Skip to main content

Foundational Trials

CRASH-3

Image: Shutterstock / ChaNaWiT

Publication

  • Title: Effects of tranexamic acid on death, disability, vascular occlusive events and other morbidities in patients with acute traumatic brain injury (CRASH-3): a randomised, placebo-controlled trial
  • Acronym: CRASH-3
  • Year: 2019
  • Journal published in: The Lancet
  • Citation: The CRASH-3 trial collaborators. Effects of tranexamic acid on death, disability, vascular occlusive events and other morbidities in patients with acute traumatic brain injury (CRASH-3): a randomised, placebo-controlled trial. Lancet. 2019;394:1713–23.

Context & Rationale

  • Background
    • Progressive intracranial haemorrhage after TBI is common and is strongly associated with raised intracranial pressure, herniation, death, and disability.
    • Bleeding and haemorrhage expansion occur predominantly in the first hours after injury, providing a biologically plausible “time-critical” window for haemostatic intervention.
    • Tranexamic acid (TXA) is an antifibrinolytic with established efficacy in reducing surgical bleeding and reducing bleeding deaths in trauma with major extracranial haemorrhage when given early.
    • Before CRASH-3, only two small randomised TBI trials existed, and meta-analysis suggested a possible mortality reduction (RR 0·63; 95% CI 0·40–0·99) but provided limited information on disability and adverse events.
  • Research Question/Hypothesis
    • Does early intravenous tranexamic acid, compared with placebo, reduce head injury-related death (and improve disability outcomes) in adults with acute traumatic brain injury, without increasing vascular occlusive events or other harms?
    • Is any effect modified by time to treatment and/or baseline severity (e.g., Glasgow Coma Scale and pupil reactivity)?
  • Why This Matters
    • TBI is a high-burden global condition with limited pharmacological therapies that plausibly improve early outcomes.
    • TXA is inexpensive, widely available, and deliverable in emergency and prehospital settings, but requires high-quality evidence specific to TBI.
    • Even modest relative benefits could translate into large absolute population benefit given the incidence of TBI worldwide.

Design & Methods

  • Research Question: In adults with acute TBI, does IV tranexamic acid given as early as possible (and within 3 hours) reduce head injury-related death (and disability), compared with placebo, without excess serious adverse events?
  • Study Type: International, multi-centre, randomised, placebo-controlled, double-blind pragmatic trial; 175 hospitals in 29 countries; predominantly emergency department / acute hospital setting.
  • Population:
    • Adults with traumatic brain injury who were within 3 h of injury at randomisation (originally within 8 h; protocol amended on 6 Sept 2016 to restrict recruitment to within 3 h).
    • Clinical eligibility threshold: Glasgow Coma Scale (GCS) score ≤12 or any intracranial bleeding on CT scan.
    • Required absence of major extracranial bleeding.
    • Core pragmatic inclusion principle: the responsible clinician was “substantially uncertain” about whether TXA was appropriate.
    • Key exclusions included: significant extracranial bleeding; receipt of TXA before randomisation; known contraindication to TXA; and situations where the clinician was not substantially uncertain about giving TXA.
  • Intervention:
    • Tranexamic acid 1 g IV loading dose over 10 minutes, followed by 1 g IV infusion over 8 hours.
  • Comparison:
    • Matching placebo, delivered with the same loading and infusion regimen (1 g-equivalent volume over 10 minutes, then infusion over 8 hours).
  • Blinding: Patients, treating clinicians/caregivers, and outcome assessors were masked; treatment packs were identical except for pack number; allocation was by selecting a numbered pack from a box containing eight packs.
  • Statistics: Sample size planning: 10,000 patients were estimated to provide 90% power (two-sided alpha 1%) to detect a 15% relative reduction in mortality (from 20% to 17%); after restricting eligibility/primary analysis to treatment within 3 h, the target was increased to 13,000 to ensure ~10,000 treated within 3 h; analysis was intention-to-treat with a prespecified sensitivity analysis excluding GCS 3 and bilateral unreactive pupils.
  • Follow-Up Period: Outcomes assessed in hospital up to 28 days after injury (or earlier discharge); disability outcomes recorded at discharge or day 28 (whichever occurred first).

Key Results

This trial was not stopped early. Recruitment ran from 20 July 2012 to 31 Jan 2019; interim analyses were overseen by an independent data monitoring committee, and enrolment concluded when the planned sample size and trial treatment supply were reached.

Outcome Tranexamic acid Placebo Effect p value / 95% CI Notes
Primary: Head injury-related death in hospital within 28 days (treated within 3 h; ITT) 855/4613 (18·5%) 892/4514 (19·8%) RR 0·94 95% CI 0·86 to 1·02; p value not reported Primary analysis population: those treated within 3 hours
Prespecified sensitivity: Head injury-related death (treated within 3 h; excluding GCS 3 and bilateral unreactive pupils) 485/3880 (12·5%) 525/3757 (14·0%) RR 0·89 95% CI 0·80 to 1·00; p value not reported Designed to reduce dilution by including patients unlikely to benefit
Subgroup: Mild-to-moderate injury (GCS 9–15; treated within 3 h) 166/2846 (5·8%) 207/2769 (7·5%) RR 0·78 95% CI 0·64 to 0·95; heterogeneity p=0·030 (vs severe) Prespecified severity subgroup
Subgroup: Severe injury (GCS 3–8; treated within 3 h) 689/1739 (39·6%) 685/1710 (40·1%) RR 0·99 95% CI 0·91 to 1·07; p value not reported No evidence of benefit in this severity stratum
Subgroup: Both pupils reactive (treated within 3 h) 440/3820 (11·5%) 493/3728 (13·2%) RR 0·87 95% CI 0·77 to 0·98; heterogeneity p=0·032 (vs any non-reactive) Suggests better effect in patients with preserved brainstem reflexes
Subgroup: Any non-reactive pupil (treated within 3 h) 415/793 (52·3%) 399/786 (50·8%) RR 1·03 95% CI 0·94 to 1·13; p value not reported Very high baseline risk; biologically plausible dilution/no benefit
Head injury-related death within 24 h (treated within 3 h) Not reported Not reported RR 0·81 95% CI 0·69 to 0·95; p value not reported Direction and timing consistent with an effect on early bleeding-related deaths
All vascular occlusive events (treated within 3 h) 69/4613 (1·5%) 60/4514 (1·3%) RR 1·13 95% CI 0·80 to 1·59; p value not reported Events required imaging or post-mortem confirmation (high specificity)
Seizure (treated within 3 h) 130/4613 (2·8%) 105/4514 (2·3%) RR 1·21 95% CI 0·94 to 1·56; p value not reported All-patient analysis: 206/6359 (3·2%) vs 186/6280 (3·0%); RR 1·09 (0·90–1·33)
Patient-derived disability: Unable to wash or dress (treated within 3 h) 580/4613 (12·6%) 583/4514 (12·9%) RR 0·97 95% CI 0·87 to 1·08; p value not reported Disability outcomes recorded at discharge or day 28 (whichever first)
Disability Rating Scale (DRS) score among survivors (treated within 3 h) Mean 4·99 (SD 7·04); n=2758 Mean 5·03 (SD 7·16); n=2748 Not reported Not reported Continuous disability measure; no between-group estimate reported
  • In the prespecified primary population (treated within 3 h), TXA showed a modest reduction in head injury-related death (RR 0·94; 95% CI 0·86–1·02), with a larger effect after excluding moribund patients (RR 0·89; 0·80–1·00).
  • Signals of benefit were concentrated in patients with milder physiological injury severity proxies (GCS 9–15; both pupils reactive), supporting the hypothesis that TXA mainly helps when there is salvageable intracranial bleeding.
  • Major safety outcomes showed no clear excess harm, although vascular occlusive event ascertainment was intentionally high-specificity (imaging/post-mortem confirmed), which can reduce sensitivity.

Internal Validity

  • Randomisation and allocation concealment: Allocation by selection of a numbered, identical treatment pack from a box containing eight packs; packs identical apart from pack number; clinicians had no foreknowledge of allocation; masking applied to patients, caregivers, and outcome assessors.
  • Blinding integrity: Unmasking occurred in 13 patients (6 TXA; 7 placebo); this is unlikely to materially bias mortality outcomes but indicates that emergency unblinding pathways were used.
  • Adherence and separation of the variable of interest: Loading dose given to 6314/6406 (98·6%) in TXA vs 6247/6331 (98·7%) in placebo; maintenance dose given to 5984/6406 (93·4%) in TXA vs 5882/6331 (92·9%) in placebo.
  • Timing feasibility: In the treated-within-3h cohort, mean time since injury at randomisation was 1·9 h (SD 0·7) in both groups; 9202/12,737 (72·2%) of all randomised participants were treated within 3 h.
  • Missing data and exclusions: Head injury-related death outcome missing in 37/4613 (0·8%) TXA vs 34/4514 (0·8%) placebo in the within-3h primary population; analysis was ITT using available outcomes (no imputation stated).
  • Baseline comparability (within 3 h): Mean age 41·7 (SD 19·0) TXA vs 41·9 (SD 19·0) placebo; men 3742/4613 (81·1%) vs 3660/4514 (81·1%); baseline pupil reactivity and GCS categories were closely balanced.
  • Outcome assessment: Primary outcome (head injury-related death) was assigned by the responsible clinician; assessment was masked to allocation, but cause attribution is inherently more subjective than all-cause mortality and could be vulnerable to non-differential misclassification.
  • Heterogeneity and effect modification: Prespecified subgroup heterogeneity by severity (GCS 9–15 vs 3–8) and pupil reactivity was statistically evident (heterogeneity p=0·030 and p=0·032, respectively); time-to-treatment subgroup RRs for head injury-related death were similar across time strata in the unadjusted analysis (≤1 h RR 0·96; 95% CI 0·79–1·17; >1–3 h RR 0·93; 0·85–1·02; >3 h RR 0·94; 0·81–1·09).
  • Protocol changes: Recruitment window changed (8 h to 3 h) and primary endpoint focus changed (to head injury-related death within 28 days among those treated within 3 h) during the trial; this was undertaken blind to unblinded outcome data but introduces methodological complexity and interpretive risk (addressed in Controversies).
  • Statistical rigour: Large sample size; ITT analysis; prespecified sensitivity analysis; multiple subgroup analyses require careful interpretation; confidence intervals for the primary endpoint include both clinically relevant benefit and no effect.

Conclusion on Internal Validity: Overall, internal validity is moderate-to-strong: randomisation and masking were robust, adherence was high, and missing outcome data were minimal, but interpretive confidence is tempered by the mid-trial protocol/endpoint modification and by reliance on clinician-assigned cause of death for the primary endpoint.

External Validity

  • Population representativeness: Adults with suspected TBI across 175 hospitals in 29 countries, including a wide range of healthcare settings; typical trauma demographics (mean age ~42 years; ~80% male in the within-3h cohort).
  • Clinical spectrum: Included a broad range of severities (including very severe injuries with extremely high mortality); the most consistent signals of benefit were in patients with mild-to-moderate injury and reactive pupils.
  • Important exclusions: Patients with major extracranial bleeding were excluded, so inference to polytrauma with haemorrhagic shock requires caution (and is better informed by CRASH-2).
  • Setting and deliverability: Intervention is simple (IV bolus + infusion) and feasible in ED and prehospital systems with IV capability; benefit appears time-dependent and therefore depends on system speed to first dose.
  • Subpopulations not directly informed: Children, penetrating TBI, delayed presentation beyond 3 h, and patients where clinicians were certain TXA was indicated or contraindicated.

Conclusion on External Validity: External validity is good for adult blunt TBI patients who can receive IV TXA rapidly (within ~3 h) and who do not have major extracranial bleeding; generalisability is more limited for paediatric, penetrating, and delayed-presentation TBI.

Strengths & Limitations

  • Strengths:
    • Very large, international, pragmatic randomised trial with double-blind placebo control.
    • High protocol adherence to the intended TXA dosing regimen and strong separation between study arms.
    • Low missingness for the primary outcome and prespecified sensitivity analyses to address predictable dilution by moribund patients.
    • Clinically implementable intervention and a clear biologically plausible time-sensitive mechanism.
  • Limitations:
    • Protocol amendment restricted enrolment time window (8 h to 3 h) and modified the primary analysis focus during the trial, increasing interpretive complexity.
    • Primary endpoint relied on clinician classification of cause of death; non-differential misclassification could attenuate observed effects.
    • Disability outcomes were assessed at discharge/28 days, which might be too early to capture longer-term neurological recovery trajectories.
    • Thromboembolic event capture used high-specificity criteria (imaging/post-mortem), which may underestimate total event incidence (but is less likely to bias relative comparisons).

Interpretation & Why It Matters

  • Time-critical therapy
    TXA should be considered an early intervention for eligible adult TBI patients, with maximal plausibility and observed signal when administered as soon as possible and within 3 hours of injury.
  • Who benefits most
    Observed benefit was concentrated in patients with milder physiological severity (GCS 9–15) and reactive pupils, consistent with an effect on preventable intracranial haemorrhage expansion rather than on established herniation/irreversible injury.
  • Safety and trade-offs
    Across prespecified safety outcomes, there was no clear evidence of excess vascular occlusive events; seizure estimates were compatible with no effect and with modest harm, emphasising the importance of local monitoring and balanced counselling.

Controversies & Subsequent Evidence

  • Endpoint selection and timing: The accompanying Lancet Comment argued that using 28-day head injury-related mortality as the primary endpoint probably biased the effect towards the null, because TXA is most plausible for early bleeding-related deaths; it advocated for future trials to focus on earlier bleeding-related endpoints and highlighted the possibility of uncaptured venous thromboembolism and uncertainties about optimal dosing/regimen.1
  • Subgroup inference and statistical thresholds: Correspondence questioned interpretive emphasis on pooled “mild-to-moderate” subgroups and highlighted the risk of over-interpreting subgroup findings without stringent interaction thresholds, particularly given the heterogeneity within “severe” TBI and the known instability of severity proxies (GCS, pupils) early after injury.2
  • Time-to-treatment confounding: Correspondence noted that patients treated earlier can be systematically more severely injured (thus at higher baseline risk), which can obscure an underlying “earlier is better” effect in unadjusted analyses; this strengthens the case for prespecified adjusted models when assessing effect modification by delay.3
  • Choice of head injury-related death vs all-cause mortality: Correspondence raised concerns about potential misclassification and interpretive bias when using cause-specific mortality as a primary outcome; this debate reflects a classic trade-off between biological specificity and measurement error in pragmatic trials of complex critical illness syndromes.4
  • Trialists’ reply: The CRASH-3 investigators responded that misclassification of cause of death assessed blind to treatment would bias towards the null rather than create a spurious benefit; they also clarified the interpretation of interaction testing and defended the analytic choices made to address confounding in time-to-treatment analyses.5
  • Subsequent evidence (overview): Post-CRASH-3 randomised and observational evidence has generally supported TXA as a low-cost, implementable early intervention for selected TBI patients, with the balance of evidence favouring early treatment and greatest plausibility in mild-to-moderate injury; uncertainty persists regarding benefit in very severe injury, delayed treatment, and the degree to which prehospital delivery changes outcomes (see Further Reading).

Summary

  • CRASH-3 randomised 12,737 adults with TBI to TXA vs placebo; 9,202 (72·2%) were treated within 3 hours and formed the primary analysis population.
  • Primary outcome (treated within 3 h): head injury-related death 18·5% with TXA vs 19·8% with placebo (RR 0·94; 95% CI 0·86–1·02).
  • Prespecified sensitivity excluding moribund patients (GCS 3 or bilateral fixed pupils) showed a larger, borderline effect (12·5% vs 14·0%; RR 0·89; 0·80–1·00).
  • Benefit signal was most evident in mild-to-moderate injury and reactive pupils; no clear benefit was observed in severe injury strata.
  • Safety outcomes showed no clear excess vascular occlusive events; seizure estimates were compatible with both no effect and modest harm.

Further Reading

Other Trials

Systematic Review & Meta Analysis

Observational Studies

Guidelines

Notes

  • Where “Not reported” appears in Further Reading, the DOI link provides definitive bibliographic details on the journal landing page.
  • CRASH-3 used high-specificity definitions for thromboembolic outcomes (imaging/post-mortem), which improves causal interpretability of group comparisons but can underestimate total incidence.

Overall Takeaway

CRASH-3 is a landmark, large-scale pragmatic randomised trial showing that early tranexamic acid in adult TBI is safe and likely reduces head injury-related death when given within 3 hours, with the clearest signal in patients who still have salvageable physiology (mild-to-moderate injury and reactive pupils). The trial’s nuanced subgroup pattern, time-critical mechanism, and low logistical burden shifted TXA from “polytrauma-only” thinking towards routine consideration in isolated TBI pathways, while preserving uncertainty for very severe injury and delayed treatment.

Overall Summary

  • Large, double-blind RCT across 29 countries; simple intervention (1 g IV bolus + 1 g infusion) deliverable in real-world systems.
  • Primary result: modest overall mortality signal within 3 h, with stronger effect after excluding moribund patients and in milder injury strata.
  • Safety profile reassuring for vascular occlusive events; seizure risk estimates remain compatible with no effect or modest harm.

Bibliography