Entropy-derived latent states identify pressure–volume regions during experimental intracranial hypertension

Intracranial compensatory reserve determines how intracranial pressure (ICP) responds to added volume, but the operating region of the pressure–volume curve is unavailable during routine monitoring. We tested whether ICP waveform complexity could recover this hidden structure in a secondary analysis of a porcine model of reversible intracranial hypertension. Normalized permutation entropy, wavelet entropy and sample entropy were computed from the 200-Hz ICP waveform in non-overlapping 0.5-s windows. A three-state Gaussian hidden Markov model used only these entropy features; absolute ICP, infused volume, pressure–volume slope and threshold labels were excluded. After inference, numerical states were named Normal, Warning and Alert from the pressure–volume regions they occupied; the names did not alter assignments. State-specific dP/dV, a local inverse surrogate of compliance, increased from 0.53 to 3.84 and 7.96 mmHg mL \(^{-1}\) . Subject-wise leave-one-out mappings defined exclusively in the training animals preserved the complete low–intermediate–high ordering in three of four held-out animals. Thus, entropy-derived states aligned with progressively steeper pressure–volume regions and progressive exhaustion of compensatory reserve during controlled loading. This proof of concept requires independent validation and does not establish a clinical alarm or a phase-independent, haemodynamically independent compliance measurement.