Skip to main content

Foundational Trials

RESCUE-ICP

Image: Shutterstock / ChaNaWiT

Publication

  • Title: Trial of Decompressive Craniectomy for Traumatic Intracranial Hypertension
  • Acronym: RESCUEicp
  • Year: 2016
  • Journal published in: New England Journal of Medicine
  • Citation: Hutchinson PJ, Kolias AG, Timofeev IS, et al. Trial of decompressive craniectomy for traumatic intracranial hypertension. N Engl J Med. 2016;375:1119-30.

Context & Rationale

  • Background
    Severe traumatic brain injury (TBI) complicated by sustained intracranial hypertension has a strong mechanistic and observational association with death and poor neurological outcome, and it is a central target of neurocritical care escalation pathways.
    Secondary decompressive craniectomy (DC) is widely used as a “last-tier” intervention when intracranial pressure (ICP) remains uncontrolled despite multimodal medical therapy, but it carries a long-recognised concern that it may convert death into survival with profound disability.
    Prior randomised evidence had focused on earlier DC for more modest ICP thresholds in diffuse TBI and suggested worse functional outcomes despite improved ICP control, leaving equipoise for truly refractory intracranial hypertension and for later “rescue” DC in contemporary tiered protocols.1
  • Research Question/Hypothesis
    In patients aged 10–65 years with TBI and refractory intracranial hypertension (ICP >25 mm Hg sustained for 1–12 hours despite stage 1 and stage 2 therapies), does last-tier secondary DC (vs continued medical management, with rescue surgery permitted) improve the 6-month Extended Glasgow Outcome Scale (GOS-E) distribution?
  • Why This Matters
    The decision to perform DC is a high-stakes, preference-sensitive intervention that trades survival against the risk of vegetative state or severe disability, and it requires robust data for patient/family counselling, trial design, and guideline development across heterogeneous neurocritical care systems.

Design & Methods

  • Research Question: Among patients with TBI and refractory intracranial hypertension despite stage 1–2 therapies, does last-tier secondary decompressive craniectomy improve the 6-month GOS-E distribution compared with continued medical management?
  • Study Type: International, multicentre, parallel-group, superiority, randomised trial; open-label clinical care with central outcome adjudication by assessors unaware of trial-group assignment; neuro-ICU/neurosciences hospital setting (52 centres in 20 countries; 71.1% recruited in the United Kingdom).
  • Population:
    • Key inclusion: Age 10–65 years; TBI with abnormal CT requiring ICP monitoring; refractory intracranial hypertension defined as ICP >25 mm Hg sustained for 1–12 hours despite stage 1 and stage 2 interventions within a tiered protocol.
    • Protocol timing anchor: Randomisation occurred at “stage 3” only (i.e., after failure of first- and second-tier measures), with a protocol-mandated pre-randomisation CT to exclude an evolving intracranial mass lesion.
    • Key exclusions: Pregnancy; severe pre-existing physical/mental disability; injury deemed unsurvivable; other exclusions as per protocol (not all itemised in the index manuscript).
  • Intervention:
    • Last-tier secondary DC with dural opening; operative approach selected by lateralisation of swelling (large unilateral fronto-temporo-parietal craniectomy for unilateral swelling vs large bifrontal/frontotemporoparietal craniectomy for bilateral diffuse swelling, with wide dural opening).
    • Delivered promptly after randomisation: median time from randomisation to DC 2.2 hours (interquartile range 1.3–5.1) in patients who received DC.
    • Ongoing neurocritical care: background care followed the same tiered protocol and local practice for ventilation, sedation/analgesia, cerebrospinal fluid diversion, osmotherapy, vasopressor support, and temperature management.
  • Comparison:
    • Continued medical management at stage 3, including barbiturate infusion as the predominant last-tier therapy (barbiturates used in 171/196 [87.2%]; median time to barbiturates 0.6 hours [interquartile range 0.3–1.2]; median duration 53.0 hours [interquartile range 24.5–115.0] among those treated).
    • Rescue DC permitted at clinician discretion for subsequent deterioration, resulting in substantial crossover to DC in the medical group (73/196 [37.2%]).
  • Blinding: Clinical teams were aware of trial assignment; GOS-E outcomes were adjudicated centrally by investigators unaware of group assignment using mailed questionnaires and structured telephone follow-up.
  • Statistics: A total of 400 patients (200 per group) were required to detect a 15% absolute increase in favourable outcome (from 45% to 60%) with 80% power at the 5% (two-sided) significance level, allowing up to 15% loss to follow-up; analysis was intention-to-treat excluding patients without outcome data and/or with withdrawn/invalid consent; the primary analysis planned an ordinal proportional-odds model for 6-month GOS-E with prespecified descriptive/chi-square reporting if the proportional-odds assumption was rejected.
  • Follow-Up Period: Primary endpoint at 6 months; secondary follow-up at 12 months (reported in the index paper); 24-month outcomes, health-related quality of life, and economic evaluation were protocol-specified but not reported in the index manuscript.

Key Results

This trial was not stopped early. Recruitment ran from 2004 through 2014 and ended after the planned sample size was achieved (408 randomised; 10 excluded from all analyses due to consent issues; 389 patients had primary outcome data at 6 months).

Outcome Decompressive craniectomy Medical management Effect p value / 95% CI Notes
GOS-E distribution at 6 months (primary outcome; n with data) 201 188 Unordered χ² (descriptive) P<0.001 Proportional-odds assumption was rejected; categories reported descriptively.
Death at 6 months (GOS-E “dead”) 54/201 (26.9%) 92/188 (48.9%) Absolute difference −22.1 percentage points 95% CI −31.5 to −12.7 Marked survival advantage with DC.
Vegetative state at 6 months 17/201 (8.5%) 4/188 (2.1%) Absolute difference +6.3 percentage points 95% CI 2.0 to 10.7 Shift towards survival with profound disability.
Upper severe disability at 6 months 31/201 (15.4%) 15/188 (8.0%) Absolute difference +7.4 percentage points 95% CI 1.1 to 13.8 Upper severe disability = independent at home.
GOS-E distribution at 12 months (secondary; n with data) 194 179 Unordered χ² (descriptive) P<0.001 Between-group distributional differences persisted to 12 months.
Death at 12 months (GOS-E “dead”) 59/194 (30.4%) 93/179 (52.0%) Absolute difference −21.5 percentage points 95% CI −31.3 to −11.8 Sustained mortality benefit at 12 months.
Vegetative state at 12 months 12/194 (6.2%) 3/179 (1.7%) Absolute difference +4.5 percentage points 95% CI 0.6 to 8.4 Persistent increase in vegetative state with DC.
Upper severe disability at 12 months 26/194 (13.4%) 7/179 (3.9%) Absolute difference +9.5 percentage points 95% CI 3.9 to 15.1 By 12 months, more DC survivors were independent at home.
Median mean ICP after randomisation (mm Hg) 14.5 (IQR 1.7–18.0) 17.1 (IQR 4.2–21.8) Median difference −3.0 mm Hg 95% CI −4.1 to −1.8; P<0.001 Physiological separation favouring DC.
Median duration of ICP >25 mm Hg after randomisation (hours) 5.0 (IQR 0.0–17.0) 17.0 (IQR 5.0–35.0) Median difference −8.0 hours 95% CI −12.0 to −5.0; P<0.001 Large reduction in ICP burden.
Median cerebral hypoperfusion index 60 after randomisation 6.8 (IQR 3.1–16.6) 11.1 (IQR 4.4–24.8) Median difference −2.8 95% CI −4.9 to −1.0; P=0.002 Lower proportion of time with CPP <60 mm Hg in DC group.
Time to discharge from ICU among survivors (median days) 15.0 (IQR 9.2–22.6) 20.8 (IQR 13.0–32.1) Not reported P=0.01 Among survivors, ICU length of stay was shorter after DC.
≥1 reported complication/adverse event 33/202 (16.3%) 18/196 (9.2%) Not reported P=0.033 Event categories included postoperative haematoma (5 vs 0) and surgical site infection (5 vs 1).
  • Last-tier DC produced a large mortality reduction at both 6 months (26.9% vs 48.9%) and 12 months (30.4% vs 52.0%), with a concomitant increase in vegetative state and severe disability proportions at both timepoints.
  • The primary ordinal model (proportional-odds) was rejected, necessitating descriptive distributional reporting; interpretation therefore rests on the pattern of category shifts rather than a single common odds ratio.
  • Physiological separation was substantial (median hours ICP >25: 5.0 vs 17.0), supporting biological plausibility for the survival effect.

Internal Validity

  • Randomisation and allocation: Computer-generated randomisation with permuted blocks of undisclosed size and stratification by site; central telephone service; allocation released only once the patient met stage 3 criteria (supports allocation concealment).
  • Dropout/exclusions: 408 randomised; 10 excluded from all analyses due to withdrawal of consent or lack of valid consent; 389/398 (97.7%) had primary outcome data at 6 months (201 vs 188), with 7 more patients lost to primary follow-up in the medical group than in the surgical group.
  • Performance/detection bias: Open-label bedside care increases risk of co-intervention and withdrawal-of-care bias; mitigation was central GOS-E adjudication by blinded investigators using questionnaires and structured telephone follow-up.
  • Protocol adherence: Timing of assigned stage 3 therapy was rapid (median time to DC 2.2 hours; median time to barbiturates 0.6 hours among those treated), consistent with urgent rescue intent.
  • Baseline characteristics: Groups were broadly similar (mean age 32.3 vs 34.8 years; males 81.7% vs 80.0%); notable imbalance included history of drug or alcohol abuse (24.8% vs 35.2%; P=0.02) and differences in initial CT classification (P=0.04), although pre-randomisation CT classification did not differ significantly.
  • Heterogeneity: 52 centres in 20 countries with 71.1% recruitment in one national system (United Kingdom); centre-level practice variation is plausible, but randomisation stratified by site reduces confounding by centre.
  • Timing: Randomisation occurred late in the clinical trajectory by design (median time from stage 1 initiation to randomisation 44.3 vs 41.8 hours), selecting a population with sustained refractory ICP and potentially heterogeneous pathophysiology (diffuse swelling, contusions, haematomas).
  • Dose/technique: DC technique was protocolised at a “large” craniectomy with dural opening, but exact size/standardisation was not quantified in the index report; among patients with known type, bifrontal DC occurred in 109/173 (63.0%) and unilateral in 64/173 (37.0%).
  • Separation of the variable of interest: DC was performed in 187/202 (92.6%) in the surgical group vs 73/196 (37.2%) in the medical group; barbiturate infusion was used in 19/202 (9.4%) vs 171/196 (87.2%); median hours ICP >25 after randomisation 5.0 vs 17.0; median mean ICP 14.5 vs 17.1 mm Hg.
  • Crossover: Crossover to rescue DC in the medical group (37.2%) likely diluted the estimated effect of assignment, biasing towards the null for both benefit and harm.
  • Outcome assessment: Primary endpoint (GOS-E) is clinically meaningful, ordinal, and patient-centred; adjudication was blinded, but data collection relied on mailed questionnaires and telephone interviews, introducing potential differential response/availability by survival and disability.
  • Statistical rigour: The primary proportional-odds model was prespecified with a prespecified fallback if assumptions failed; sample size was achieved; however, the analytic “ITT” excluded those without outcome data (no imputation), which can compromise randomisation if missingness is not ignorable.

Conclusion on Internal Validity: Internal validity is moderate-to-strong: randomisation, allocation concealment, and blinded central outcome adjudication support causal inference, but open-label care, substantial crossover, and exclusions/missing outcome data introduce non-trivial risk of bias in effect magnitude and disability distribution.

External Validity

  • Population representativeness: Enrolled patients reflect tertiary neurocritical care practice for severe TBI requiring ICP monitoring, including patients with intracranial haematoma/contusions (approximately 20% had non-diffuse injury on initial CT classification).
  • Key exclusions limit generalisability: Age >65 years was excluded; pregnancy and severe pre-existing disability were excluded; outcomes may differ in older patients, in resource-limited systems, and where rehabilitation capacity is constrained.
  • Comparators and practice patterns: The medical arm used barbiturate infusion commonly (87.2%); centres where barbiturate coma is uncommon or where decompressive surgery thresholds differ may see a different balance of benefit/harm.
  • System-level dependency: Translation depends on timely neurosurgical capability for large DC, intensive monitoring, and sustained neurorehabilitation resources; these contextual factors influence the acceptability of survival with disability.

Conclusion on External Validity: Findings are highly applicable to well-resourced neurotrauma systems managing refractory intracranial hypertension with tiered protocols, but less generalisable to older populations, settings without rapid neurosurgical access, and systems where post-ICU rehabilitation support is limited.

Strengths & Limitations

  • Strengths: Large pragmatic neurosurgical RCT; clinically relevant “last-tier” question aligned with real escalation pathways; international multicentre conduct; central randomisation with site stratification; blinded central adjudication of the primary functional outcome; strong physiological separation in ICP burden.
  • Limitations: Open-label bedside care; substantial crossover to rescue DC in the medical group; modified ITT excluding patients without outcome data and/or with consent issues; long recruitment period (2004–2014) with evolving neurocritical care and rehabilitation practices; heterogeneity in surgical approach and post-cranioplasty timing (not systematically captured in the index report); key patient-centred and economic outcomes were protocol-specified but not reported in the index manuscript.

Interpretation & Why It Matters

  • Survival–disability trade-off
    The central signal is not a uniform “improvement” in neurological outcome but a redistribution across the GOS-E: fewer deaths with DC, offset by more vegetative state and severe disability categories at both 6 and 12 months, and more patients living independently at home (upper severe disability) by 12 months.
  • Time horizon matters
    The disability distribution evolved between 6 and 12 months (e.g., upper severe disability 15.4% at 6 months vs 13.4% at 12 months with DC, alongside increased good recovery from 4.0% to 9.8%), underscoring that 6-month outcome may be an early and potentially pessimistic snapshot for severe TBI survivors.
  • Bedside decision-making
    RESCUEicp provides the quantitative backbone for shared decision-making about last-tier DC: families and teams can discuss expected mortality reduction alongside explicit probabilities of vegetative state, dependence, or independence at home, rather than relying on anecdote or uncontrolled series.

Controversies & Subsequent Evidence

  • Reconciling “early” vs “late” DC trials: The key methodological distinction versus the earlier DECRA trial is timing and threshold (DECRA: early bifrontal DC for more modest ICP elevations in diffuse TBI; RESCUEicp: last-tier DC for sustained ICP >25 mm Hg after failure of stage 1–2 therapies), which plausibly explains divergent functional outcome signals while both trials demonstrate robust ICP reduction with surgery.1
  • Outcome valuation and “acceptable” survival: The trial quantifies the historically contentious proposition that DC can increase survival with severe disability; interpretation is therefore inherently preference-sensitive, and the results force explicit value judgements about outcomes such as vegetative state (8.5% vs 2.1% at 6 months) and dependency (lower severe disability 21.9% vs 14.4% at 6 months).
  • Cranioplasty timing as a potential modifier of functional assessment: Published correspondence queried whether delayed or incomplete cranial reconstruction by 6 months could influence GOS-E scoring and therefore the apparent distribution at the primary endpoint.2
  • Longer-term functional outcomes: A 24-month analysis reported persistent mortality reduction (32.0% with DC vs 47.5% with medical management; absolute reduction 15.5% [95% CI 5.9 to 25.2]) with more survivors in good recovery (26.4% vs 17.7%) but also more vegetative state (4.9% vs 1.9%).3
  • Health-economic interpretation: A published economic evaluation using trial data reported higher costs alongside improved quality-adjusted survival for DC and concluded DC was likely cost-effective at commonly used willingness-to-pay thresholds in the evaluated setting (model-based estimates).4
  • Guidelines and synthesis: A Brain Trauma Foundation update integrating DECRA and RESCUEicp recommends secondary DC for late refractory ICP elevation to improve mortality and favourable outcomes, and recommends against early secondary DC for diffuse injury and <72-hour ICP elevation to improve mortality and favourable outcomes, emphasising that DC decreases ICP and ICU duration but carries functional trade-offs.5
  • Systematic review signal: A Cochrane review synthesising DC trials in closed TBI highlights improved survival with DC while underscoring uncertainty and heterogeneity in functional outcome distribution and the necessity of contextualising “benefit” against survival with disability.6

Summary

  • In severe TBI with sustained refractory intracranial hypertension (ICP >25 mm Hg for 1–12 hours after stage 1–2 therapies), last-tier DC significantly altered the 6-month GOS-E distribution (P<0.001).
  • DC reduced mortality substantially (26.9% vs 48.9% at 6 months; 30.4% vs 52.0% at 12 months) but increased vegetative state and severe disability proportions.
  • Physiological separation was strong: median hours with ICP >25 mm Hg after randomisation were 5.0 vs 17.0, supporting mechanistic plausibility.
  • The medical group had substantial rescue DC crossover (37.2%), which likely diluted the treatment effect of assignment.
  • Later work suggests mortality benefit persists to 24 months with an increasing proportion of good recovery over time, but residual excess vegetative state remains a key harm signal.

Further Reading

Other Trials

Systematic Review & Meta Analysis

Observational Studies

Guidelines

Notes

  • RESCUEicp-related long-term and economic analyses are listed under “Observational Studies” as non-randomised secondary analyses derived from trial data.

Overall Takeaway

RESCUEicp established that, in carefully selected patients with truly refractory post-traumatic intracranial hypertension within tiered management protocols, last-tier secondary decompressive craniectomy substantially reduces mortality and the physiological burden of raised ICP. The price of this survival benefit is a predictable redistribution towards vegetative state and severe disability categories, making the intervention a quintessential preference-sensitive treatment that requires explicit counselling about the probability of survival with dependence versus independence. Its landmark status lies in converting a long-standing neurosurgical “rescue” practice into quantifiable outcomes that can underpin shared decision-making, guideline recommendations, and future trial designs.

Overall Summary

  • Last-tier DC for sustained refractory ICP >25 mm Hg reduced mortality by ~22 percentage points at both 6 and 12 months, with P<0.001 for overall GOS-E distribution shifts.
  • Survival gains were accompanied by higher vegetative state and severe disability proportions, necessitating explicit value-based decisions rather than binary “benefit” framing.
  • Physiological separation was large (median hours ICP >25: 5.0 vs 17.0), supporting mechanistic plausibility for the mortality signal.

Bibliography