Skip to main content

Trial Summaries & Critiques

TRACTION

Houston BL, McIsaac DI, Breau RH, Kanji S, Greenstreet P, Andrews M, et al. Hospital policy of tranexamic acid to reduce transfusion in major noncardiac surgery. N Engl J Med 2026;epublished June 10th

Image: Shutterstock / luchschenF

Box of tranexamic acid

Publication

Context & Rationale

  • Background
    • Perioperative bleeding is a major driver of red-cell transfusion in hospitalised patients.
    • Red-cell transfusion is a constrained, costly, and clinically consequential resource, with risks including transfusion reactions, infection, immune modulation, prolonged hospital stay, and downstream morbidity.
    • Tranexamic acid is an inexpensive antifibrinolytic drug that inhibits clot breakdown and reduces bleeding in several surgical and bleeding contexts.
    • Before TRACTION, tranexamic acid was already established in cardiac surgery and hip and knee arthroplasty, but use in other major noncardiac surgery was inconsistent.
    • The TRACTION protocol was designed specifically to test a hospital policy rather than a narrow individual-patient efficacy question, because real-world adoption depends on institutional workflows, anaesthesia practice, pharmacy delivery, and clinician confidence in safety.1
    • A pre-TRACTION systematic review of 69 RCTs in major noncardiac surgery at high risk for transfusion found that tranexamic acid reduced red-cell transfusion, with RR 0.59; 95% CI 0.48 to 0.72; 6157 patients, but certainty was limited by heterogeneity and many modest-sized trials.2
    • The POISE-3 trial enrolled 9535 adults undergoing noncardiac surgery at risk for bleeding and cardiovascular complications; tranexamic acid reduced a major bleeding composite, but noninferiority was not established for the primary 30-day cardiovascular safety composite.3
    • Persistent concern about venous thromboembolism was especially important in patients with cancer or other prothrombotic conditions, who are frequently excluded or under-represented in perioperative tranexamic acid trials.
  • Research Question/Hypothesis
    • TRACTION tested whether implementing a hospital-wide policy of intraoperative tranexamic acid in adults undergoing major noncardiac surgery at high risk for red-cell transfusion would reduce transfusion during the index hospitalisation.
    • The trial simultaneously tested whether such a policy was noninferior to placebo with respect to venous thromboembolism within 90 days.
    • The central hypothesis was that routine perioperative tranexamic acid would reduce red-cell transfusion without a clinically important increase in deep-vein thrombosis or pulmonary embolism.
  • Why This Matters
    • Major noncardiac surgery represents a large, heterogeneous, daily source of perioperative blood use.
    • A policy-level intervention that safely prevents transfusion is attractive because tranexamic acid is cheap, widely available, simple to administer, and scalable.
    • The trial addressed the implementation question most relevant to hospitals: whether routine policy use across eligible operations is beneficial and safe, not merely whether tranexamic acid works in selected operations under explanatory trial conditions.
    • The large proportion of oncologic surgery made TRACTION especially relevant to the group in whom thrombosis concerns most often constrain tranexamic acid adoption.

Design & Methods

  • Research Question:
    • In adults undergoing major noncardiac surgery with an estimated red-cell transfusion risk of at least 5%, does a hospital policy of intraoperative tranexamic acid reduce red-cell transfusion during the index hospitalisation without increasing venous thromboembolism within 90 days?
  • Study Type:
    • Multicentre, double-blind, placebo-controlled, registry-based, cluster-randomised, cluster-crossover trial.
    • Investigator-initiated phase IV pragmatic trial.
    • Setting: operating theatres and perioperative inpatient care in 10 Canadian hospitals.
    • Hospitals were randomly assigned at 4-week intervals to a hospital-wide policy of tranexamic acid or matching placebo.
    • The trial used linked clinical and administrative data for central electronic outcome ascertainment.
  • Population:
    • Adults aged ≥18 years.
    • Undergoing major noncardiac inpatient surgery.
    • Estimated risk of red-cell transfusion ≥5%.
    • Eligible surgeries broadly included open operations or laparoscopic/endoscopic operations expected to last at least 3 hours.
    • Examples included major general, orthopaedic trauma, spine, otolaryngology, thoracic, vascular, gynaecological, urological, plastic, and other operations judged by the surgical team to carry ≥5% transfusion risk.
    • Hospitals were eligible if they performed at least 100 noncardiac operations per month and anaesthesia, surgical, and hospital leadership agreed to implement the assigned policy.
    • Key exclusions were active thromboembolic disease, cardiac surgery, hip or knee total arthroplasty, surgery involving a free flap, trauma surgery in which tranexamic acid had been administered within the previous 3 hours, and pregnancy.
  • Intervention:
    • Hospital policy of intraoperative tranexamic acid.
    • Tranexamic acid 1 g IV bolus at the start of surgery.
    • Patients weighing >100 kg received a 2 g IV bolus.
    • An additional 1 g IV was administered before skin closure, with timing at the discretion of the anaesthesiologist.
    • The dosing regimen reflected clinical use and pharmacokinetic rationale from earlier perioperative studies.
  • Comparison:
    • Hospital policy of matching placebo.
    • Placebo was 0.9% sodium chloride, supplied and delivered to match the tranexamic acid intervention.
    • Other perioperative care, including transfusion decisions, cell salvage, topical tranexamic acid, and thromboprophylaxis, was left to treating clinicians and documented but not protocolised.
    • Blinded emergency investigational product containing the opposite of the site’s current allocation was available for life-threatening haemorrhage if the clinical team needed to ensure tranexamic acid exposure without unblinding the cluster allocation.
  • Blinding:
    • Double-blind.
    • Patients, healthcare providers, research staff, investigators, and the trial statistician were unaware of the randomisation schedules and investigational product administered.
    • Research pharmacy staff prepared study drug according to the interval assignment but were separated from clinical and research teams.
    • Blinding was methodologically important because transfusion decisions can be clinician-sensitive, even though the key safety outcome was centrally ascertained from electronic data.
  • Statistics:
    • Power calculation: A total sample size of approximately 8320 patients was planned; assuming an average cluster-period size of 130, intraclass correlation 0.005, and cluster autocorrelation 0.85, the trial had 99.9% power to detect a 6% absolute reduction in red-cell transfusion from a baseline of 18%, and 83% power to exclude a ≥1% absolute increase in 3-month VTE from a baseline of 2.2% with a one-sided alpha of 0.025 for noninferiority.1
    • The coprimary effectiveness outcome was red-cell transfusion during the index hospitalisation.
    • The coprimary safety outcome was diagnosis of deep-vein thrombosis or pulmonary embolism within 90 days.
    • The VTE noninferiority margin was defined as an upper boundary of 1.46 for the 95% CI of the relative risk, corresponding to a 1-percentage-point absolute increase if baseline VTE risk was 2.2%.
    • Both coprimary outcomes had to be satisfied to declare the tranexamic acid policy beneficial, so no multiplicity adjustment was required for the coprimary decision.
    • The primary effectiveness analysis used the intention-to-treat population.
    • The primary noninferiority safety analysis used the per-protocol population, with an intention-to-treat analysis also reported.
    • Mixed-effects models accounted for the cluster-crossover design and period effects.
    • No formal interim analyses were planned because the 90-day VTE safety outcome needed to mature.
  • Follow-Up Period:
    • Index hospitalisation for red-cell transfusion and several in-hospital secondary outcomes.
    • 30 days for days alive and out of hospital.
    • 90 days for venous thromboembolism and survival.

Key Results

This trial continued to completion. No formal interim analyses were planned; enrolment ran from February 2022 through March 2024 across 25 consecutive 4-week periods.

Outcome Tranexamic acid Placebo Effect p value / 95% CI Notes
Patients randomised 4222 assigned 4199 assigned NA NA 8273/8421 (98.2%) had successful data linkage and formed the primary intention-to-treat population.
Red-cell transfusion during index hospitalisation 306/4156 (7.4%) 403/4117 (9.8%) RR 0.73; adjusted difference -2.7 percentage points 95% CI 0.61 to 0.86; adjusted difference 95% CI -4.2 to -1.4; no p value reported Coprimary effectiveness outcome; number needed to treat 37.
VTE within 90 days, per-protocol population 86/4128 (2.1%) 85/4052 (2.1%) RR 0.96; adjusted difference -0.1 percentage points 95% CI 0.65 to 1.38; adjusted difference 95% CI -0.9 to 0.7; no p value reported Coprimary safety outcome; met noninferiority because the upper 95% CI boundary was below 1.46.
VTE within 90 days, intention-to-treat population 86/4156 (2.1%) 90/4117 (2.2%) RR 0.92 95% CI 0.62 to 1.30; no p value reported Consistent with the per-protocol noninferiority analysis.
Red-cell units transfused, patient level per 100 patients 24.7 ± 3.6 34.2 ± 2.6 Mean difference -9.5 units per 100 patients 95% CI -19.2 to -0.2; no p value reported Fewer red-cell units were used under the tranexamic acid policy.
Red-cell units transfused, hospital level per 100 patients 23.5 ± 27.2 39.6 ± 80.6 Mean difference -16.2 units per 100 patients 95% CI -33.5 to 1.1; no p value reported Direction favoured tranexamic acid; wide interval reflected cluster-level variability.
Myocardial infarction, in hospital 29/4156 (0.7%) 34/4117 (0.8%) Not analysed Not reported Limited event rate precluded reliable between-group comparison.
Stroke, in hospital 9/4156 (0.2%) 9/4117 (0.2%) Not analysed Not reported Limited event rate.
Deep-vein thrombosis, in hospital <6 <6 Not analysed Not reported Small cell sizes suppressed for privacy.
Pulmonary embolism, in hospital 8/4156 (0.2%) 6/4117 (0.1%) Not analysed Not reported Low event rate.
Hospital length of stay 6.20 ± 8.89 days 6.26 ± 9.42 days Mean difference -0.15 days 95% CI -0.69 to 0.29; no p value reported No clinically meaningful separation.
ICU admission 693/4156 (16.7%) 721/4117 (17.5%) Difference -0.01 95% CI -0.03 to 0.02; no p value reported Similar.
Hospital survival 4112/4156 (98.9%) 4076/4117 (99.0%) Not analysed Not reported High and similar.
Overall survival at 90 days 4066/4156 (97.8%) 4017/4117 (97.6%) HR for death 0.95 95% CI 0.71 to 1.28; no p value reported No survival signal; model convergence issues meant the conditional estimate was reported.
Median days alive and out of hospital to day 30 26 days (IQR 22 to 27) 26 days (IQR 22 to 27) Mean difference 0.03 days 95% CI -0.33 to 0.39; no p value reported No separation in this patient-centred recovery metric.
Prespecified serious adverse events 1 perioperative seizure 1 perioperative anaphylaxis Not analysed Not reported Serious adverse events of special interest were rare.
Oncologic surgery subgroup: red-cell transfusion 161/2536 (6.3%) 241/2466 (9.8%) RR 0.63 95% CI 0.49 to 0.77; no p value reported Clinically important subgroup because 5002/8273 (60.5%) operations were oncologic.
Oncologic surgery subgroup: VTE within 90 days 60/2526 (2.4%) 64/2421 (2.6%) RR 0.92 95% CI 0.68 to 1.48; no p value reported No signal of excess VTE in patients with cancer.
Vascular surgery subgroup: red-cell transfusion 68/286 (23.8%) 53/296 (17.9%) RR 1.28 95% CI 0.83 to 1.95; no p value reported Imprecise subgroup estimate; vascular surgery represented only 582/8273 (7.0%) procedures.
Vascular surgery subgroup: VTE within 90 days <6/284 <6/286 RR 1.83 95% CI 0.50 to 10.75; no p value reported Very sparse events; no reliable vascular-surgery safety inference.
  • TRACTION met both coprimary requirements: tranexamic acid reduced transfusion and was noninferior to placebo for 90-day VTE.
  • The absolute transfusion reduction was modest at the individual level, but policy use translated into fewer red-cell units per 100 patients and is potentially important at hospital-system scale.
  • The most practice-relevant reassurance is that VTE was 2.1% in both groups in the per-protocol analysis, including no signal of excess VTE in the large oncologic subgroup.

Internal Validity

  • Randomisation and Allocation:
    • Randomisation occurred at the hospital-cluster level at 4-week intervals.
    • A central, secure, web-based randomisation system generated the allocation scheme.
    • The cluster-crossover structure reduced confounding by stable hospital-level practice, because each hospital contributed both tranexamic acid and placebo periods.
    • Allocation concealment was strong: clinical teams and investigators were unaware of interval assignments, with research pharmacy handling preparation.
    • The final linked intention-to-treat groups were closely balanced: 4156 tranexamic acid and 4117 placebo.
  • Dropout or Exclusions:
    • 9927 patients undergoing major noncardiac surgery were assessed.
    • 1506 were not randomised: 66 had active venous thromboembolism, 339 had perceived transfusion risk <5%, 834 had anaesthesiologist, surgeon, or both decline participation, 263 had process barriers, and 4 declined participation.
    • 8421 eligible patients underwent randomisation.
    • Data linkage failed in 148/8421 (1.8%): 66 assigned to tranexamic acid and 82 assigned to placebo, owing to out-of-province healthcare numbers.
    • The primary intention-to-treat population therefore included 8273 linked patients.
    • The per-protocol population included 8180 patients: 4128 tranexamic acid and 4052 placebo.
    • There were no missing primary or secondary outcomes among successfully linked participants.
    • The post-randomisation linkage exclusion was small and reasonably balanced, so it is unlikely to materially undermine internal validity.
  • Performance/Detection Bias:
    • Double-blinding substantially reduced performance bias, particularly for transfusion decision-making.
    • Matching placebo and blinded emergency investigational product avoided the need for routine emergency unblinding.
    • The primary effectiveness outcome was pragmatic and objective in the sense that issued red cells were captured electronically, although the clinical threshold for transfusion remained clinician-dependent.
    • The primary safety outcome was electronically ascertained rather than based on systematic screening imaging.
    • Validated VTE administrative ascertainment has imperfect sensitivity: DVT sensitivity 75.24% and PE sensitivity 74.83%, with specificity 95.77% and 91.86%, respectively.4
  • Protocol Adherence:
    • Study-drug non-receipt was uncommon: 28/4222 assigned to tranexamic acid did not receive tranexamic acid and 20/4199 assigned to placebo did not receive placebo.
    • Open-label tranexamic acid use occurred in 124/8421 patients (1.5%).
    • Blinded emergency investigational product was used in 334/8421 patients (4.0%).
    • Topical tranexamic acid was rare, with <6 patients in each group.
    • Cell-saver use was similar: 80 patients in the tranexamic acid group versus 84 in the placebo group.
    • Intraoperative DVT prophylaxis was also broadly similar: 2138 versus 2042 patients.
  • Baseline Characteristics:
    • Baseline characteristics were well balanced.
    • Median age was 64 years in both groups: IQR 54-72 with tranexamic acid and 54-73 with placebo.
    • Male sex was 1985/4156 (47.8%) versus 1908/4117 (46.3%).
    • Median weight was 80 kg in both groups.
    • Median Charlson Comorbidity Index was 2 in both groups.
    • Median preoperative haemoglobin was 13.4 g/dL in both groups among those with available data.
    • Oncologic surgery was common and balanced: 2536/4072 (62.3%) versus 2466/4027 (61.2%).
    • Elective surgery predominated: 3806/4093 (93.0%) versus 3753/4052 (92.6%).
  • Heterogeneity:
    • Clinical heterogeneity was intentional and central to the policy question.
    • Procedures included general surgery, gynaecology, urology, thoracic surgery, vascular surgery, spine surgery, and other specialties.
    • The most common specialties were general surgery 2742/8273 (33.1%), gynaecological surgery 1540/8273 (18.6%), urological surgery 1434/8273 (17.3%), vascular surgery 582/8273 (7.0%), and spine surgery 461/8273 (5.6%).
    • The cluster-crossover model accounted for hospital and period effects.
    • The intracluster correlation was 0.68 for transfusion and 0.03 for VTE; the cluster autocorrelation coefficients were -0.13 and 0.05, respectively.
    • Subgroup estimates were generally directionally consistent, although vascular surgery estimates were imprecise and not independently practice-defining.
  • Timing:
    • The first dose was given at the start of surgery, which is biologically appropriate for prevention of intraoperative fibrinolysis and bleeding.
    • The second dose was administered before skin closure, with timing left to the anaesthesiologist.
    • The trial did not report granular timing-fidelity data beyond receipt of study intervention.
    • Because the tested intervention was a hospital policy, some flexibility in timing improves pragmatic relevance.
  • Dose:
    • The regimen was clinically familiar: 1 g IV at surgery start, 2 g for patients >100 kg, plus 1 g before closure.
    • The dose was designed to be easy to operationalise and consistent with perioperative practice.
    • It was not a dose-finding trial and does not establish the minimum effective dose.
    • The rare seizure signal was clinically reassuring in this dosing context: one perioperative seizure occurred in the tranexamic acid group.
  • Separation of the Variable of Interest:
    • The intervention achieved measurable biological and clinical separation: red-cell transfusion was 7.4% versus 9.8%.
    • Red-cell units transfused at patient level were 24.7 ± 3.6 per 100 patients versus 34.2 ± 2.6 per 100 patients.
    • Red-cell units transfused at hospital level were 23.5 ± 27.2 per 100 patients versus 39.6 ± 80.6 per 100 patients.
    • VTE was not separated in a harmful direction: 86/4128 (2.1%) versus 85/4052 (2.1%) in the per-protocol analysis.
    • Low open-label tranexamic acid use and low topical tranexamic acid use preserved contrast between treatment periods.
  • Key Delivery Aspects:
    • The hospital-policy design directly tested implementation under routine operating-room conditions.
    • Transfusion thresholds were not protocolised, which improves pragmatism but permits inter-clinician and inter-hospital variation.
    • Use of linked data enabled efficient near-complete follow-up.
    • Hospitals needed research pharmacy and data-linkage infrastructure, which is both a methodological strength and an implementation constraint.
  • Crossover:
    • Crossover was relevant but uncommon.
    • Open-label tranexamic acid use occurred in 59 patients assigned to tranexamic acid periods and 65 assigned to placebo periods.
    • The per-protocol safety analysis excluded placebo-period patients who received open-label tranexamic acid, preserving the primary noninferiority contrast.
    • Any residual crossover would be expected to dilute, rather than exaggerate, the treatment effect.
  • Adjunctive Therapy Use:
    • Cointerventions were documented rather than protocolised.
    • Cell salvage was used in 80 versus 84 patients.
    • DVT prophylaxis was used in 2138 versus 2042 patients.
    • Topical tranexamic acid was used in fewer than six patients in each group.
    • There was no evidence that the placebo group systematically received more adjunctive haemostatic therapy that could mask a tranexamic acid effect.
  • Outcome Assessment:
    • Red-cell transfusion was objectively captured during the index hospitalisation.
    • VTE ascertainment used provincial administrative sources and imaging-related codes or reports.
    • In Manitoba, VTE ascertainment was supplemented by manual imaging-report review by two independent adjudicators.
    • MI, stroke, DVT, and PE during hospitalisation were too infrequent for reliable comparative modelling.
    • Subclinical VTE could have been missed because routine screening imaging was not performed.
  • Statistical Rigor:
    • The statistical analysis plan was finalised before investigators were aware of trial results.
    • The analytic approach matched the cluster-crossover design.
    • The prespecified primary effectiveness and safety questions were both answered in the planned populations.
    • The covariate-adjusted sensitivity analysis for transfusion was directionally consistent: 231/2800 (8.3%) versus 279/2769 (10.1%); RR 0.79; 95% CI 0.66 to 0.93.
    • The covariate-adjusted sensitivity analysis was complete-case because preoperative haemoglobin was not available for all patients, so the unadjusted prespecified primary analysis remains the principal result.
    • Several secondary outcomes were not analysed because event numbers were too low.

Conclusion on Internal Validity: Internal validity is strong. The trial combined concealed cluster-crossover randomisation, double-blinding, matching placebo, near-complete electronic follow-up, low contamination, and prespecified analyses; the main caveats are clinician-dependent transfusion decisions, imperfect administrative VTE ascertainment, and limited power for rare arterial safety events.

External Validity

  • Population Representativeness:
    • The enrolled population was broad and clinically relevant for major noncardiac surgery with meaningful transfusion risk.
    • The median age was 64 years, consistent with many major noncardiac surgical cohorts.
    • Sex distribution was balanced for a broad surgical population: 47.1% male overall.
    • Oncologic procedures were strongly represented: 5002/8273 (60.5%).
    • Patients undergoing cardiac surgery and hip or knee total arthroplasty were excluded because tranexamic acid is already routine in those settings.
    • Patients with active thromboembolic disease, pregnancy, free-flap surgery, and recent trauma tranexamic acid exposure were excluded.
    • Race, ethnicity, and gender identity were not systematically collected, limiting assessment of representativeness across sociodemographic groups.
  • Applicability:
    • Findings apply best to adults undergoing inpatient noncardiac surgery with estimated transfusion risk ≥5% in health systems capable of perioperative policy implementation.
    • The intervention is inexpensive and technically simple, so clinical applicability is broader than the Canadian trial setting.
    • The trial’s data-linkage infrastructure is not required to administer tranexamic acid clinically, but it was important for efficient trial conduct.
    • Generalisability to low-risk minor procedures is indirect because TRACTION enriched for operations with ≥5% transfusion risk.
    • Generalisability to vascular surgery is less certain because vascular procedures were only 7.0% of the trial population and subgroup estimates were imprecise.
    • Generalisability to patients with active thrombosis, pregnancy, free-flap reconstruction, and trauma patients already receiving tranexamic acid is limited.
    • Resource-limited settings may benefit greatly from transfusion-sparing treatment, but TRACTION was conducted entirely in Canadian hospitals with pharmacy and data systems capable of supporting a blinded policy trial.

Conclusion on External Validity: External validity is strong for adult inpatient major noncardiac surgery with meaningful transfusion risk in high-resource perioperative systems. It is more limited for low-risk surgery, vascular-surgery-specific practice, pregnancy, free-flap surgery, active thromboembolic disease, and systems without reliable perioperative delivery workflows.

Strengths & Limitations

  • Strengths:
    • Large sample size: 8421 randomised and 8273 in the primary linked intention-to-treat population.
    • Double-blind, placebo-controlled design despite policy-level cluster randomisation.
    • Cluster-crossover structure tested hospital implementation while controlling for stable hospital-level differences.
    • Prespecified coprimary effectiveness and safety framework.
    • Near-complete follow-up through linked electronic health data.
    • High pragmatic relevance because transfusion decisions and cointerventions were embedded in routine care.
    • Large oncologic surgery subgroup, addressing a population of high clinical interest for thrombosis safety.
    • Patient and caregiver involvement in trial design, consent processes, and outcome prioritisation.
    • Low contamination with open-label or topical tranexamic acid.
  • Limitations:
    • Conducted only in Canada.
    • Participation required hospitals capable of policy implementation and data linkage.
    • 148 randomised patients could not be linked because of out-of-province health numbers.
    • 834 otherwise assessed cases were not randomised because the anaesthesiologist, surgeon, or both declined participation, which may affect representativeness.
    • Transfusion decisions were not protocolised.
    • VTE ascertainment relied on administrative and imaging-code strategies rather than systematic screening.
    • MI, stroke, in-hospital DVT, and in-hospital PE were too infrequent for reliable between-group modelling.
    • Race, ethnicity, and gender identity were not systematically available.
    • Limited inference for vascular surgery, pregnancy, active thromboembolic disease, free-flap reconstruction, and low-risk operations.

Interpretation & Why It Matters

  • Clinical practice
    • For adults undergoing major noncardiac surgery with estimated transfusion risk ≥5%, perioperative tranexamic acid should be considered a default hospital policy unless a specific contraindication exists.
    • The transfusion reduction was clinically modest for an individual patient but operationally important across large surgical volumes.
    • The 90-day VTE result directly addresses the commonest safety objection to wider tranexamic acid use.
    • The oncologic subgroup is particularly practice-informing because VTE risk concerns often limit antifibrinolytic use in cancer surgery.
  • Assumptions tested
    • TRACTION confirms that the haemostatic benefit of tranexamic acid persists when delivered as a hospital-wide policy rather than as an individual-patient explanatory intervention.
    • It challenges the assumption that broad use in major noncardiac surgery necessarily carries an important venous thrombotic penalty.
    • It extends POISE-3 by focusing on high-transfusion-risk surgery, longer 90-day VTE ascertainment, and policy implementation.
  • Methodological importance
    • TRACTION is an exemplar of pragmatic perioperative policy evaluation.
    • It demonstrates that cluster-crossover randomisation, blinding, and electronic outcome ascertainment can be combined for large, efficient, practice-shaping trials.
    • The trial is important not merely because tranexamic acid works, but because it tested whether a routine hospital policy is safe enough to implement across a broad surgical case mix.

Controversies & Subsequent Evidence

  • Implementation rather than efficacy is now the dominant problem:
    • The accompanying New England Journal of Medicine editorial argued that millions of surgical patients remain exposed to unnecessary transfusion risk because tranexamic acid is not consistently implemented, despite accumulated evidence for efficacy, safety, and cost-effectiveness.5
    • The same editorial highlighted TRACTION as an exemplar of institutional policy randomisation integrated with electronic health data.5
    • The editorial’s practical implication is that evidence generation is no longer the only bottleneck; behavioural, anaesthetic, surgical, pharmacy, and governance implementation strategies are now central.
  • TRACTION complements, rather than replaces, POISE-3:
    • POISE-3 enrolled 9535 adults undergoing noncardiac surgery and showed reduced major bleeding with tranexamic acid, but did not establish noninferiority for its cardiovascular safety composite at 30 days.3
    • TRACTION asked a different question: whether a hospital policy in high-transfusion-risk noncardiac surgery reduces red-cell transfusion and is noninferior for 90-day VTE.
    • TRACTION did not use the POISE-3 major bleeding composite and was not powered for rare arterial safety outcomes such as myocardial infarction or stroke.
  • Thrombosis concern is much reduced, but not abolished for every subgroup:
    • TRACTION’s VTE result was reassuring: 86/4128 (2.1%) versus 85/4052 (2.1%) in the per-protocol population.
    • Administrative VTE ascertainment may miss subclinical or miscoded events, but this limitation should be nondifferential because allocation was blinded.
    • Active thromboembolic disease was excluded, so the result should not be extrapolated to patients with very recent or active arterial or venous thrombosis.
    • Vascular surgery remains less certain because the subgroup was small and the VTE estimate was extremely imprecise: RR 1.83; 95% CI 0.50 to 10.75.
  • Cancer surgery is a major contribution:
    • The trial included 5002 patients undergoing oncologic surgery, a population often excluded or under-represented in earlier tranexamic acid trials.
    • In oncologic surgery, transfusion was reduced from 241/2466 (9.8%) to 161/2536 (6.3%), RR 0.63; 95% CI 0.49 to 0.77.
    • Oncologic surgery VTE was not increased: 60/2526 (2.4%) versus 64/2421 (2.6%), RR 0.92; 95% CI 0.68 to 1.48.
    • This is one of TRACTION’s most clinically important extensions beyond prior evidence.
  • Evidence before TRACTION was broad but fragmented:
    • CRASH-2 showed that tranexamic acid reduced death due to bleeding in trauma without increasing vascular occlusive events.6
    • The WOMAN trial showed that early tranexamic acid reduced death due to bleeding in post-partum haemorrhage.7
    • ATACAS showed that tranexamic acid reduced bleeding-related outcomes in coronary-artery surgery.8
    • BART remains historically important because it compared aprotinin and lysine analogues in high-risk cardiac surgery and helped shape the modern antifibrinolytic evidence base.9
  • Meta-analytic safety evidence is consistent but statistically difficult:
    • A 2023 systematic review, meta-analysis, and trial sequential analysis of prophylactic IV tranexamic acid in noncardiac surgery found no increased thromboembolic risk, but trial sequential analysis had reached only 65% of the required information size to confirm safety definitively.10
    • A broader 2021 JAMA Surgery systematic review and meta-regression found no association between intravenous tranexamic acid and thromboembolic events or non-bleeding mortality across diverse bleeding and surgical settings.11
    • A 2021 critical care systematic review found no clear thrombotic signal but raised dose-related concern about seizures in bleeding patients, especially at higher-dose exposure; TRACTION’s regimen was comparatively modest, and seizure was rare.12
  • Guidelines are moving beyond the TRACTION population:
    • NICE NG24, updated in 2026 before TRACTION publication, recommends offering tranexamic acid to adults having surgery in an operating theatre if there is any risk of bleeding and the procedure breaches the skin or mucous membranes.13
    • This recommendation is broader than TRACTION, which studied major inpatient noncardiac surgery with estimated transfusion risk ≥5%.
    • TRACTION strongly supports the high-transfusion-risk component of that direction of travel, but it does not by itself prove benefit for very low-risk operations.
  • Why practice variation persists:
    • A qualitative interview study found that tranexamic acid use varies because of differences in awareness, perceived thrombosis risk, specialty norms, workflow, decision ownership, and feedback systems.14
    • TRACTION reduces uncertainty about policy safety in high-transfusion-risk surgery, but implementation still requires clinical decision support, agreed local policy, anaesthetic and surgical engagement, and audit-feedback systems.
  • Further reading and related evidence:
    • Ker et al. provide an older but pivotal cumulative meta-analysis showing that tranexamic acid reduces surgical bleeding and transfusion across surgical trials.15
    • The 2020 noncardiac surgery systematic review framed the residual uncertainty that TRACTION was designed to resolve.2
    • Park et al. analysed POISE-3 general surgery patients and found that tranexamic acid reduced perioperative bleeding without increasing cardiovascular risk, providing procedure-domain evidence that is concordant with TRACTION’s direction.16
    • The TRACTION protocol is essential reading for the cluster-crossover, risk-adapted consent, and registry-based design features.1

Summary

  • TRACTION randomised 8421 adults across 10 Canadian hospitals to 4-week hospital-policy periods of intraoperative tranexamic acid or placebo.
  • Among 8273 linked participants, red-cell transfusion during the index hospitalisation was lower with tranexamic acid: 306/4156 (7.4%) versus 403/4117 (9.8%); RR 0.73; 95% CI 0.61 to 0.86.
  • VTE within 90 days was 86/4128 (2.1%) versus 85/4052 (2.1%) in the per-protocol population; RR 0.96; 95% CI 0.65 to 1.38, meeting the prespecified noninferiority criterion.
  • The large oncologic subgroup showed both transfusion reduction and no VTE excess.
  • The trial supports routine perioperative tranexamic acid policy for adults undergoing major noncardiac surgery at meaningful transfusion risk.

Overall Takeaway

TRACTION is practice-shaping because it answers the hospital-policy question that previous tranexamic acid trials did not fully resolve. In major noncardiac surgery at high risk of transfusion, routine intraoperative tranexamic acid reduced red-cell transfusion and did not increase 90-day venous thromboembolism.

Overall Summary

  • Tranexamic acid reduced red-cell transfusion in major noncardiac surgery.
  • 90-day VTE was 2.1% in both treatment groups in the primary safety population.
  • The result is especially reassuring for oncologic surgery, which comprised 60.5% of procedures.
  • The strongest application is hospital policy for adult inpatient noncardiac surgery with ≥5% estimated transfusion risk.

Bibliography

Added June 20th, 2026