A spatiotemporal state-inference framework for adaptive immunotherapy in glioblastoma.

Although immunotherapy has transformed outcomes in several solid tumors, it has yielded little survival benefit in glioblastoma (GBM). This limited efficacy may in part reflect not only modest drug activity and a chronically immunosuppressive microenvironment, but also a temporal mismatch between fixed treatment schedules and a tumor-immune ecosystem that evolves across space and time. This review outlines the GBM Immune-Spatiotemporal Feedback Loop (GBM-ISFL), a clinician-governed, hypothesis-generating framework that conceptualizes adaptive immunotherapy as a process of longitudinal sensing, biologic state inference, phase-matched intervention, and iterative feedback. Drawing on single-cell and spatial multi-omics, radiologic assessment, and liquid-biopsy studies, we outline a four-phase atlas of GBM evolution and define a patient-specific Critical Transition Window. This window may functionally overlap with the post-radiotherapy interval highlighted by Response Assessment in Neuro-Oncology (RANO) 2.0, but it should not be treated as a fixed calendar block or as a validated clinical interval. To narrow the resulting observability gap, we position spatiotemporal graph neural networks (STGNNs) as candidate tools for noninvasive inference of latent tumor-immune states from serial multimodal data, including imaging dynamics, treatment exposure, and circulating biomarker trajectories. We further describe how uncertainty-aware state inference could support exploratory phase-specific therapeutic reasoning, translational validation, and lifecycle governance. By reframing GBM immunotherapy around biologic phase rather than chronology alone, the GBM-ISFL offers a testable route toward adaptive, state-informed, and clinically governed precision intervention, but it should not be interpreted as a current standard-of-care algorithm.