Refractory intracranial hypertension is the point at which monitoring stops being merely descriptive and begins forcing high-risk, sometimes irreversible choices. The first principle is that “refractory” is an operational designation, not a distinct disease and not a single ICP value. The Brain Trauma Foundation recommends treating ICP above 22 mm Hg in severe traumatic brain injury, but also explicitly states that ICP must be interpreted with the examination and CT findings. The SIBICC tiers similarly represent increasing treatment risk rather than a rigid ladder through which every patient must pass. A patient with ICP of 24 mm Hg, preserved compliance, intact brain oxygenation, stable pupils, and no tissue shift is not physiologically equivalent to a patient with the same ICP, a rising P2 waveform, falling CPP, worsening PRx, declining PbtO₂, and progressive cisternal effacement. Brain Trauma Foundation guidelinesSIBICC algorithm

That distinction has gained additional support from a 2026 CENTER-TBI high-resolution substudy. Using causal-inference methods in 201 patients, the investigators found that cerebrovascular reactivity and treatment intensity were more directly related to outcome than mean ICP itself. That study does not prove that ICP is harmless; it used observational data, reduced a dynamic process to first-week averages, and remains vulnerable to unmeasured confounding. It does, however, reinforce the larger concept that ICP is embedded within the injury, autoregulatory state, and toxicity of the treatments used to control it. A low ICP achieved through severe hypotension, cerebral ischemia, or destructive surgery is not a physiologic success. 2026 CENTER-TBI substudy

Before labeling an episode refractory, perform a rapid diagnostic reset. Confirm the transducer level, waveform quality, and ventricular catheter patency. Make sure MAP and ICP are referenced at compatible levels. Separate a true sustained rise from suctioning, repositioning, coughing, agitation, ventilator dyssynchrony, fever, shivering, seizure, or an obstructed EVD. Reassess the head and neck position, tight cervical collars, elevated intrathoracic or intra-abdominal pressure, PEEP, hypercapnia, hypoxemia, hyponatremia, and systemic hypotension. Reexamine the pupils and determine whether the pattern has changed from intermittent B waves to a sustained plateau. A new refractory pattern, especially with pupillary change or loss of prior treatment responsiveness, should usually trigger repeat imaging rather than another blind osmotic bolus. Enlarging contusions, recurrent hematoma, evolving infarction, acute hydrocephalus, venous obstruction, and device malposition are not failures of medical ICP therapy; they are diagnoses requiring different treatment.

The next distinction is between an imminent herniation event and persistent intracranial hypertension without current herniation. In impending herniation, minutes matter, and several interventions may appropriately occur in parallel: temporary hyperventilation, concentrated osmotherapy, CSF drainage if immediately available, hemodynamic support, and movement toward definitive surgery. In sustained but nonherniating intracranial hypertension, there is more opportunity to determine which intracranial compartment is responsible and to test one intervention at a time. The response should be measured not only by the change in ICP, but by its speed, magnitude, and durability, together with CPP, PbtO₂, autoregulation, systemic blood pressure, and the examination when available. An intervention that reduces ICP from 30 to 18 mm Hg but drops MAP by 25 mm Hg may have worsened the brain’s actual perfusion.

---

Hyperventilation is the fastest example of a therapy that makes the ICP look better while potentially making the brain worse. Carbon dioxide crosses the blood–brain barrier rapidly. A fall in PaCO₂ raises extracellular pH, constricts cerebral arterioles, reduces cerebral blood flow and cerebral blood volume, and lowers ICP within minutes. It does not remove edema or treat the expanding lesion. It borrows intracranial space from the vascular compartment by deliberately reducing blood flow. That is useful when the alternative is active herniation, but dangerous when marginal tissue is already flow-limited.

The default remains normocapnia, generally a PaCO₂ around 35–40 mm Hg. For acute deterioration or impending herniation, a temporary target around 30–35 mm Hg is reasonable while osmotherapy, imaging, CSF diversion, or surgery is being mobilized. Prolonged prophylactic hyperventilation to PaCO₂ values of 25 mm Hg or lower is not recommended, and hypocapnia is particularly hazardous during the first postinjury day, when cerebral blood flow may already be reduced. If hyperventilation must continue beyond a very brief bridge, PbtO₂ or jugular venous oxygen saturation should be followed when available. The PaCO₂ should subsequently be restored gradually because cerebrospinal fluid bicarbonate adaptation reduces the effect over time and rapid normalization can produce rebound vasodilation and ICP elevation. The Brain Trauma Foundation therefore treats hyperventilation as a temporizing maneuver, not durable ICP therapy. Brain Trauma Foundation ventilation recommendations

---

Escalating osmotherapy is similarly a bridge rather than a destination. A symptom-triggered bolus of hypertonic saline or mannitol can rapidly decrease tissue water and sometimes improve blood rheology, but repeated administration should remain response-driven. Hypertonic saline is often favored when intravascular support is needed, whereas mannitol may be attractive when hypernatremia or volume overload limits additional sodium, provided renal function and blood pressure are adequate. After each dose, determine the ICP reduction, duration of benefit, CPP response, urine output, sodium, chloride, renal function, acid–base consequences, and, with mannitol, the osmolar gap. Diminishing duration of response is not an instruction to shorten the dosing interval indefinitely; it may indicate worsening edema, an evolving mass lesion, exhausted compliance, or loss of an effective osmotic gradient.

Continuous hypertonic saline designed to maintain prophylactic hypernatremia should not be confused with a concentrated rescue bolus. In the 370-patient COBI randomized trial, continuous 20% hypertonic saline did not improve six-month neurologic outcome compared with standard care. This does not negate the acute ICP-lowering efficacy of bolus osmotherapy. It demonstrates again that manipulating a physiologic variable continuously does not necessarily modify the underlying injury or improve recovery. COBI randomized trial

---

Before metabolic suppression, a brief neuromuscular-blockade trial can be diagnostically useful if coughing, shivering, posturing, or ventilator dyssynchrony appears to drive ICP. If ICP does not change after paralysis in an already deeply sedated patient, continuing paralysis adds immobility, obscures motor examination and clinical seizures, and increases the importance of continuous EEG without treating the dominant mechanism. Sedation should similarly be titrated to a demonstrable physiologic purpose. Propofol can control ICP by reducing cerebral metabolism, but high-dose prolonged exposure introduces hypotension and propofol-infusion syndrome without evidence that it improves neurologic outcome.

Barbiturate coma takes metabolic suppression much further. Barbiturates reduce neuronal activity and CMRO₂; flow–metabolism coupling then decreases cerebral blood flow and blood volume, lowering ICP. Once adequate burst suppression is achieved, additional drug provides little further metabolic reduction while cardiovascular toxicity continues to increase. The Brain Trauma Foundation therefore advises against prophylactic barbiturate-induced burst suppression but retains high-dose barbiturates for ICP refractory to maximal medical and surgical treatment, with hemodynamic stability required before and throughout therapy. Brain Trauma Foundation barbiturate recommendations

A traditional adult pentobarbital regimen derived from the original trial literature is a 10-mg/kg load over approximately 30 minutes, followed by 5 mg/kg hourly for three doses and then a maintenance infusion around 1 mg/kg per hour, subsequently titrated to ICP response and continuous EEG. Contemporary institutional protocols vary, so the exact loading and maintenance strategy should follow local pharmacy and neurocritical-care standards. The goal is the lowest dose that controls ICP, commonly accompanied by several EEG bursts per minute, rather than an arbitrarily deep isoelectric tracing. Barbiturate therapy should be continued only when a meaningful ICP response is demonstrated.

The evidentiary limitation is important. The randomized barbiturate literature is small and old. A systematic review found no demonstrated reduction in death or disability and found clinically significant hypotension in approximately one additional patient for every four treated. That hypotension can erase the benefit of lowering ICP by lowering MAP and therefore CPP even more. Cochrane barbiturate review Practical complications also include myocardial depression, vasodilation, prolonged drug accumulation, ileus, impaired nutrition, infection risk, temperature suppression that can conceal fever, electrolyte disturbances, and loss of the clinical examination. Continuous EEG is required, and PbtO₂ is particularly valuable because a falling ICP does not guarantee preserved oxygen delivery. Long drug clearance can also confound later prognostication and death-by-neurologic-criteria evaluation.

---

Induced hypothermia is physiologically compelling for many of the same reasons. Cooling reduces cerebral metabolism, cerebral blood flow, inflammatory activity, and sometimes ICP. Yet the systemic consequences are extensive: shivering may paradoxically raise metabolism and ICP unless suppressed; cardiac output falls; bradycardia, arrhythmia, coagulopathy, infection, insulin resistance, electrolyte shifts, and delayed drug clearance become more likely; and rewarming may precipitate rebound intracranial hypertension. Eurotherm3235 showed that cooling to 32–35°C as an ICP-directed strategy after TBI worsened functional outcome, despite its physiologic rationale. POLAR subsequently found no neurologic benefit from early prophylactic hypothermia. Eurotherm3235POLAR randomized trial

This must be separated from fever prevention. A 2024 ESICM/NACCS consensus strongly supported continuous core-temperature monitoring and controlled normothermia, pragmatically 36.0–37.5°C, as part of early ICP management. The panel did not reach consensus that induced hypothermia should be used as a bridge during impending herniation or before barbiturate coma. Some tiered algorithms retain mild hypothermia around 35–36°C as an extraordinary last-resort maneuver, but this is consensus-based salvage practice, not an intervention proven to improve outcome. If used after all better-supported options are exhausted, shivering control, hemodynamics, coagulation, electrolytes, infection surveillance, brain oxygenation, and very slow rewarming become integral parts of the treatment. 2024 temperature-control consensus

---

Decompressive craniectomy differs fundamentally because it modifies the container rather than the contents. Removing a large section of skull and opening the dura converts a nearly closed cranial compartment into an expandable one. Compliance improves, ICP falls, and compressed vessels may reopen. The operation does not remove diffuse axonal injury, reverse infarcted tissue, restore mitochondrial function, or stop the molecular edema cascade. It allows the injured brain to swell without immediately dying from internal compression. This explains both its lifesaving capacity and its central ethical problem: decompression can change the outcome from death to survival anywhere along the disability spectrum.

The DECRA and RESCUEicp trials are not contradictory once their populations are separated. DECRA studied relatively early, modest intracranial hypertension in diffuse TBI without a mass lesion: ICP above 20 mm Hg for more than 15 minutes within an hour despite first-tier therapy during the first 72 hours. Bifrontotemporoparietal decompression lowered ICP and shortened intensive-care exposure, but produced worse six-month functional outcomes; at 12 months the difference was no longer statistically significant, but there remained no evidence of functional benefit. The lesson is not that decompression cannot work. It is that operating early for a relatively low ICP burden can expose patients to major surgical harm when continued medical management might have succeeded. DECRA trial

RESCUEicp asked a later and more severe rescue question. Patients aged 10–65 years had ICP above 25 mm Hg for one to twelve hours despite first- and second-tier therapies. Bilaterally fixed dilated pupils, an unsurvivable injury, and major bleeding diathesis were exclusions. At six months, mortality was 26.9% with surgical intent versus 48.9% with medical intent, but vegetative state and severe disability were more common among surgical survivors. At 24 months, mortality remained lower at 33.5% versus 54.0%. Expressed in clinically useful terms, for every 100 patients assigned to surgery rather than medical treatment, approximately 21 additional patients were alive at 24 months: about four were vegetative, nine had severe disability, and eight had moderate disability. Good-recovery rates were essentially identical, although improvement by at least one GOS-E category between six and 24 months was more frequent after surgery. RESCUEicp 24-month analysis

The 2020 Brain Trauma Foundation update therefore recommends secondary decompression for “late refractory” ICP patterns resembling RESCUEicp, but not for “early refractory” patterns resembling DECRA. These labels should not be misread as a simple postoperative clock. They encode different ICP thresholds, durations, treatment failures, and patient populations. The recommendation against early secondary decompression also should not be extrapolated to evacuation of an expanding hematoma or to leaving the bone flap off when the brain cannot safely accommodate its replacement. For secondary decompression, the Foundation continues to favor a large frontotemporoparietal opening—at least 12 by 15 cm or approximately 15 cm in diameter—over a small opening. Brain Trauma Foundation decompressive-craniectomy update

Recent evidence adds uncertainty without yet overturning that recommendation. A 2026 CENTER-TBI comparative-effectiveness study examined 295 hemicraniectomies and found wide variation in operative size. Only four patients actually received a decompression meeting the guideline dimensions, and larger size was not associated with better 12-month GOS-E. Because almost no patients occupied the guideline-sized or very-small extremes, and because surgical indication was heterogeneous, the study cannot establish equivalence or justify intentionally small decompressions. It primarily shows that real-world practice differs substantially from the dimensions used in guidelines. 2026 CENTER-TBI craniectomy-size study

Primary decompression during hematoma evacuation is another separate question. In RESCUE-ASDH, 450 patients requiring evacuation of an acute traumatic subdural hematoma were randomized to replacement or nonreplacement of the bone flap when either approach was technically possible. Twelve-month functional outcomes were similar. Craniotomy produced more early reoperations, whereas decompressive craniectomy produced more wound complications and obligated surviving patients to later cranioplasty. The practical inference is that the bone flap should generally be replaced when it can be replaced without compressing the brain; it should be left off when intraoperative swelling makes replacement unsafe, not merely because decompression lowers future ICP. RESCUE-ASDH trial

The complications of decompression follow directly from its physiology. Removing the skull’s restraint may permit external cerebral herniation and venous kinking. Loss of tamponade can allow contusions or contralateral hematomas to expand. Subsequent CSF disturbances include subdural hygromas and hydrocephalus. Before cranioplasty, excessive CSF drainage, upright positioning, or systemic volume depletion can produce sinking-flap physiology and paradoxical herniation. Infection, seizures, wound failure, cranioplasty complications, and prolonged vulnerability of the unprotected brain must be considered part of the original decision rather than deferred as unrelated postoperative issues.

---

Choosing between decompression, barbiturate coma, and continued medical rescue is therefore not simply choosing surgery versus medication. A lateralized expanding lesion, progressive tissue shift, or focal swelling strongly favors an anatomic solution. Diffuse intracranial hypertension without an operable lesion may permit a time-limited barbiturate trial, particularly if hemodynamics and brain oxygenation are robust. A patient already requiring escalating vasopressors may tolerate craniectomy better than profound metabolic suppression, while severe coagulopathy or prohibitive operative instability may temporarily favor medical therapy. Neither option should be selected solely because its immediate effect on ICP is larger.

The goals-of-care conversation must occur before the last possible minute. Families should be told that decompression has strong evidence for reducing death in carefully selected patients with sustained post-traumatic intracranial hypertension, but it does not preferentially create good recovery. It broadens the range of possible survival outcomes, including dependence and vegetative state, and recovery may continue well beyond six months. The 2024 Neurocritical Care Society neuroprognostication guideline recommends against basing prognosis on a single variable or only on ultra-early features during the first three days. Unless there is irreversible brainstem destruction, imminent death, or documented prior wishes limiting support, it suggests at least three days of full critical care and, when possible, one to two weeks before definitive prognostication. That recommendation does not mean waiting while a salvageable patient herniates; it means not withholding timely rescue because of premature nihilism. 2024 TBI neuroprognostication guideline

Every rescue therapy should be framed as a monitored therapeutic trial with an explicit target and a stopping rule. Hyperventilation should lower ICP without producing tissue hypoxia and should end once definitive control is available. Osmotherapy should produce a reproducible benefit without unsustainable hypernatremia, hyperchloremia, hypotension, or renal injury. Barbiturates should continue only if they control ICP while an acceptable CPP and oxygen-delivery state can be maintained. Decompression should be followed by reassessment of PbtO₂, CPP, autoregulation, EEG, and imaging rather than declaring victory because ICP normalized. If maximal burst suppression does not control ICP, if decompression fails to restore viable physiology, or if progressive injury makes the agreed neurologic outcome unattainable, adding further toxic therapy is not escalation—it is failure to recognize nonresponse.

De-escalation is equally physiologic. PaCO₂ should be normalized gradually. Hyperosmolar therapy should stop when there is no active indication rather than being continued to preserve a sodium target. Barbiturates should be tapered only after sustained ICP stability, with awareness of prolonged accumulation and rebound excitability. Induced cooling, if exceptionally used, requires controlled rewarming. Following craniectomy, CSF drainage and positioning must account for the newly open cranial compartment. The question during withdrawal is the same question used during initiation: what physiologic function was this intervention providing, and has the brain recovered enough to assume that function again?

Finally, the TBI evidence cannot be generalized indiscriminately. In malignant hemispheric or cerebellar infarction, decompression is often driven by anatomy and clinical deterioration before a globally measured ICP becomes impressive. In aneurysmal subarachnoid hemorrhage or spontaneous intracerebral hemorrhage, rescue decompression remains more selectively applied and the outcome evidence is substantially weaker. Refractory intracranial hypertension is therefore not one treatment pathway shared by every diagnosis. It is the final common physiologic expression of several diseases, and the definitive intervention must still target the disease producing it.