Population Dynamics and Short-Horizon Forecastability of the Burst-Suppression Ratio After Cardiac Arrest: Implications for Closed-Loop Brain-State Monitoring

Objective. Closed-loop brain-state control uses the burst-suppression ratio as its manipulated variable, assuming the signal is estimable, non-saturated, and predictable over the control horizon. Whether the injured brain's burst-suppression ratio meets these conditions under routine care is unknown. Approach. We analysed continuous electroencephalography from 607 comatose cardiac-arrest survivors across 5 hospitals in the International Cardiac Arrest Research Consortium database. For each patient we streamed one 24-minute window near 24 hours after return of spontaneous circulation, computed the algorithmic burst-suppression ratio on a 5-second grid, and derived non-saturated occupancy, within-window stability, and short-horizon forecastability. Forecastability was benchmarked against temporal-shuffle surrogates. Main results. The burst-suppression ratio occupied a non-saturated band (0.10 to 0.90) for a median of 0.250 of monitored time (95% CI 0.200 to 0.306) and sat toward suppression (median operating level 0.824). The operating level was stable within the window yet not forecastable beyond its mean: absolute out-of-sample R-squared was near zero at 1 to 5 minute horizons, matched temporal-shuffle surrogates, remained near zero in higher-occupancy windows, and was not improved by a sequence model. An autoregressive forecast showed an apparent skill of 0.46 against a persistence baseline, an artifact exposed by the surrogate analysis. Greater occupancy and a lower operating level were associated with good discharge outcome after adjustment (adjusted odds ratio 2.34 and 0.38; both p <= 0.002). Significance. Passive burst-suppression-ratio history provided little multi-minute predictive information beyond the current operating level, although that level was itself estimable and stable within the window. A closed-loop system with a reliable actuator could still regulate the signal by feedback, but model-predictive control from passive dynamics alone would add little.