CRASH-3 is the latest massive, pragmatic multi-center RCT brought to us by the London School of Hygiene and Tropical Medicine Trials Unit. These investigators have created an extensive network of hospitals throughout the world with the capability of performing truly impressive trials. Especially in the critical care arena, these trials have uniquely high power which allows them to detect small differences in mortality, while evaluating for uncommon adverse events.
Pragmatic trials are designed to emulate the realities of clinical practice, thereby answering a clinical question in a direct fashion. This is unlike most critical care trials, which are explanatory (aiming to be as controlled as possible, thereby answering a mechanistic question with greatest precision; see table above from Coates 2011).1 To folks who are more accustomed to explanatory trials, pragmatic trials may seem a bit rough around the edges.
prelude: CRASH-2 traumatic brain injury sub-study
CRASH-2 was a large, pragmatic RCT investigating the use of tranexamic acid in trauma patients at risk of hemorrhage.2 The study showed a mortality benefit favoring the use of tranexamic acid (for more on this seminal study, see TheBottomLine & EMCrit).
Nested within CRASH-2 was a sub-study evaluating the use of tranexamic acid among patients with significant trauma plus traumatic brain injury.3 This study included 270 patients who met both the CRASH-2 inclusion criteria (trauma with concern for hemorrhage) plus the added requirements of a Glasgow Coma Scale score <15 and a brain CT scan compatible with traumatic brain injury.
Baseline properties of these patients are shown below. Most patients had mild or moderate traumatic brain injury, with an average Glasgow Coma Score of 10.
Patients in the tranexamic acid group tended to have less hematoma expansion and fewer new intracranial hemorrhages. There was also a trend towards improved mortality:
CRASH-3 was designed to further investigate using tranexamic acid for patients with traumatic brain injury. This study utilized the following inclusion criteria:
- Enrollment within <3 hours of injury
- Either Glasgow Coma Scale <13 or intracranial hemorrhage on CT scan
- No major extracranial bleeding
This was a massive, pragmatic, double-blind RCT involving 175 hospitals in 29 countries, with a target enrollment of 10,000 patients. Patients were randomized to receive either saline or tranexamic acid (1 gram loading dose over 10 minutes followed by a second gram infused over the following 8 hours; this is the same regimen used in CRASH-2). The primary endpoint was head injury-related death in the hospital within 28 days of injury.
The study was undertaken from 2012-2019. The protocol was amended slightly in 2016 based on the emergence of new knowledge (blinded to data from within the trial). The main change was narrowing the time window for recruitment from <8 hours post injury to <3 hours. Recruitment was extended, with a goal of recruiting ~10,000 patients meeting these more stringent inclusion criteria.
Groups were well matched at baseline, as shown below:
The primary endpoint is shown below. In addition to the entire study, a pre-planned analysis was performed excluding patients with a GCS of 3 or with bilateral unreactive pupils:
In both of these analyses, there is a very strong trend towards reduced mortality. However, these differences are technically not statistically significant (the p-values hover a wee bit over 0.05). Even when excluding patients with a GCS of 3 or bilateral fixed pupils, the p-value is 0.059 (using a Fisher Exact test).
Subgroup analyses show stark differences between patients of varying severity. Benefit from tranexamic acid is seen in subgroups of patients with less severe injury (either GCS 9-15 or patients with bilateral reactive pupils). Alternatively, among patients with more severe injury, tranexamic acid has no effect:
A similar distinction is evident when evaluating the relationship between time to treatment and the effect on outcomes (below). Among patients with severe injury, tranexamic acid made no difference (regardless of when it was given). Alternatively, among patients with mild/moderate injury, tranexamic acid exerted a time-dependent benefit:
many patients were too sick to benefit?
When comparing CRASH-3 to the CRASH-2 traumatic brain sub-study, the most notable difference is that patients in CRASH-3 were often sicker. The overall rate of head injury-related death was 12% in the CRASH-2 sub-study, versus 19% in CRASH-3.
It appears that a population of severely ill patients in CRASH-3 were simply beyond saving, regardless of intervention (e.g. patients with GCS 3 and bilateral fixed pupils). The presence of these patients diluted out beneficial effects from tranexamic acid.
Some evidence supporting this concept is that within the entire CRASH-3 study population, tranexamic acid did cause a statistically significant reduction in head injury-related death within the first 24 hours (OR 0.81, 95% confidence interval 0.69-0.95). This suggests efficacy in reducing hematoma expansion. However, many of these patients were so sick that they eventually went on to die within 28 days anyway. So, although tranexamic acid showed signs of efficacy within the entire study population, this efficacy often wasn’t sufficient to affect mortality endpoints. In short, tranexamic acid reduces brain injury, but it can’t raise the dead.
Many RCTs are underpowered to adequately evaluate uncommon adverse events. This is not a problem for CRASH-3. Below are the adverse events for all 12,639 patients enrolled in the study (including patients enrolled between 3-8 hours, who were subsequently removed from the final analysis).
The strength of this study is a well-powered look at these safety endpoints. There is no signal demonstrating any increase in adverse events (including vascular occlusive events, pulmonary emboli, or stroke).
There might be a tantalizing signal here showing a reduction in gastrointestinal bleeding among patients treated with tranexamic acid (relative risk 0.68, confidence interval 0.4-1.14). This may foreshadow the results of an upcoming RCT evaluating the use of tranexamic acid in gastrointestinal bleeding (the HALT-IT trial).4
Tranexamic acid reduced mortality from traumatic brain injury. The next question, then, is whether it increased the number of patients who have good clinical outcomes (or whether it merely increases the number of patients who survive with poor neurologic outcomes). There wasn't any signal that tranexamic acid increases the number of patients living with severe disability (table above). This would indirectly imply that tranexamic acid is increasing the number of patients who survive with good neurologic outcome.
The study includes a small meta-analysis to illustrate the results of CRASH-3 within the larger context of other trials. There appears to be a consistent signal here across multiple studies indicating benefit from tranexamic acid:
methodological crux: changing practice based on subgroup analyses?
CRASH-3 is technically a “negative” study, as the primary endpoint falls slightly short of reaching statistical significance (p>0.05). This likely resulted from the inclusion of some patients who were profoundly injured and unlikely to survive regardless of any therapy. Such patients may have generated statistical noise, which diminished the signal of benefit from tranexamic acid.
Subgroup analysis shows benefit from tranexamic acid among patients with a greater hope of recovery. Specifically, tranexamic acid reduced head injury-related death in the subgroup of patients with GCS>8 and also the subgroup of patients with reactive pupils. This makes a lot of sense. As shown below, the maximum benefit from an intervention may often occur among patients with intermediate disease severity:
Technically, since the primary endpoint is negative, subgroups should be used only for hypothesis generation. However, I feel that the subgroups in this study may be used to drive clinical management, for several reasons:
- Increased benefit among the subgroup of moderately injured patients is predictable and easily understood. In some ways, subgroup analysis could be viewed as correcting a design flaw inherent in the original study (inclusion of moribund patients).
- There are numerous signals throughout the study that tranexamic acid is more beneficial in patients with less severe injury (thus, this doesn’t represent a single subgroup, but rather a consistent finding across multiple subgroup analyses).
- The subgroups are huge (the smallest subgroup contains 2,769 patients). We’re not talking about a small handful of patients – each of these subgroups contains far more patients than would typically be included within an entire critical care RCT.
- Data supporting the safety of tranexamic acid is robust (across all patient subgroups).
- The use of tranexamic acid in traumatic brain injury is congruent with a broader body of literature supporting the use of tranexamic acid in trauma.2 Likewise, the safety of tranexamic acid is consistent with external literature.
- As sentient beings, we are allowed to exercise judgement (rather than being obligated to mechanically follow elementary statistical algorithms). Ultimately, the decision to use tranexamic acid should be made on the basis of integrating all available evidence across multiple trials (rather than a single p-value for one statistical comparison).
[TICH-2 + CRASH-3] = [TXA for ICH]?
Based on new insights from CRASH-3, it may be worth re-visiting the TICH-2 trial investigating spontaneous intracranial hemorrhage.5 TICH-2 randomized patients up to 8 hours after spontaneous intracranial hemorrhage to receive placebo versus tranexamic acid. The primary endpoint was neutral (good functional outcome at 90 days). However, many secondary endpoints were positive (including statistically significant improvements in hematoma expansion and in mortality after a week). Reasons that TICH-2 were negative could include delayed use of tranexamic acid (up to 8 hours after onset) or simply lack of adequate power to detect small clinical improvements.
CRASH-3 doesn’t directly study spontaneous intracranial hemorrhage, but may indirectly shed some light on that disease process. First, CRASH-3 suggests that it’s safe to use tranexamic acid in patients with cerebral edema and/or intracranial hemorrhage (allaying some concerns regarding the risk of stroke or seizure). Second, CRASH-3 (along with the sub-study of intracranial hemorrhage in CRASH-23) suggests that tranexamic acid may offer benefit for acute intracranial hemorrhage.
A pre-planned sub-study of CRASH-3 is still pending, which will evaluate the effect of tranexamic acid on hematoma expansion. If that sub-study shows a beneficial effect from tranexamic acid, this could further support the use of tranexamic acid for spontaneous intracranial hemorrhage.
- The conclusion of this article sums things up nicely: “tranexamic acid is safe in patients with TBI and treatment within 3 hours of injury reduces head injury-related death. Patients should be treated as soon as possible after injury.”
- The greatest strength of this study might be an extremely thorough evaluation for possible adverse events among 12,639 patients. Tranexamic acid was found to be safe, without increased rates of any adverse events (including thrombosis, seizure, and stroke).
- The primary endpoint of this study was technically negative (p-value slightly above 0.05). This likely reflects the inclusion of moribund patients, who diluted out the signal of benefit from tranexamic acid. Numerous subgroup analyses indicate that among patients with a greater hope of recovery, tranexamic acid is beneficial (figure below). As a statistical rebel, I would consider this trial to be positive, despite having a technically negative primary endpoint.
- EM Nerd (Rory Spiegel)
- St Emlyn's (Simon Carley)
- Rebel EM (Salim Rezaie)
- The Resus Room (including interview with author Ian Roberts)
- EM Lit of Note (Ryan Radecki)
- SGEM (Ken Milne, Salim Rezaie)
- BadEM (Craig Wyile)
- Tranexamic acid, CRASH-2, and pragmatism with Tim Coates (EMCrit, Scott Weingart)
- How CRASH2 got it right (Scott Weingart on the Maryland CC project)
- The Bottom Line (Adrian Wong)
- Its time for TXA in massive hemorrhage (REBEL EM, Salim Rezaie)
- CRASH-2: Tranexamic acid in major trauma (Core EM, Anand Swaminathan)
- Wiki Journal Club
The Tranexamic Acid Denier's Handbook (Karim Brohi at SMACC)
Great news! Most of these strategies will work nicely for CRASH-3!
- 1.Coats T. Future research in emergency medicine: explanation or pragmatism? Large or small? Simple or complex? Emerg Med J. 2011;28(12):1004-1007. https://www.ncbi.nlm.nih.gov/pubmed/21785146.
- 2.Roberts I, Shakur H, Coats T, et al. The CRASH-2 trial: a randomised controlled trial and economic evaluation of the effects of tranexamic acid on death, vascular occlusive events and transfusion requirement in bleeding trauma patients. Health Technol Assess. 2013;17(10):1-79. https://www.ncbi.nlm.nih.gov/pubmed/23477634.
- 3.Perel P, Al-Shahi S, Kawahara T, et al. CRASH-2 (Clinical Randomisation of an Antifibrinolytic in Significant Haemorrhage) intracranial bleeding study: the effect of tranexamic acid in traumatic brain injury–a nested randomised, placebo-controlled trial. Health Technol Assess. 2012;16(13):iii-xii, 1-54. https://www.ncbi.nlm.nih.gov/pubmed/22417901.
- 4.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. https://www.ncbi.nlm.nih.gov/pubmed/25409738.
- 5.Sprigg N, Flaherty K, Appleton J, et al. Tranexamic acid for hyperacute primary IntraCerebral Haemorrhage (TICH-2): an international randomised, placebo-controlled, phase 3 superiority trial. Lancet. 2018;391(10135):2107-2115. https://www.ncbi.nlm.nih.gov/pubmed/29778325.