Intra-arterial chemotherapy (IAC) is used to increase the delivery of chemotherapy agents for intracranial tumours, but its clinical value across diseases remains uncertain. We evaluated the clinical outcomes of IAC across intracranial tumours. We systematically searched PubMed, Embase, Scopus and Web of Science until April 2025 (PROSPERO CRD420251056066). Eligible clinical studies evaluated IAC as primary, neoadjuvant, concurrent or salvage therapy and reported quantitative outcomes. For time-to-event outcomes we preferentially extracted and reconstructed hazard ratios (HRs). Random-effects meta-analysis was undertaken where comparable data permitted; small-study effects were explored with funnel plots and Egger's test. Of 1302 records, 155 studies were included in the qualitative synthesis (retinoblastoma, malignant gliomas, head-and-neck squamous-cell carcinoma (HNSCC), and other tumours). IAC achieved high globe-salvage with low systemic toxicity in retinoblastoma. However, in gliomas and skull-based HNSCC, radiographic responses occurred in subsets. A meta-analysis of four studies in glioma showed no significant difference in survival (pooled HR 1.07, 95 % CI 0.95-1.21; prediction interval 0.70-1.65), with evidence of funnel-plot asymmetry (Egger's p = 0.001). Between-study heterogeneity was extreme, and most studies were at serious or critical risk of bias. Toxicities were regimen-dependent, and catheter-related neurovascular events were uncommon. Current evidence supports the role of IAC in ocular retinoblastoma. For gliomas, HNSCC and other skull-base tumors, IAC remains investigational where survival benefit over systemic therapy is unproven. Generally, the pharmacologic advantages of IA delivery alone have not translated into a consistent survival benefit outside retinoblastoma.

Ischemic stroke remains a leading cause of death in the United States, with the COVID-19 pandemic exacerbating disparities. Prior studies have been limited by small sample sizes and lack of generalizability. We used nationally representative data to assess the effects of COVID-19 on ischemic stroke care, outcomes, and predictive modeling. We performed a retrospective cohort study using the 2019-2021 HCUP National Inpatient Sample (NIS). Ischemic stroke hospitalizations and COVID-19 diagnosis were identified using ICD-10-CM diagnosis codes. Hospitalizations were categorized as pre-pandemic calendar year 2019 versus pandemic-era 2020-2021. Calendar-defined pandemic epochs were evaluated as temporal phases and were not interpreted as patient-level SARS-CoV-2 variant assignments. Survey-weighted analyses and machine-learning models were used to evaluate in-hospital mortality. We identified 517,750 weighted ischemic stroke hospitalizations. Compared with the pre-pandemic year 2019, the pandemic era (2020-2021) had higher in-hospital mortality, slightly longer length of stay, higher total hospital charges, and shifted discharge disposition patterns. IV/IA thrombolysis-coded procedure rates were similar between eras, whereas mechanical endovascular reperfusion-coded procedures were more frequent during the pandemic. In the expanded survey-weighted mortality model, COVID-19 diagnosis was associated with higher in-hospital mortality (aOR 2.74, 95% CI 2.42-3.11). Most calendar-defined pandemic epochs were not independently associated with mortality, although the 2021Q4 Delta-to-Omicron transition epoch showed a modest association. In the NIHSS-coded subgroup, COVID-19 diagnosis remained associated with mortality after NIHSS adjustment. Admission-time machine-learning models showed modest discrimination, with AUROC values ranging from 0.62 to 0.64 and AUPRC values ranging from 0.12 to 0.14. COVID-19 diagnosis was strongly associated with in-hospital mortality after ischemic stroke, whereas broad calendar-defined pandemic timing alone showed limited independent association. Calendar epochs should not be interpreted as variant-specific effects because NIS lacks patient-level SARS-CoV-2 sequencing, viral subtype, or variant data. Admission-time machine-learning models provided modest mortality prediction and require cautious interpretation.