TRACTION: A tranexamic acid default cuts transfusion in major noncardiac surgery

Tranexamic acid (TXA) is routine in cardiac surgery and hip and knee arthroplasty, yet its use in other major noncardiac operations still varies by surgeon, anesthesia clinician, and hospital. TRACTION asked a different question from most earlier TXA trials. It did not test whether TXA works in one selected operation. It tested whether a hospital policy of giving TXA to every eligible high-transfusion-risk patient reduces red-cell transfusion without increasing venous thromboembolism (VTE). For both coprimary outcomes, the answer was yes. Reading it well means keeping three things apart: a clear but modest per-patient transfusion benefit, a reassuring VTE result that was designed to exclude a specific margin of harm, and arterial safety events that were too rare to compare.

What the trial asked and why it matters

The case for perioperative TXA in noncardiac surgery was already substantial but fragmented. A 2020 systematic review of 69 trials (6,157 patients) in noncardiac, non-arthroplasty surgery at high transfusion risk found that TXA reduced the proportion of patients transfused (relative risk [RR] 0.59; 95% CI, 0.48–0.72), but the certainty of evidence was low and thrombotic outcomes were limited by short follow-up and no systematic screening. Houston et al., 2020. POISE-3 randomized 9,535 patients having noncardiac surgery to TXA 1 g at the start and end of surgery or placebo. The composite bleeding outcome fell from 11.7% to 9.1% (hazard ratio [HR] 0.76; 95% CI, 0.67–0.87), but the composite cardiovascular safety outcome (14.2% vs 13.9%; HR 1.02; 95% CI, 0.92–1.14) narrowly missed its noninferiority margin (one-sided P = .04). POISE-3. A 2023 meta-analysis of 191 trials (40,621 patients) found no increase in thromboembolic events (RR 1.02; 95% CI, 0.94–1.11), but trial sequential analysis showed only 64.6% of the required information size had been reached. Tsan et al., 2023.

That leftover thrombosis uncertainty is a large part of why TXA use is inconsistent. In the accompanying editorial, Murphy and Roberts argue that the gap reflects "poor implementation into clinical practice and lingering concerns" about whether thrombosis risk had been excluded with enough confidence, rather than missing evidence of efficacy. NEJM editorial. TRACTION was built to address both: whether a policy works when delivered as routine care, and whether it is noninferior for VTE out to 90 days.

How the study was designed

TRACTION was a multicenter, double-blind, placebo-controlled, cluster-randomized crossover trial at 10 Canadian hospitals. Each hospital was randomly assigned at 4-week intervals to a hospital-wide policy of intraoperative TXA or matching placebo (0.9% saline), so every site contributed both TXA and placebo periods. Enrollment ran from February 2022 to March 2024, over 25 consecutive 4-week periods. Research pharmacy prepared the study drug and was separated from clinical and research teams. Patients, clinicians, investigators, and the trial statistician were blinded. A blinded "emergency" product containing the opposite of the current allocation was available for life-threatening hemorrhage, so clinicians could ensure TXA exposure without unblinding. Original report; Critical Care Reviews summary; ClinicalTrials.gov NCT04803747.

Patients. Adults having inpatient major noncardiac surgery with an estimated red-cell transfusion risk of at least 5% were eligible, including open operations and laparoscopic operations expected to last at least 3 hours. Hospitals had to perform at least 100 noncardiac operations a month, and their anesthesia, surgical, and hospital leadership had to agree to the policy. Exclusions included active thromboembolic disease, cardiac surgery, total hip or knee arthroplasty (where TXA is already routine), free-flap surgery, trauma surgery with TXA given in the preceding 3 hours, and pregnancy. Because TXA use for these patients already fell within usual-care variation, sites enrolled under altered or waived consent as directed by their ethics boards. Published protocol; Critical Care Reviews summary.

Regimen. TXA 1 g IV (2 g if weight exceeded 100 kg) within 10 minutes of the first incision, followed by 1 g IV at 2–4 hours or before skin closure. The anesthesia clinician could give the second gram as a bolus or as a continuous infusion through the procedure. Transfusion thresholds, cell salvage, topical TXA, and thromboprophylaxis were left to the treating teams and documented but not protocolized. ClinicalTrials.gov NCT04803747.

Outcomes and analysis. The coprimary effectiveness outcome was red-cell transfusion during the index hospitalization. The coprimary safety outcome was VTE (deep-vein thrombosis or pulmonary embolism) diagnosed within 90 days, assessed for noninferiority: the upper bound of the 95% CI for the relative risk had to fall below 1.46. That margin corresponds to about a 1-percentage-point absolute increase from an assumed 2.2% baseline. Outcomes came from linked hospital, transfusion, and provincial administrative data, including a validated combination of physician billing and imaging codes for VTE. Mixed-effects models accounted for the cluster-crossover design. Both coprimary outcomes had to be met to call the policy beneficial. The effectiveness analysis used the intention-to-treat population, and the safety analysis used the per-protocol population. Original report; Published protocol.

The trial was planned for about 8,320 patients, assuming a baseline transfusion rate of 18% and powered to detect a 6-percentage-point absolute reduction. Published protocol. The observed placebo-period rate was roughly half that assumption, which matters for interpreting the absolute effect.

Results

Of 9,927 patients assessed, 8,421 were randomized (4,222 in TXA periods, 4,199 in placebo periods). Data linkage failed for 148 (1.8%), mostly because of out-of-province health numbers, leaving 8,273 in the primary analysis. Median age was 64 years, and 93% of operations were elective. Oncologic surgery made up 60.5% of procedures (5,002 of 8,273). The most common specialties were general surgery (33.1%), gynecology (18.6%), urology (17.3%), vascular surgery (7.0%), and spine surgery (5.6%). Original report; ASH 2025 abstract; Critical Care Reviews summary.

Sources: Original report, Abstract (first two rows); Critical Care Reviews summary (ITT VTE row).

The upper bound of the per-protocol VTE interval (1.38) sat below the 1.46 margin, so the safety criterion was met. The transfusion effect works out to a number needed to treat of about 37 to prevent one patient from being transfused. That is modest for any one patient, but meaningful across a hospital's major-surgery volume. Original report.

Secondary and subgroup results reported in the trial are summarized below. Several arterial and in-hospital thrombotic outcomes were not formally compared because events were too few. Critical Care Reviews summary.

Adherence was high. Of patients assigned to TXA periods, 28 of 4,222 did not receive TXA, and 20 of 4,199 in placebo periods did not receive placebo. Open-label TXA was given to 124 of 8,421 patients (1.5%), and the blinded emergency product to 334 (4.0%). Cell salvage (80 vs 84 patients) and topical TXA (fewer than 6 per arm) were similar between arms. Serious adverse events of special interest were rare: one perioperative seizure in the TXA arm and one perioperative anaphylaxis in the placebo arm. A covariate-adjusted, complete-case sensitivity analysis pointed in the same direction (8.3% vs 10.1%; RR 0.79; 95% CI, 0.66–0.93). Critical Care Reviews summary.

Strengths and limitations

TRACTION's strengths are its size, double-blind placebo control within a policy-level design, a crossover structure in which each hospital served as its own control, high adherence, minimal contamination, and near-complete follow-up through linked data. It also enrolled 5,002 patients having cancer surgery, a group in whom thrombosis concern most often limits TXA use. The limitations matter for applying it:

• VTE was found, not screened for. VTE came from administrative and imaging-code data rather than protocolized ultrasound. Validation work cited by reviewers puts the sensitivity of this approach at roughly 75%. Because allocation was blinded, missed events should be nondifferential, but subclinical clots were not counted. Critical Care Reviews summary.
• Arterial safety is not settled. TRACTION was not powered for myocardial infarction or stroke, and it did not use POISE-3's cardiovascular composite, which included myocardial injury after noncardiac surgery. The arterial question POISE-3 left open is not answered here.
• Selected hospitals and clinicians. Hospitals had to agree to the policy, and 834 otherwise eligible patients were not randomized because the anesthesiologist, surgeon, or both declined. Practice in less engaged settings may differ.
• Lower-than-planned baseline risk. Placebo-period transfusion was 9.8%, against the 18% assumed in planning. The absolute benefit (about 2.7 points) will be smaller in lower-risk operations and could be larger in higher-risk ones.
• Excluded groups. Patients with active thromboembolic disease, pregnant patients, free-flap reconstruction, recent trauma TXA, cardiac surgery, and arthroplasty were not studied. Race and ethnicity were not systematically collected.
• Vascular surgery is uncertain. The vascular subgroup (7% of patients) pointed in the opposite direction for transfusion, with a wide interval, and had too few VTE events for any safety inference. Subgroups were not powered for separate conclusions.
• Transfusion was not protocolized. Thresholds were left to clinicians. Blinding protects the comparison, but absolute rates will shift with local transfusion practice.
• Not a dose-finding trial. TRACTION tested one practical regimen. It does not establish the minimum effective dose, and it does not compare a single dose with two.

An earlier presentation of the trial at the American Society of Hematology 2025 meeting reported the results as odds ratios with a slightly different margin. The published NEJM analysis, using relative risks and the 1.46 margin, is the reference version. ASH 2025 abstract.

Keeping a TXA default safe: dose, timing, and wrong-route errors

A policy shifts TXA from an individual decision to a default, so the work moves to the system: when the drug is drawn up, who gives it, and how it is stored. Three practical points follow from the trial and from current guidance.

• Match the trial regimen if you are adopting the trial's evidence. TRACTION gave 1 g IV within 10 minutes of incision (2 g above 100 kg) and a second gram at 2–4 hours or before closure, by bolus or infusion. ClinicalTrials.gov NCT04803747. POISE-3 used 1 g at the start and end of surgery. POISE-3. The NICE guideline, updated in February 2026, recommends a single 1 g dose given just before the start of surgery for adults. It advises weighing any extra doses against risks, including accumulation in renal impairment. NICE NG24, section 1.3. TRACTION does not tell you whether one dose would have been enough.
• The guideline population is broader than the trial population. NICE now recommends offering TXA to adults having surgery in an operating theatre whenever there is any risk of bleeding and the procedure breaches skin or mucous membranes. NICE NG24. TRACTION directly supports that direction for inpatient operations with at least 5% transfusion risk. It provides only indirect evidence for lower-risk or day-case surgery.
• Engineer out neuraxial mix-ups. The FDA has required a boxed warning and a neuraxial contraindication for TXA injection after cases in which TXA was given intrathecally or epidurally in place of a local anesthetic, with serious outcomes including death. Storage near local anesthetics and failure to verify the product contributed. The FDA recommends keeping TXA vials and ampules away from local anesthetics and out of spinal or epidural kits, using premixed IV bags when possible, barcode scanning, labeling every syringe, and adding TXA to high-alert medication lists. FDA safety update. NICE likewise calls for safety measures against accidental intrathecal or epidural administration. NICE NG24, 1.3.9. When a hospital moves to a TXA default for cases that are often done with a spinal or epidural, the timing and location of TXA preparation should be part of the policy.

With TRACTION's 2 to 3 g perioperative regimen, one perioperative seizure occurred in the TXA arm. Critical Care Reviews summary. That is reassuring at this exposure but says nothing about higher-dose regimens.

Bottom line for practice

For adults having inpatient major noncardiac surgery with an estimated transfusion risk of at least 5%, TRACTION supports intraoperative TXA as a hospital default. In 8,273 patients, it reduced red-cell transfusion from 9.8% to 7.4% (RR 0.73), and 90-day VTE was 2.1% in both arms, which met a prespecified noninferiority margin. That includes patients having cancer surgery. The per-patient benefit is modest, with about 37 patients treated to spare one a transfusion, so the gains come from applying it consistently at scale. TRACTION does not resolve arterial safety, the minimum effective dose, vascular surgery, or patients with active thrombosis. Any TXA default should be built with wrong-route safeguards from the start.