Cumulative barriers to access: An AI-driven analysis of eligibility restrictions, clinical setting, and molecular gatekeeping in 509 interventional glioblastoma trials (2016-2025).

e14065 Background: The 2021 WHO Classification redefined glioblastoma (GBM) via strict molecular parameters. It is unclear how this shift influenced the accessibility of clinical trials and the prevalence of molecular subtypes in study protocols. We utilized a large language model (LLM) to quantify the evolution of eligibility complexity, clinical setting allocation, and biomarker mandates (BM) from 2016-2025. Methods: Glioblastoma trials were extracted from ClinicalTrials.gov (Jan 2016–Dec 2025). An LLM (Gemini 2.5-Flash) parsed unstructured eligibility text from 573 unique protocols (excluding IDH-mutant). Trials were classified as "therapeutic" (disease-modifying intent) or “supportive." Therapeutic protocols were further stratified by clinical setting (newly diagnosed [ND], recurrent restricted, or recurrent open) and BM (requiring biomarkers beyond diagnosis, e.g., MGMT, EGFR). Accuracy was validated via manual review of 30 randomly sampled trials, yielding 93% agreement on clinical setting and 90% on biomarker mandates. Results: 509 (88.7%) of the 573 protocols were classified as therapeutic. Across the decade, protocol complexity remained static (mean exclusion criteria: 13.4, P=0.40). 1) Front-Line Allocation: Clinical settings showed a substantial allocation to the ND population, with 38.3% (n=195) of all therapeutic trials restricted to the front-line setting. 2) Recurrence & Salvage: Among the remaining 314 trials accessible at recurrence, 12.1% (n=38) were restricted to first recurrence only. While 54.2% (n=276) of the therapeutic trials permitted salvage (2nd+ line) entry, 21.4% (n=59) of these enforced additional BM. 3) Molecular Gatekeeping: 24.8% (n=126) of therapeutic trials enforced BM. This proportion fluctuated between 12.8% (2016) and 30.4% (2025) without a statistically significant upward trend (P=0.08). Conclusions: Our AI-driven analysis reveals that while most trials allow salvage entry, access is obstructed by cumulative barriers. Patients face a landscape where 38% of trials are front-loaded to ND disease, and the remaining options are gated by persistent BM and static complexity. This suggests that even trials allowing salvage entry may remain inaccessible to broad patient populations due to these overlapping exclusivity criteria. Comparison of the mean exclusion criteria, biomarker mandates, and clinical setting allocation pre- and post-2021 WHO CNS tumor classification updates. Characteristic 2016-2020 2021-2025 P-Value Total Protocols (N) 235 273 Mean Exclusion Criteria (SD) 14.7 (±7.7) 13.9 (±8.3) 0.25 Biomarker Mandates 53 (22.6%) 72 (26.4%) 0.37 Clinical Setting Allocation 0.53 Newly Diagnosed (Front-line) 87 (37.0%) 107 (39.2%) Recurrent (Restricted/1st Line) 15 (6.4%) 23 (8.4%) Recurrent (Salvage/Open) 133 (56.6%) 143 (52.4%)