Pre-stimulus brain states predict and control variability in stimulation responses

Introduction: Brain stimulation outcomes vary across trials even under identical parameters and targets, limiting reproducibility and clinical translation. Previous work suggests that ongoing brain states, including oscillatory and network-level activity, can modulate stimulation effects, yet the most reliable predictive features and their generalizability across modalities remain unclear. Materials and Methods: We analyzed simultaneous high-density EEG and stereotactic EEG recordings from 36 epilepsy patients across approximately 320 sessions, comprising more than 10,000 single-pulse electrical stimulations. Trial-by-trial relationships between pre-stimulus spontaneous activity and post-stimulus evoked responses were screened across 127 candidate metrics, after which primary analyses were restricted to the 5 most reliable metrics of interest. These metrics captured signal dynamics, synchronization, functional connectivity, and complexity, and were quantified at both local and whole-brain scales Results: Pre-stimulus brain states explained a substantial fraction of post-stimulation variance within session, with higher mean ρ 2 in SEEG than hd-EEG (17.6% vs 7.1%; maxima 85.0% and 71.8%, respectively), and also supported out-of-sample prediction, with positive generalization observed on average in 42.1% of SEEG sessions and 19.0% of hd-EEG sessions. Whole-brain pre-stimulus measures outperformed local measures in both modalities. Explanatory power was also network-dependent and, in SEEG, followed a hierarchy, with the highest explained variance in primary systems and lower values in higher-order networks. Retrospective and prospective closed-loop analyses further converged in showing that conditioning stimulation on favorable pre-stimulus states reduced post-stimulation response variability. Conclusion: This work establishes that trial-to-trial variability in stimulation outcomes is not purely stochastic, but is systematically shaped by ongoing brain dynamics. These findings shift the focus from fixed stimulation settings toward adaptive, state-aware strategies. By identifying pre-stimulus neural signatures that explain and predict stimulation responses, and by showing convergent retrospective and prospective support for state-dependent control, this study provides candidate biomarkers for real-time monitoring and closed-loop neuromodulation. Together, these results support the development of more reproducible and translational stimulation protocols in both research and clinical settings.