Publication

• Title: Effects of a high-dose 24-h infusion of tranexamic acid on death and thromboembolic events in patients with acute gastrointestinal bleeding (HALT-IT): an international randomised, double-blind, placebo-controlled trial
• Acronym: HALT-IT
• Year: 2020
• Journal published in: The Lancet
• Citation: Roberts I, Shakur-Still H, Afolabi A, et al; HALT-IT Trial Collaborators. Effects of a high-dose 24-h infusion of tranexamic acid on death and thromboembolic events in patients with acute gastrointestinal bleeding (HALT-IT): an international randomised, double-blind, placebo-controlled trial. Lancet. 2020;395(10241):1927-1936.

Context & Rationale

• Background

  • Acute severe gastrointestinal bleeding is common, often requires urgent resuscitation and haemostatic interventions, and carries substantial mortality risk.
  • Tranexamic acid (TXA) reduces death due to bleeding in other haemorrhage syndromes (eg, trauma, postpartum haemorrhage) and reduces surgical bleeding, motivating interest in gastrointestinal bleeding as an additional high-burden indication.
  • Pre-HALT-IT evidence in gastrointestinal bleeding comprised small trials with important bias concerns and inadequate power to evaluate thromboembolic harms.
  • A protocolised, large-scale evaluation was needed to reliably estimate benefit–harm, including venous and arterial thromboembolic events and seizures.¹
• Research Question/Hypothesis

  • In adults with clinically significant acute upper or lower gastrointestinal bleeding, does TXA (1 g loading dose plus 3 g maintenance infusion over 24 h) reduce death due to bleeding (and related clinical outcomes) compared with placebo, without excess thromboembolic events or seizures?
• Why This Matters

  • TXA is inexpensive, widely available, and already embedded in major haemorrhage pathways in other domains; even modest benefit in gastrointestinal bleeding could have high population impact.
  • Conversely, gastrointestinal bleeding populations (including patients with cirrhosis/varices) may have altered coagulation–fibrinolysis balance, raising plausible harm signals that require definitive quantification in a large randomised trial.
  • The trial’s pragmatic, international design was intended to support generalisable decision-making, and to resolve the credibility gap between small-trial meta-analyses and large multicentre RCTs.²

Design & Methods

• Research Question: Among adults with clinically significant acute gastrointestinal bleeding where the responsible clinician is substantially uncertain about TXA use, does TXA (high-dose 24-h infusion) reduce death due to bleeding compared with placebo?
• Study Type: International, multicentre, randomised, double-blind (participants and trial staff), placebo-controlled, pragmatic trial conducted in 164 hospitals across 15 countries (including emergency and inpatient settings).
• Population:
  • Setting: Hospital care across 15 countries (UK, Pakistan, Nigeria, Egypt, Malaysia, Georgia, Romania, Nepal, Sudan, Saudi Arabia, Spain, Ireland, Albania, Papua New Guinea, Australia).
  • Inclusion: Aged above the minimum adult age in the relevant country (≥16 or ≥18 years) with clinically diagnosed significant upper or lower gastrointestinal bleeding where the clinician was substantially uncertain about TXA use.
  • “Significant” bleeding: Defined clinically as risk of bleeding to death, including hypotension, tachycardia, or signs of shock, or likely need for transfusion or urgent endoscopy or surgery.
  • Exclusions: Not fully enumerated in the main report; key reported exclusions/events included age below eligibility threshold, and TXA given prior to randomisation in a small number of protocol violations.
• Intervention:
  • TXA 1 g loading dose added to 100 mL 0.9% sodium chloride and infused intravenously over 10 minutes, then TXA 3 g added to 1 L isotonic solution infused at 125 mg/h for 24 hours.
• Comparison:
  • Matching placebo (0.9% sodium chloride) delivered using identical loading and maintenance infusion procedures and masking.
• Blinding: Double-blind with identical treatment packs and labelling; clinicians, patients, and outcome assessors masked; limited unmasking occurred for clinical reasons.
• Statistics: A total of 12,000 patients provided ~85% power (two-sided α 5%) to detect a 25% relative reduction in death due to bleeding within 5 days (from 4% to 3%); primary analysis was modified intention-to-treat (excluding patients receiving neither dose and those without death outcome data); a prespecified statistical analysis plan was published prior to unblinding.³
• Follow-Up Period: Outcomes collected to death, discharge from the randomising hospital, or 28 days after randomisation (whichever occurred first).

Key Results

This trial was not stopped early. Recruitment stopped when the planned sample size was reached (12,009 randomised; primary analysis population 5,956 TXA vs 5,981 placebo).

• TXA did not reduce death due to bleeding within 5 days (RR 0.99; 95% CI 0.82 to 1.18) despite a large, adequately powered sample.
• Harms were clinically important: venous thromboembolic events increased (0.8% vs 0.4%; RR 1.85; 95% CI 1.15 to 2.98) and seizures increased (0.6% vs 0.4%; RR 1.73; 95% CI 1.03 to 2.93).
• Prespecified subgroup analyses showed no heterogeneity for the primary endpoint by time to treatment (≤3 h: RR 1.10; 95% CI 0.75 to 1.61; >3 h: RR 0.96; 95% CI 0.78 to 1.18; heterogeneity P=0.53), suspected bleeding site (upper: RR 0.97; 95% CI 0.81 to 1.17; lower: RR 1.61; 95% CI 0.59 to 4.40; P=0.34), suspected variceal/liver disease (yes: RR 1.01; 95% CI 0.81 to 1.24; no/unknown: RR 0.99; 95% CI 0.70 to 1.40; P=0.94), or pre-endoscopy Rockall score (P=0.32).

Internal Validity

• Randomisation and Allocation:
  • Randomisation numbers generated independently (Sealed Envelope) and implemented through sequentially numbered treatment packs; block randomisation without stratification.
  • Allocation concealment operationalised via identical packs and labelling; “lowest numbered pack” selection reduces clinician discretion and selection bias risk.
• Drop out or exclusions:
  • Randomised: 5,994 TXA vs 6,015 placebo.
  • Received loading dose: 5,964/5,994 (99.5%) TXA vs 5,988/6,015 (99.5%) placebo.
  • Received neither dose: 29 TXA vs 23 placebo.
  • Primary analysis population: 5,956 TXA vs 5,981 placebo (modified intention-to-treat; excludes those receiving neither dose and those without death outcome data).
  • Primary outcome data missing: 3 patients in TXA group; effectively complete ascertainment for death outcomes otherwise.
• Performance/Detection Bias:
  • Double blinding reduces differential co-intervention and ascertainment bias.
  • Unmasking was uncommon: 28 total unmasked (13 TXA; 15 placebo), largely for clinical reasons (including desire to administer TXA and adverse events).
• Protocol Adherence:
  • Maintenance dose received: 5,483/5,994 (91.5%) TXA vs 5,510/6,015 (91.6%) placebo.
  • Open-label antifibrinolytic use occurred in both groups: 105 TXA vs 118 placebo (potential dilution towards the null; sensitivity analysis excluding these patients was similar).
  • Protocol violations were rare (14 total); inclusion violations included TXA given prior to randomisation and one under-age participant.
• Baseline Characteristics:
  • Age: mean 58.1 years (SD 17.0) in both groups.
  • Female sex: 2,142/5,994 (36%) TXA vs 2,124/6,015 (35%) placebo.
  • Time from bleeding onset to randomisation: mean 21.4 h (SD 36.4) TXA vs 22.5 h (SD 37.8) placebo; ≤3 h: 960 (16%) vs 975 (16%); >8 h: 3,427 (57%) vs 3,488 (58%).
  • Suspected bleeding site: upper 5,320 (89%) vs 5,361 (89%); lower 674 (11%) vs 654 (11%).
  • Suspected variceal bleeding: 2,694 (45%) vs 2,739 (46%).
  • Physiology: SBP ≤75 mm Hg in 181 (3%) vs 201 (3%); SBP 76–89 mm Hg in 577 (10%) vs 577 (10%).
• Heterogeneity:
  • No evidence of heterogeneity for the primary endpoint across prespecified subgroup factors (time to treatment, bleeding location, suspected variceal/liver disease, Rockall score).
  • Exploratory signal: VTE excess appeared larger in suspected variceal bleeding/liver disease (14 vs 2 events; RR 7.26; 95% CI 1.65 to 31.90) than other/unknown causes (34 vs 24; RR 1.38; 95% CI 0.82 to 2.32; heterogeneity P=0.035); exploratory and not prespecified.
• Timing:
  • Treatment was frequently late relative to reported bleeding onset (only 16% within 3 h), potentially limiting the biological plausibility of benefit if any effect is strongly time-dependent.
• Dose:
  • High-dose, prolonged regimen (total 4 g over 24 h) ensures biological exposure but may contribute to seizure and thrombotic risks; the observed seizure/VTE excess supports a clinically material harm signal.
• Separation of the Variable of Interest:
  • Intervention separation: TXA infusion regimen versus matched placebo with high adherence to dosing.
  • Co-interventions were broadly balanced: therapeutic endoscopy 42.7% vs 44.5% (RR 0.96; 95% CI 0.92 to 1.00); diagnostic endoscopy 80.3% vs 79.1% (RR 1.02; 95% CI 1.00 to 1.03).
• Key Delivery Aspects:
  • Pragmatic eligibility and delivery maximised real-world relevance but reduced precision in onset timing and aetiology at enrolment.
• Outcome Assessment:
  • Primary endpoint (death due to bleeding within 5 days): locally assigned cause with narrative review by the masked chief investigator.
  • Thromboembolic events: strict diagnostic criteria including imaging/biomarkers/post-mortem (high specificity; potential under-ascertainment but directionally less likely to bias relative effects if non-differential).
  • Rebleeding: clinician diagnosis using established criteria (more subjective than mortality endpoints).
• Statistical Rigor:
  • Prespecified statistical analysis plan was published before unblinding, including rationale for the amended primary outcome and sample size increase.³
  • Four non-masked interim analyses were reviewed by an independent data monitoring committee.
  • Effect estimates presented as risk ratios with 95% CIs; sensitivity analyses (covariate-adjusted and per-protocol) were directionally consistent with the primary analysis.

Conclusion on Internal Validity: Overall, internal validity appears strong given robust concealment and masking, near-complete follow-up, high protocol adherence with limited unmasking, and prespecified analysis planning; the principal interpretive vulnerability is outcome misclassification for cause-specific death and the pragmatic uncertainty in onset/aetiology at enrolment.

External Validity

• Population Representativeness:
  • Large, international cohort across high-income and low- and middle-income settings; enrolled patients reflected the spectrum of suspected upper and lower gastrointestinal bleeding encountered in routine practice.
  • High prevalence of suspected variceal bleeding/liver disease (≈45–46%) increases applicability to mixed-cause “real-world” emergency bleeding, including cirrhosis-heavy case-mix.
  • Entry criterion of “clinician uncertainty” may exclude patients where TXA use was already standardised or contraindicated locally.
• Applicability:
  • Findings apply directly to the tested regimen (4 g over 24 h); extrapolation to lower-dose or shorter-duration TXA in gastrointestinal bleeding is not directly supported.
  • Most patients were treated >3 h after bleeding onset (mean ~22 h), so generalisability to very early presenters is limited; the ≤3 h subgroup did not show benefit, but precision remains limited (RR 1.10; 95% CI 0.75 to 1.61).
  • Resource-limited systems: pragmatic delivery and broad hospital inclusion support feasibility, but diagnostic adjudication for thromboembolic events may be less complete where imaging capacity is constrained.

Conclusion on External Validity: Generalisability is high for typical in-hospital management of significant gastrointestinal bleeding across diverse health systems, but is more limited for early-presenting patients and for alternative TXA dosing strategies not tested in HALT-IT.

Strengths & Limitations

• Strengths:
  • Very large sample size (12,009 randomised) with near-complete ascertainment of mortality outcomes.
  • International, multicentre pragmatic design across 164 hospitals in 15 countries improves relevance and reduces single-system idiosyncrasy.
  • Double-blind placebo-controlled methods with robust operational masking and limited unmasking.
  • Clinically meaningful safety endpoints captured, including venous thromboembolism and seizures.
  • Statistical analysis plan published prior to unblinding, with explicit rationale for the amended primary outcome and increased sample size.³
• Limitations:
  • Treatment was frequently administered long after reported bleeding onset (only 16% within 3 h), potentially attenuating any time-sensitive efficacy signal.
  • Primary outcome was cause-specific mortality (death due to bleeding) which can be misclassified despite masking and central narrative review; however, all-cause mortality also showed no benefit.
  • Pragmatic enrolment before definitive endoscopic diagnosis introduces uncertainty in aetiology and site classification for subgroup analyses.
  • Modified intention-to-treat analysis excluded patients receiving neither dose and a small number without death outcome data; magnitude of exclusions was small but directionally could reduce pure ITT interpretability.
  • High-dose, prolonged infusion strategy may not reflect all settings’ TXA practices and may have contributed to seizure risk.

Interpretation & Why It Matters

• Clinical signal

  TXA (4 g over 24 h) did not reduce bleeding death or all-cause mortality in acute gastrointestinal bleeding, and increased venous thromboembolism and seizures; routine use for this indication is not supported.
• Mechanistic alignment

  Observed harms are biologically credible given fibrinolysis heterogeneity in gastrointestinal bleeding (especially cirrhosis/varices) and dose-related neurotoxicity concerns; benefit signals seen in trauma/postpartum haemorrhage did not translate to this syndrome.
• Methods-to-practice lesson

  HALT-IT illustrates the fragility of conclusions from small-trial meta-analyses in high-stakes haemostatic therapy, reinforcing the necessity of adequately powered, bias-resistant pragmatic RCTs for adoption decisions.²

Controversies & Subsequent Evidence

• Primary outcome change during the trial:
  • The trial amended its primary outcome from all-cause mortality to death due to bleeding within 5 days, with sample size increase; rationale centred on observed non-bleeding death predominance and TXA’s short half-life, and was pre-specified in the published analysis plan.³
  • The accompanying editorial regarded the amendment as methodologically defensible but emphasised that all-cause mortality remains the most patient-relevant endpoint and that adverse event ascertainment warrants scrutiny when harms are uncommon.²
• Timing of TXA relative to bleeding onset:
  • Bleeding onset is difficult to time in gastrointestinal haemorrhage, and the majority were randomised >3 h after onset (≈84%), raising concerns that HALT-IT primarily tested late TXA; prespecified analysis showed no benefit even in ≤3 h subgroup, but with limited precision.
• Dose and harm profile:
  • The high-dose 24-h infusion strategy produced an unfavourable safety signal (VTE and seizures), strengthening the inference that net clinical benefit is unlikely for this regimen in unselected gastrointestinal bleeding.
• Guideline incorporation:
  • Subsequent major endoscopy guidance recommends against TXA for nonvariceal upper gastrointestinal haemorrhage in routine care, reflecting evidence including HALT-IT and the observed harm profile.⁴
• Position relative to pre-HALT-IT evidence:
  • HALT-IT effectively rules out the large mortality benefit suggested by earlier small-trial syntheses, and shifts the evidentiary balance towards harm in broad gastrointestinal bleeding populations.⁵

Summary

• HALT-IT randomised 12,009 adults with clinically significant acute gastrointestinal bleeding to TXA (1 g + 3 g/24 h) versus placebo across 164 hospitals in 15 countries.
• TXA did not reduce death due to bleeding within 5 days (3.7% vs 3.8%; RR 0.99; 95% CI 0.82 to 1.18) and did not improve all-cause mortality at 28 days (9.5% vs 9.2%; RR 1.03; 95% CI 0.92 to 1.16).
• TXA did not materially reduce rebleeding or downstream haemostatic interventions (eg, therapeutic endoscopy 42.7% vs 44.5%; RR 0.96; 95% CI 0.92 to 1.00).
• TXA increased venous thromboembolic events (0.8% vs 0.4%; RR 1.85; 95% CI 1.15 to 2.98) and seizures (0.6% vs 0.4%; RR 1.73; 95% CI 1.03 to 2.93).
• The trial’s pragmatic design and scale make its “no benefit + harm” conclusion highly practice-shaping for routine TXA use in gastrointestinal bleeding.

Overall Takeaway

HALT-IT is a landmark haemostasis trial because it definitively demonstrated that high-dose 24-hour TXA does not reduce bleeding death or overall mortality in acute gastrointestinal bleeding, while increasing clinically meaningful harms (VTE and seizures). Its scale, pragmatic design, and robust masking make the “no benefit + harm” conclusion highly practice-shaping and a cautionary counterpoint to earlier small-trial meta-analytic signals.

Bibliography

• 1. Roberts I, Coats T, Edwards P, et al. HALT-IT—tranexamic acid for the treatment of gastrointestinal bleeding: study protocol for a randomised controlled trial. Trials. 2014;15:450.
• 2. Perner A, Møller MH. Tranexamic acid for severe gastrointestinal bleeding. Lancet. 2020;395(10241):1885-1886.
• 3. Brenner A, Afolabi A, Ahmad SM, et al. Tranexamic acid for acute gastrointestinal bleeding (the HALT-IT trial): statistical analysis plan for an international, randomised, double-blind, placebo-controlled trial. Trials. 2019;20:467.
• 4. Gralnek IM, Stanley AJ, Morris AJ, et al. Endoscopic diagnosis and management of nonvariceal upper gastrointestinal haemorrhage: European Society of Gastrointestinal Endoscopy (ESGE) Guideline. Endoscopy. 2021;53(3):300-332.
• 5. Bennett C, Klingenberg SL, Langholz E, Gluud LL. Tranexamic acid for upper gastrointestinal bleeding. Cochrane Database Syst Rev. 2014;CD006640.