Heterogeneity of immune checkpoint inhibitor-related inflammatory central nervous system adverse event reporting signals in primary and metastatic brain tumors: a pharmacovigilance study with single-cell and spatial transcriptomic contextualization.

Background: Immune checkpoint inhibitors (ICIs) can induce immune-related adverse events (irAEs) across multiple organ systems. Although inflammatory central nervous system irAEs (CNS inflammatory irAEs) are uncommon, they are often severe. Primary CNS tumors and brain metastases have distinct immune microenvironments, yet the heterogeneity of ICI-related inflammatory CNS irAE reporting signals across tumor phenotypes remains poorly understood. Methods: We used pharmacovigilance signal discovery, external corroboration, and transcriptomic contextualization of pre-existing brain tumor immune landscapes. We constructed an ICI-exposed cohort from the FDA Adverse Event Reporting System (FAERS) and compared inflammatory CNS irAE disproportionality signals across primary CNS tumors, brain metastases, and non-CNS solid tumors. External comparison used the Japanese Adverse Drug Event Report database (JADER). Public single-cell RNA sequencing datasets were analyzed to characterize baseline strict inflammatory and broad stress-related modules across cellular compartments, with spatial transcriptomics used as secondary descriptive visualization in brain metastasis tissue. Results: In FAERS, inflammatory CNS irAE reporting signals suggested tumor phenotype-associated heterogeneity, with adjusted odds ratios of 1.65 (95% CI, 1.02-2.65) for primary CNS tumors and 3.12 (95% CI, 2.45-3.98) for brain metastases versus non-CNS solid tumors. Signals were stronger under a strict noninfectious phenotype and attenuated under a broad neuroinflammatory phenotype. Thyroid comparator analyses showed no comparable enrichment, whereas the myocarditis-related comparator was too sparse in the brain metastasis subgroup for meaningful inference. JADER showed a broadly similar pattern, although primary CNS tumor estimates were sparse and exploratory. Baseline single-cell analyses localized strict inflammatory module activity mainly to myeloid and T/NK compartments, while spatial maps served only as secondary descriptive visualization. Conclusions: ICI-related inflammatory CNS irAE reporting signals suggested tumor phenotype-associated differences, most prominently in brain metastases. Stricter phenotype definitions appeared more specific than broader neuroinflammatory definitions. Public single-cell datasets characterized pre-existing immune-rich myeloid/T-NK compartments, while spatial maps provided only secondary descriptive tissue-level visualization and did not demonstrate irAE-onset tissue states. These pharmacovigilance findings should be interpreted as hypothesis-generating reporting associations, not evidence of incidence, absolute risk, or causality.