with internal validation performed by 10-fold cross-validation and bootstrap resampling. Results GFAP and NfL showed the highest individual discriminative performances (AUC = 0.995 and 0.957, respectively). A multivariate model integrating GFAP, NfL, Tau, NSE, MPO, and PMN elastase showed excellent discrimination between GBM and healthy controls (AUC = 0.999; sensitivity = 98.6%; specificity = 100%). The model’s negative predictive value remained above 0.999 in simulated population settings, supporting its use as a potential high-sensitivity exclusion tool. Conclusions This study reports the possibility of using a plasma-based multivariate biomarker framework to discriminate GBM from healthy individuals with high sensitivity, providing proof-of-concept evidence supporting the future development of a plasma-based exclusion tool. This needs validation in clinically representative symptomatic populations. By integrating astrocytic, neuronal, and inflammatory markers into a single logistic model, this approach bridges analytical precision with clinical applicability, paving the way for a minimally invasive blood test to support diagnostic rule-out in neuro-oncology.