e13746 Background: Up to 50% of patients with metastatic non-small cell lung cancer (NSCLC) will develop brain metastases in their disease course, with 10% to 20% having them at the time of diagnosis. While modern immunotherapy (IO) agents demonstrate meaningful intracranial efficacy, clinical trial (CT) eligibility criteria often mandate central nervous system (CNS) stability or exclude CNS disease entirely. This study utilized a large language model (LLM) to audit whether eligibility criteria in the modern IO era have evolved to include patients with active or untreated CNS metastases in the advanced/metastatic setting. Methods: Interventional phase I to phase III CTs involving "immunotherapy" in "NSCLC stage IV" were extracted from ClinicalTrials.gov from 2016 to 2025. Using each trial’s inclusion and exclusion criteria, a LLM (Gemini 2.5-Flash) was used to classify CNS metastases eligibility into four categories: 1) total exclusion (explicitly excludes history of CNS metastases), 2) treated/stable only (only allowed if CNS metastases have been treated and are stable), 3) active allowed (explicitly allow active, asymptomatic, or untreated CNS metastases), and 4) silent (CNS metastases policy not mentioned). The accuracy of LLM categorization was validated via manual review of 30 randomly sampled trials, yielding 100% agreement. Results: In this CT cohort (n = 71), only 16.9% of trials explicitly permitted patients with active/untreated CNS metastases. 14.1% of trials did not have a statement regarding CNS metastases and were categorized as silent. The majority (63.4%) of CTs required treated/stable brain metastases, with 8.5% explicitly excluding those with brain metastases. Temporal analysis revealed a non-significant 2.3% decrease in the number of clinical trials allowing for active/untreated metastases (p = 0.21). Conclusions: Even when analysis is restricted to advanced/metastatic NSCLC trials, where brain metastases are a common clinical reality, 69% of modern IO protocols will explicitly exclude patients with active CNS disease. This "CNS exile" creates a profound evidence-practice gap, generating pivotal trial data that is not representative of a significant portion of the patient population. Future protocols must explicitly expand inclusion for stable, untreated CNS metastases to improve equity and real-world applicability. CNS eligibility policy in metastatic NSCLC trials by year, 2016-2025 (%). Year ACTIVE ALLOWED SILENT TOTAL EXCLUSION TREATED STABLE Total (N) 2016 33.3 0 66.6 0 3 2017 11.1 11.1 0 77.8 9 2018 42.9 0 14.3 42.9 7 2019 25 16.7 0 58.3 12 2020 0 0 14.3 85.7 7 2021 12.5 12.5 0 75 8 2022 16.7 33.3 0 50 6 2023 0 25 0 75 4 2024 33.3 0 0 66.7 6 2025 0 33.3 0 66.7 9 Total 16.9 14.1 5.6 63.4 71
Ng et al. (Thu,) studied this question.