Surgical Traumatic Brain Injury — Evacuation Indications, Craniectomy Trials, ICP Control and Rescue

The head-injured patient gives you one operation to get right and a lifetime of disability to answer for. The three randomised answers that govern the knife are stark: early craniectomy for moderate pressure worsened function without saving lives, late craniectomy for pressure above 25 rescued one in five from death at the price of disability, and for the evacuated subdural the bone flap question changes complications but not function.^[1][2][5] Everything else on this page — which clot to evacuate, when the scan mandates theatre, what tranexamic acid buys inside three hours, how to climb the pressure ladder, and what to tell the family — hangs off those three trials and the cohorts beside them.

The numbers that anchor the viva: DECRA randomised 155 adults with diffuse injury refractory to first-tier therapies to bifrontotemporoparietal craniectomy or standard care — worse functional scores (odds ratio 1.84) and more unfavourable outcomes (odds ratio 2.21) with identical 6-month mortality (19% against 18%).^[1] RESCUEicp randomised 408 patients aged 10 to 65 with pressure above 25 mm Hg refractory for 1 to 12 hours — death at 6 months 26.9% surgical against 48.9% medical, with the survivors shifted toward vegetative and severely disabled states.^[2] RESCUE-ASDH randomised 450 subdural evacuations to flap-off versus flap-on — common odds ratio 0.85, statistically null, deaths 30.2% against 32.2%.^[5] Three traps decide the viva: the 4-hour rule quoted without severity adjustment, the urine-output-style single-number reassurance from one ICP reading, and the bone flap presented as the outcome instead of the swelling beneath it.

A 24-year-old motorcyclist arrives intubated with a blown right pupil, a right convexity subdural with 8 mm of shift, and a frontal contusion. His father takes warfarin for atrial fibrillation — no, the patient takes nothing, but the man in the next bay is 78, fell downstairs on warfarin, and has a thin subdural with a normal exam. The examiner wants your evacuation indications with the millimetre and millilitre numbers, which flap trial governs each clot, your tranexamic acid timing, your pressure ladder with its monitoring evidence, your seizure and clot prophylaxis, and what you tell each family with the 24-month numbers. This page answers each with the number from the paper beside it.^[14][18][2]

Overview & Definition — the operation treats the mass, the ICU defends the brain

Surgical traumatic brain injury is the mass lesion plus the pressure it creates. The primary injury — torn vessels, contused parenchyma, diffuse axonal shearing — is over at impact and no operation reverses it. The secondary injury — expanding haematoma, oedema, raised intracranial pressure, falling perfusion, ischaemia — is what the surgeon evacuates and the intensivist defends. Every decision below separates the two: evacuate what compresses, monitor what swells, and never promise the operation fixes the impact.^[6][2]

The Brain Trauma Foundation's fourth edition exists precisely to mark which practices evidence supports and which it cannot — it synthesises evidence into recommendations only where evidence exists, and it does not constitute a complete protocol for clinical use.^[6] The 2020 decompressive-craniectomy update folded RESCUEicp and the DECRA 12-month data into 3 new level-IIA recommendations with a fourth restated, plus a practice-integration section for the bedside.^[4] Where the guidelines stop, the SIBICC consensus takes over: 42 active severe-TBI specialists from six continents, 18 interventions declared fundamental, ten declared never-use, arranged in three tiers of 10, 4 and 3 interventions.^[7]

Classification — severity bands, lesion types, and the trial entry gates

Classify first by consciousness, because every trial entry gate is a Glasgow Coma Scale band. The scale itself is the 1974 Teasdale-Jennett assessment of coma and impaired consciousness — among the most cited papers in head-injury history with more than 10,000 citations — still the emergency department's standard consciousness overview.^[46] The working bands are 13-15 mild, 9-12 moderate, 3-8 severe, exactly as the extradural outcome cohorts report them: 70% of Hong Kong EDH admissions at 13-15, 10% at 9-12, 20% at 3-8.^[15] When no GCS was ever recorded — absent in 74% of retrospective events — the Mayo system classifies from whatever survives (death, imaging abnormality, GCS, amnesia, loss of consciousness, post-concussive symptoms) with 89% sensitivity and 98% specificity for moderate-severe injury.^[42]

Classify second by lesion, because the lesion picks the trial. Extradural haematoma is the lens-shaped arterial bleed of the young fractured skull — 2% of head injuries — with the best surgical prognosis when caught before coma.^[14][15] Acute subdural haematoma is the crescent over a bruised brain — commoner, deadlier (33% in-hospital mortality in surgical series), and the subject of the flap-on-versus-flap-off randomisation.^[12][5] Traumatic contusion and intracerebral haemorrhage enter surgery through the STITCH gate: no more than two intraparenchymal bleeds of 10 mL or more, within 48 hours of injury, early evacuation within 12 hours of randomisation against initial conservative care.^[16]

Classify third by pressure behaviour, because refractory hypertension is itself a diagnosis with a trial definition: traumatic brain injury with pressure above 25 mm Hg sustained for 1 to 12 hours despite pressure-controlling measures, age 10 to 65 — the RESCUEicp doorway to last-tier surgery.^[2] DECRA's doorway sat earlier and lower — severe diffuse injury with moderate intracranial hypertension refractory only to first-tier therapies — and that single difference in timing explains why the two trials point opposite ways.^[1]

Epidemiology & Risk — falls, warfarin, and the severity behind the clock

The modern surgical head injury is an elderly fall on anticoagulation. In 133 consecutive over-65s needing cranial surgery, mean age 76.6, the fall caused 80.4%, mild injury still comprised 57% of operative cases, comorbidity touched 80%, and preinjury anticoagulation ran at 19% with antiplatelets at 17%.^[45] Mortality was 42.1% — 84% of deaths in the first month — with 73% unfavourable at discharge; severity, absent pupil reactivity, acute bleeding, cisternal obliteration, coagulopathy and transfusion need predicted it.^[45]

Warfarin is the drug that kills. Among 1552 elderly intracranial-haemorrhage patients — 543 aspirin-only, 97 clopidogrel-only, 218 warfarin-only, 193 dual, 501 none — warfarin predicted in-hospital death (odds ratio 2.27) while antiplatelets alone did not, after adjustment including reversal (given to 77% of anticoagulated patients).^[32] The direct oral anticoagulants soften but do not erase the risk: across 49 studies and 15,180 elderly patients, DOACs against vitamin K antagonists cut immediate haemorrhage (odds ratio 0.58) and operations (odds ratio 0.59) with shorter stay — but progression, delayed bleeding and mortality did not differ.^[33]

The baseline that judges every timing claim is severe. Surgical acute subdural series carry 33% in-hospital mortality; trial control arms sit near 40-50% at 6 months.^[12][2] Crude comparisons always flatter delay: early-operated subdurals die at 50% against 15.6% late — because the early group arrives with lower GCS, more axonal injury and more contusion — and after adjustment the 4-hour threshold carries no independent signal (adjusted odds ratio 1.46, confidence interval crossing 1) while each GCS point protects (odds ratio 0.75) and axonal injury devastates (odds ratio 7.56).^[12]

Pathophysiology — pressure, perfusion, and the autoregulation floor

The skull is rigid, so blood plus swelling must steal from brain. Rising intracranial pressure lowers cerebral perfusion pressure — mean arterial pressure minus ICP — and when perfusion fails, ischaemia extends the primary injury. Three facts from the pressure literature organise management: the Foundation fences perfusion at 60-70 mm Hg; the personalised optimum (CPPopt, derived from the pressure-reactivity index) marks each patient's own floor; and dips below that floor predict worse outcome (odds ratios 1.04-1.09 per dose, rising to 1.11-1.26 toward the lower limit of autoregulation) while rises above it do not.^[24] The practical moral is asymmetric: guard the floor, tolerate the ceiling — which is why CPPopt is proposed as the lower limit rather than a midpoint target.^[24][25]

Hyperventilation buys pressure at blood flow's expense. When pressure exceeded 15 mm Hg for 5 minutes across 196 treatment episodes, drainage cut ICP in 90% of observations (by 8.6 mm Hg), mannitol in 90% (by 7.4), hyperventilation in 88% (by 6.3) — but jugular venous saturation, the flow-metabolism index, rose 2.49% with mannitol and only 0.39% with drainage, marking hyperventilation's vasoconstriction mechanism as flow-reducing where mannitol's is not.^[28] That is why hyperventilation is a herniation bridge, never a maintenance therapy — and why it demands oxygenation monitoring when used.^[28][11]

Expansion is the tranexamic acid mechanism and its boundary. Across 1767 CRASH-3 scan-substudy patients, tranexamic acid did not shrink established intraparenchymal bleeding (estimate 1.09) nor raise infarction overall (hazard ratio 1.28, confidence interval crossing 1) — but excluding unreactive pupils, it prevented new haemorrhage (adjusted risk ratio 0.80).^[20] The drug stops the next bleed, not the current clot — which is why pupils, not just the clock, select the patient.^[20][18]

Presentation & the First Look — pupils, motor score, and the deterioration trigger

Examine in the order the herniation kills: consciousness band, pupils, lateralising motor signs, vital-signs Cushing response, then scalp and fracture survey. The single most prognostic bedside pair is the motor score with pupil reactivity — absent reactivity joins severity, acute bleeding, cisternal obliteration and coagulopathy as the independent outcome drivers in surgical series.^[45] Document reactivity precisely, because "at least one reactive pupil" is the tranexamic acid sensitivity boundary and bilaterally fixed pupils bound survival in every cohort on this page.^[18][15]

Read the tempo, not just the number. The lucid interval with a fractured skull and a lens-shaped clot is the extradural — operate on the scan before the coma returns.^[14][15] Immediate coma with a crescent subdural over contused brain is the graver lesion — 33% die despite surgery, and deterioration with mass effect makes it a true emergency regardless of what the 4-hour analyses say about thresholds.^[12] GCS 3 with bilaterally fixed and dilated pupils after extradural bleeding marks severe primary brain injury where emergency evacuation is unlikely to improve outcome — the one setting where the surgeon counsels against the drill.^[15]

The elderly betray the history. More than half of elderly operative cases arrive mild (57%) yet die at 42% — chronic subdural from a forgotten fall presents with gait disturbance and drowsiness, anticoagulated and comorbid, and the thin acute film with a normal exam still earns admission, reversal and a repeat scan.^[45][32] When coma has a non-traumatic explanation — intoxication, post-ictal state, hypotension, hypoglycaemia — the scan plus a lateralising pupil or focal deficit redirects you; without either, resuscitate first and re-examine.^[42]

Scoring & Imaging — Rotterdam predicts death, Marshall predicts theatre, BIG triages the door

Read every trauma head CT with the surgeon's five lines: basal cisterns (open, compressed, absent), midline shift in millimetres, lesion volume and thickness, subarachnoid or intraventricular blood, and skull fracture. The Rotterdam score turns those features into 6-month-mortality prediction: in external validation of 127 severe injuries it predicted 26% against 29% observed with an area under the curve of 0.825 — useful for early death, bounded for function.^[40] The adapted Marshall scheme does the opposite job: across 4895 rehabilitation admissions its categories predicted who needed craniotomy or craniectomy during acute care but added nothing to functional-outcome models — so never counsel rehabilitation from the CT grade alone.^[41]

Apply the extradural numbers exactly, because this is the one lesion with millimetre rules. Non-comatose patients with haematoma under 30 cm in volume, under 15 mm thick, causing under 5 mm of shift may be observed; every haematoma above 30 cm³ evacuates regardless of Glasgow Coma Scale.^[14] Carry the elderly caveat beside the rule: older brains accommodate larger extra-axial volumes, and the thresholds merit renewed consideration rather than blind application.^[14] For the subdural there is no millimetre rule in the randomised literature — indication rests on coma, deterioration, mass effect and equipoise, with the 4-hour benchmark applied alongside severity-stratified prognosis rather than as a standalone criterion.^[12][13]

Triage the mild end with the Brain Injury Guidelines, built by acute-care surgeons from 3803 blunt injuries with 1232 abnormal scans. BIG-1 patients — normal examination, small lesion, no intoxication or anticoagulation — neither worsened clinically nor radiographically nor needed any intervention; every patient needing neurosurgery (13%) sat in BIG-3; assigned-versus-verified agreement reached kappa 0.98.^[8] Use it as a supplement to examination, not a substitute — with prospective validation still warranted before it becomes dogma.^[8]

Monitoring — what the pressure trials actually proved

Hold three monitoring truths together because each tempers the others. BEST-TRIP randomised 324 severe injuries to an ICP-at-or-below-20 protocol versus imaging-and-examination care: composite outcome 56 against 53, mortality 39 against 41% — monitored care not superior, though the monitored arm needed fewer brain-treatment days (3.4 against 4.8).^[9] SYNAPSE-ICU observed 2395 brain injuries across 146 units in 42 countries: monitored patients died less (34 against 49%) with the signal strongest in unreactive pupils (hazard ratio 0.35) — but monitored patients also received markedly more therapy (intensity level 9 against 5), so intensity confounds the survival.^[10] OXY-TC randomised 318 severe injuries to ICP-plus-brain-oxygen versus ICP alone with targets of pressure at or below 20 and oxygen above 20: poor outcome 52 against 51% (odds ratio 1.0), with more catheter dysfunction (8 against 1%) and catheter haematomas (4% against none).^[11] The synthesis for the viva: monitor the severe brain at risk, treat the number as one input among scan and exam, and do not add a second catheter routinely.^[9][10][11]

Drain before you drug-escalate. Ventriculostomy lowers pressure in 90% of episodes and, uniquely among tier-1 measures, does not cut blood flow.^[28] The dose is measured: 3 mL of cerebrospinal fluid drops ICP 10.1% and lifts perfusion 2.2% for at least 10 minutes in severe (GCS at or below 8) patients.^[26] And the clock matters: among 647 drained moderate-to-severe injuries, late (beyond 24 hours) versus early drain placement more than doubled the odds of death or severe disability at 6 months (adjusted odds ratio 2.14).^[27] An EVD placed within a day is both monitor and treatment — the one monitoring intervention with a timing signal favouring haste.^[26][27]

Set targets from the trials, not memory. Intracranial pressure at or below 20 is the randomised protocol target; sustained pressure above 25 for 1 to 12 hours despite therapy is the rescue doorway; perfusion 60-70 is the Foundation fence with the personalised optimum as the floor to defend.^[9][2][24] Autoregulation-guided perfusion targeting is feasible and safe — 60 randomised, target held 46.5% of monitored time against a 36% feasibility bar, no therapy-intensity excess — but not yet outcome-proven.^[25]

Resuscitation — tranexamic acid by the clock, reversal by the drug, osmotherapy without illusions

Give tranexamic acid to the right head injury at the right minute. CRASH-3 randomised 12,737 brain-injured adults — GCS 12 or below or CT-positive, no major extracranial bleeding, 72% within 3 hours — to 1 g over 10 minutes then 1 g over 8 hours or placebo: head-injury death 18.5 against 19.8% (risk ratio 0.94), sharpening to 12.5 against 14.0% (risk ratio 0.89) excluding GCS 3 and bilaterally unreactive pupils, with vascular events identical (risk ratio 0.98).^[18] The benefit concentrates in mild-to-moderate injury (risk ratio 0.78) and vanishes in severe injury (0.99) — and the scan substudy explains why: no shrinkage of established bleeding (estimate 1.09), no excess infarction (hazard ratio 1.28, null), but fewer new haemorrhages once unreactive pupils are excluded (adjusted risk ratio 0.80).^[18][20]

Time it inside 2 hours. Across 28,448 CRASH-2 and CRASH-3 patients the relative reduction peaks within 2 hours and falls fast after; the statistically optimal rule is treatment within 2 hours or with GCS below 9 — with the severe-brain-injury tail possibly persisting beyond 2 hours.^[19] The bay rule follows: give it on clinical criteria before the scan when the clock is running, because nothing about the later course restores the lost minutes.^[18][19]

Reverse the anticoagulated bleeder in parallel with the scan. Hold the agent, correct warfarin by INR, assess platelet function for antiplatelets, and let the head CT plus examination set the pace — rapid protocolised reversal may cut mortality by as much as 38%, though that figure is proposal-level, not trial-proven.^[34] Desmopressin sits in the same honest frame: across 3 cohorts and 5841 antiplatelet-associated haemorrhages it trimmed radiographic expansion in mild injury (adjusted odds ratio about 0.26) with no mortality benefit anywhere and longer stays with platelet strategies — a scan-level gain, not a survival claim.^[35]

Run osmotherapy for pressure episodes, not for prognosis. Six randomised trials and 287 patients find no mortality or functional difference between hypertonic saline and mannitol at very-low-quality evidence; the physiology review finds lower pressure at 90-120 minutes (by 2.33 mm Hg) and higher perfusion (by 5 to 9) with saline but declares the cases insufficient for long-term claims; and 502 CENTER-TBI intensive-care patients show identical mortality (odds ratio 1.0) and 6-month outcome (odds ratio 0.9) whichever agent the centre prefers — with centre habit, not patient features, choosing the drug.^[21][22][23] Prescribe by sodium, volume status and familiarity, capped by the osmotherapy ceiling the barbiturate-coma series priced: 50% mortality at sodium 160 and osmolality 330.^[43]

Definitive Surgery I — the extradural evacuates best

The extradural is the most cost-effective operation in neurosurgery when done promptly — and the most forgiving lesion when caught conscious. Ninety consecutive Hong Kong extradurals: 90% survived, 91% of survivors left with good or moderate function, a third operated, median survivor stay just over 10 days.^[15] The shape of that success is instructive: 70% arrived mild, skull fracture in three-quarters, road crashes dominant — the lucid, fractured, deteriorating young adult is the classic save.^[14][15]

Operate above 30 cm³ whatever the consciousness; observe below 30 cm with under 15 mm thickness and under 5 mm shift only while non-comatose, with serial examination and a low re-scan threshold.^[14] Two cautions ride with the rule. First, the elderly accommodate more — apply the millimetres with age-adjusted judgment, not as statute.^[14] Second, GCS 3 with bilaterally fixed pupils marks primary injury beyond surgery's reach even in this best-prognosis lesion — consent accordingly.^[15]

Definitive Surgery II — the subdural and the flap question

RESCUE-ASDH settled the flap without settling the timing. In 450 randomised evacuations — flap replaced (craniotomy, 228) against flap left out (craniectomy, 222) — function at 12 months did not differ (common odds ratio 0.85, confidence interval 0.6-1.18), deaths matched (30.2 against 32.2%), vegetative states matched (2.3 against 2.8%), and good recoveries ran 25.6 against 19.9%: the price differed instead, with more additional surgery after craniotomy and more wound complications after craniectomy.^[5] The operating rule follows the swelling, not superiority: replace the flap unless the brain forbids it — the randomised price of leaving it out is wound complications, and of replacing it is re-operation.^[5]

Timing follows deterioration, not the dial. The 4-hour benchmark fails severity adjustment — crude early mortality doubles, adjusted signal vanishes, admission consciousness and axonal injury own the outcome — yet no analysis licenses delay in the deteriorating comatose patient with mass effect.^[12] And for the equipoise patient the conservative path is legitimate: across 18 American level-1 centres treating 711 subdurals, centre preference for acute surgery (0 to 86% of cases) bore no relation to 6-month function (odds ratio 1.05), with one in eight conservatively managed patients converting to delayed surgery.^[13] Say it plainly at the bedside: the clot decision is clinical, the flap decision is intraoperative, and the clock decision is severity-stratified.^[12][13][5]

Definitive Surgery III — the contusion and the STITCH signal

Contusions kill by blossoming, which makes early surgery a bet on prevention. STITCH(Trauma) randomised 170 patients with up to two bleeds of at least 10 mL within 48 hours to evacuation within 12 hours versus initial conservative care: favourable outcome 63 against 53% (odds ratio 0.65, short of significance in the under-recruited trial) with an absolute gain of 10.5% — but mortality separated decisively, 15 against 33%.^[16] CENTER-TBI's 367 large-haematoma patients refine where the bet pays: no overall advantage (adjusted odds ratio 1.1), benefit in moderate injury and isolated haematoma (adjusted odds ratios 1.5 and 1.8), and harm signals in mild injury and small (under 33 cc) bleeds where conservative care wins (adjusted odds ratio 0.6).^[17] The contusion rule for the viva: operate the moderate, isolated, sizeable bleed early; watch the mild, small, multifocal one.^[16][17]

Diffuse Swelling & Craniectomy — two trials, opposite lessons, one synthesis

DECRA and RESCUEicp randomised different doors and must never be merged. DECRA took 155 diffuse-injury adults at moderate pressure after first-tier failure to early bifrontotemporoparietal flaps: pressure time, interventions and ICU days all fell — yet functional scores worsened (odds ratio 1.84), unfavourable outcomes doubled (odds ratio 2.21), and deaths matched at 19 against 18%.^[1] RESCUEicp took 408 patients aged 10-65 at pressure above 25 sustained 1-12 hours despite full medical therapy to late rescue flaps: deaths halved at 6 months (26.9 against 48.9%) with survivors redistributed upward through vegetative and severe disability, at more adverse events (16.3 against 9.2%).^[2] Early moderate-pressure craniectomy buys shorter ICU at worse function; late rescue craniectomy buys life at disability's price — timing and threshold, not technique, separate them.^[1][2]

The 24-month follow-up hardens the counsel. Surgical mortality stayed down (33.5 against 54.0%, absolute gap 20.5): per 100 operated, 21 extra alive — 4 vegetative, 9 severely disabled, 8 moderately disabled, zero extra good recoveries — while surgical survivors kept improving (30% gaining at least a grade against 14% medical).^[3] Quote it whole to the family, because the gap between "alive" and "recovered" is the entire consent conversation — and note that improvement continues well past the 6-month horizon most trials stop at.^[3][4]

Refractory Ladder — tiers, coma, cold, and what not to do

Climb SIBICC's three tiers in order: ten tier-1 measures, four tier-2 escalations, three tier-3 last resorts — with 18 interventions declared fundamental, ten declared never-use, MAP-challenge autoregulation guidance, and explicit neuroworsening rules for the declining patient.^[7] The algorithm is consensus, class III by its own declaration — expert opinion scaffolding where randomised evidence thins, not standard of care.^[7]

Place barbiturate coma as salvage with measured numbers. In 55 severe injuries failing osmotherapy and drainage, pentobarbital burst-suppression rescued 40% to discharge, with 68% of followed survivors independent at a year or more — priced at sodium 160 and osmolality 330 as the 50%-mortality ceilings and higher perfusion in survivors (42 against 34 mm Hg).^[43] Coma where theatre cannot reach; craniectomy where it can — with the goals-of-care conversation preceding either at these odds.^[43][3]

Do not cool prophylactically. POLAR randomised 511 severe injuries to 33-35 °C for at least 72 hours against normothermia: favourable outcome 48.8 against 49.1% (relative risk 0.99), with more pneumonia (55 against 51%) and more intracranial bleeding (18 against 15%).^[44] Treat fever; never prescribe cold as neuroprotection.^[44]

Seizure Prophylaxis — seven days, then stop

Phenytoin prevents early seizures and nothing else. Temkin's 404-patient double-blind trial — intravenous load within 24 hours, high-therapeutic levels, one year of treatment — cut week-1 seizures from 14.2 to 3.6% (risk ratio 0.27) with no late effect at one year (21.5 against 15.7%) or two (27.5 against 21.1%).^[29] Cochrane's ten trials and 2326 patients agree: early seizures halved (risk ratio 0.42, very-low quality), late epilepsy untouched (0.91), mortality untouched (1.08).^[30] Seven days of cover, then stop — prophylaxis against the week's seizure, never against epileptogenesis.^[29][30]

Choose the agent by practicality. Levetiracetam matched phenytoin for seizure activity (32 prospective cases against a 41-patient historical phenytoin cohort, p = 0.556) with more EEG seizure-tendency signals (p = 0.003) — no monitoring, no interactions, against the deeper phenytoin evidence base and its fever and metabolism costs.^[31] Either drug for 7 days in severe injury is defensible; neither beyond it is.^[29][31]

VTE & Anticoagulation — early drugs on a stable scan

Start pharmacologic prophylaxis within 72 hours once the 24-hour scan is stable. Across 12 studies and 8747 moderate-to-severe injuries, early beats late for effectiveness (odds ratio 0.46) with no safety difference — and after urgent neurosurgery specifically, initiation within 72 hours cut clots from 13 to 5% (2 against 11% in the craniotomy-craniectomy subgroup) without more re-bleeding or re-operation.^[37][38] The 4951-patient post-neurosurgery cohort prices the trade precisely: each day of delay raises clot odds (adjusted odds ratio 1.08) while lowering re-operation odds (0.72 per day over the first 3 days, 0.85 after) — so start early, watch the scan hardest on days 1-3, and involve neurosurgery in the order.^[36]

Restart therapeutic anticoagulation by indication with a stable scan and a documented multidisciplinary plan — atrial fibrillation, mechanical valve and recent embolism each run their own clock, and none is served by indefinite delay any more than by day-1 heparin.^[36][38] Mechanical measures (stockings, pumps) start at admission for everyone; drugs start when the CT allows.^[37]

Complications & Pitfalls — what the operation and the disease each charge

The disease charges expansion, swelling, seizures, hydrocephalus and medical decay — pneumonia in half, bleeding progression in one in six even without cooling.^[44] The pressure therapies charge their own: osmotherapy capped by sodium and osmolality ceilings, hyperventilation by ischaemia, dual oxygen catheters by dysfunction and haematoma without outcome gain.^[43][28][11] The operation charges re-operation after flap replacement and wound failure after flap removal, with equipoise between them — plus post-traumatic hydrocephalus after discharge.^[5] And the system charges delay: late drains double bad outcomes, late tranexamic acid halves its benefit, late VTE drugs grow clots daily.^[27][19][36]

Prognosis & Disposition — models, scans, and the 24-month truth

Quote IMPACT and CRASH with their case-mix honesty: across 9036 patients in five datasets, discrimination runs 0.83-0.87 in registry populations but 0.65-0.71 in trial populations — the spread is case-mix, not model quality — with no meaningful difference between the two models and good calibration throughout.^[39] More complex models discriminate only slightly better than simpler variants, so the basic models hold their own where data are sparse.^[39] The Rotterdam scan score adds its own validated number: 26% predicted against 29% observed 6-month mortality, area under the curve 0.825.^[40]

Place every survivor deliberately: intensive care through the pressure days, neurosurgical ward for wound and drain care, rehabilitation for cognition, behaviour and epilepsy — with the discharge bundle naming seizure duration, VTE plan, cranioplasty timing and the follow-up scan.^[3][45] Improvement continues to 24 months, disproportionately in the operated — schedule the prognosis conversation in months, not days.^[3]

Special Populations — the elderly faller owns this topic

Four of five elderly operative injuries are falls; half the scans show chronic subdural, over a third acute subdural; comorbidity is near-universal and transfusion needed in three of five.^[45] Warfarin, not aspirin, predicts death — reverse the former aggressively, respect both with admission and repeat imaging.^[32] Direct oral anticoagulants injure less than warfarin at presentation yet kill the same once bleeding — counsel the drug difference without promising a mortality one.^[33] And remember the mild-but-lethal paradox: most elderly operative cases arrive talking and nearly half still die — age, pupils, cisterns and coagulation, not the admission GCS alone, set the ceiling.^[45]

Evidence, Guidelines & Regional Differences — who proved what

The Brain Trauma Foundation owns the evidence map (fourth edition scope plus the 2020 craniectomy update integrating RESCUEicp and DECRA-12-month into new level-IIA guidance); SIBICC owns the bedside algorithm (consensus tiers where trials are silent); BIG owns the mild-end triage for acute-care surgeons.^[6][4][7][8] The three craniectomy randomisations own the knife: DECRA against early flaps for moderate pressure, RESCUEicp for late rescue above 25, RESCUE-ASDH neutral on the flap for evacuated subdurals.^[1][2][5] CRASH-3 owns the bay drug with its 3-hour fence and 2-hour optimum; STITCH plus CENTER-TBI own the contusion conversation; Temkin plus Cochrane own the 7-day seizure rule; the VTE cohorts own the 72-hour start.^{[18][19][16][17][29][30][37]}

Unproven stays unproven: routine brain-oxygen catheters (OXY-TC null with harms), prophylactic hypothermia (POLAR null with harms), desmopressin survival benefit (radiographic only), personalised perfusion targets (COGiTATE feasible, not outcome-proven), and hypertonic-versus-mannitol superiority (physiology yes, outcomes no).^{[11][44][35][25][21]} Omitted as unverifiable in the source abstracts: GCS component maxima, Marshall six-category enumeration, universal blood-pressure targets, and paediatric dosing — this is an adult topic, and omission beats invention.^[46][41]

Exam Pearls — the one-liners that score

• DECRA: early flap, moderate pressure — worse function OR 1.84, unfavourable OR 2.21, deaths 19 vs 18%.^[1]
• RESCUEicp: rescue above 25 — deaths 26.9 vs 48.9% at 6 months, 21 extra alive per 100 at 24 months, none extra recovered.^[2][3]
• RESCUE-ASDH: flap on vs off — common OR 0.85 null, deaths ~31% either way, re-operation vs wound failure.^[5]
• EDH millimetres: observe under 30/15/5 while awake; operate over 30 at any GCS; GCS 3 plus both pupils fixed is unsurvivable.^[14][15]
• ASDH timing: crude early death doubles, adjusted 4-hour signal vanishes (aOR 1.46 ns) — deteriorate with mass effect still means now.^[12]
• Equipoise subdural: centre preference 0-86% changes nothing (OR 1.05) — conservative care is defensible, one in eight converts late.^[13]
• STITCH: early contusion surgery — favourable 63 vs 53%, deaths 15 vs 33%; moderate isolated bleeds win most.^[16][17]
• CRASH-3: 1 g plus 1 g over 8 h within 3 h — death 18.5 vs 19.8%, mild-moderate RR 0.78, severe 0.99, vessels neutral.^[18]
• TXA clock: best within 2 h, rule within-2-h-or-GCS-below-9, severe tail beyond.^[19]
• Pupils select: new-bleed prevention only with reactive pupils (RR 0.80); established clot never shrinks (1.09).^[20]
• Osmotherapy: both agents drop pressure, neither changes outcome — HS lower by 2.33 at 2 h, centre habit decides.^[22][23]
• Drain early: 3 mL drops ICP a tenth; drains after 24 h double bad odds (aOR 2.14).^[26][27]
• Phenytoin: week-1 seizures 3.6 vs 14.2% (RR 0.27), then nothing — 7 days and stop.^[29]
• Warfarin kills (OR 2.27), antiplatelets alone do not; DOACs bleed less (OR 0.58) but kill the same.^[32][33]
• VTE: drugs by 72 h on a stable scan (OR 0.46); post-op delay grows clots daily while early days risk re-operation.^[37][36]
• Prognosis: IMPACT/CRASH AUC 0.83-0.87 registry, models equivalent; Rotterdam 0.825; Marshall predicts theatre, not function.^[39][40][41]
• Rescue ladder: SIBICC 18 fundamental/10 never-use across 10-4-3 tiers; coma rescues 40% at Na-160 limits; never cool prophylactically.^[7][43][44]
• Elderly: falls 80%, dead 42%, unfavourable 73% — severity, pupils, cisterns, coagulation decide.^[45]