FormalPara Take-home message

In patients with acute TBI, TXA probably has no effect on mortality or disability. The use of TXA probably does not increase the risk of adverse events.

Introduction

Traumatic brain injury (TBI) is a leading cause of mortality and morbidity worldwide [1,2,3], with the vast majority of patients presenting with intracranial bleeding [4]. In particular, TBI is more common in low- and middle-income countries with youth and adolescents being disproportionately affected [5, 6]. Progressive hematoma expansion secondary to high levels of fibrinolysis and coagulopathy has been associated with worse prognosis and increased risk of intracranial hypertension, brain herniation, and death in patients with TBI [7]. Tranexamic acid (TXA) is an antifibrinolytic agent that reduces bleeding by inhibiting plasmin production and preventing fibrin degradation. Trauma guidelines have recommended the early administration of TXA in severely injured adult trauma patients with extracranial bleeding based on the results of the CRASH-2 trial which demonstrated survival benefit in bleeding trauma patients with no increase in adverse events [8,9,10]. Other large RCTs have further confirmed the safety of TXA administration in a number of heterogenous populations [8, 11, 12]; however, in a recently published RCT, TXA was associated with an increased rate of venous thromboembolic events in patients with gastrointestinal bleeding [13].

The role of TXA in patients with TBI or intracranial bleeding is controversial, with conflicting trial results [14]. Previously published meta-analyses examining the effect of TXA in TBI have suggested benefit of TXA in this population, but conclusions are limited by imprecision [15]. With the recent publication of the CRASH-3 trial [16], the largest examining this question, as well as the out-of-hospital TXA versus placebo trial [17], we conducted a systematic review and meta-analysis examining the efficacy and safety of TXA in acute TBI.

Methods

The protocol for this systematic review was registered on PROSPERO (CRD42020164232) April 28, 2020. We last updated our search on June 24, 2020 to ensure that there were no new trials that would meet the inclusion criteria of our systematic review and meta-analysis. We have submitted an update to PROSPERO which reflects this search update. Any deviations from the published protocol are highlighted with an accompanying explanation.

Systematic search

We conducted a comprehensive search of MEDLINE, PubMed EMBASE, CINHAL, American College of Physicians Journal Club (ACPJC), Google Scholar, and unpublished sources including WHO ICTRP, PROSPERO, Clinicaltrials.gov, and the Cochrane trial registry from inception until June 24, 2020 for RCTs investigating the role of TXA in adult patients with TBI. We did not apply language restrictions. We developed the search strategy with the assistance of an expert medical librarian and included three search terms: ‘Tranexamic acid’, ‘Traumatic Brain injury’ and ‘Randomized Controlled Trials’ (see supplementary appendix for search strategy, appendix 1–7). We used the Medical Subject Headings database for identification of synonyms. We examined the reference list of full-text articles for additional relevant studies. We also searched conference proceedings within the last 2 years for the Society of Critical Care Medicine (SCCM), the European Society of Intensive Care and Emergency Medicine (ESICM), the American Association for the Surgery of Trauma (AAST), and the Eastern Association for the Surgery of Trauma (EAST).

Study selection

We included RCTs if they examined patients with TBI who were randomized to intravenous TXA administration as compared to placebo or usual care. We included studies of adolescent (≥ 15 years of age) and adult patients with any type of intracranial hemorrhage secondary to TBI and who received TXA at any dose. We included studies which reported on the following outcomes: mortality, disability (as measured by the Glasgow Outcome Scale (GOS), the Glasgow Outcome Scale-Extended (GOS-E), or the Disability Rating Scale (DRS)), hematoma expansion on subsequent neuroimaging, need for neurosurgical intervention, hospital and intensive care unit (ICU) length of stay, and adverse events including pulmonary embolism (PE), deep vein thrombosis (DVT), stroke, and seizure. For outcomes reported at multiple timepoints, we used the longest reported follow-up timepoint.

After implementation of the search strategy, two reviewers screened all potentially relevant citations independently and in duplicate. Citations deemed potentially relevant by either screener were advanced to second-stage full-text review. Full texts were subsequently reviewed for eligibility, with disagreements resolved by consensus, and third-party adjudication if required. We captured reasons for exclusion at the full-text screening stage.

Data extraction and quality assessment

Reviewers extracted data independently and in duplicate using pre-piloted data abstraction forms. We extracted the following information from included studies: study title, first author, demographic data, details of the intervention, and control, outcome data, and risk of bias (RoB) for each study. We contacted study authors for clarification when the population characteristics, method of follow-up, or outcome data were unclear or not reported. In particular, we acquired all-cause mortality data from the CRASH-3 authors. We assessed RoB independently and in duplicate using a modified Cochrane RoB tool [18] for which each domain is rated as “low”, “probably low”, “high”, or “probably high”. We examined the following RoB domains: sequence generation, allocation sequence concealment, blinding, selective outcome reporting, and other bias (such as stopping early and funding source). We rated the overall RoB for an individual study as the highest risk attributed to any domain.

We assessed the overall certainty of evidence for each outcome using the Grading Recommendations Assessment, Development and Evaluation (GRADE) approach [19]. We resolved disagreements for RoB or GRADE assessment by consensus. We used the Guideline Development Tool (https://www.gradepro.org) to formulate the Summary of Findings table.

Statistical analysis

We used DerSimonian and Laird random-effects models to conduct the meta-analysis [20] with RevMan 5.3 (Cochrane Collaboration, Oxford) software. We generated study weights using the inverse variance method. We present results as relative risks (RRs) and risk difference (RD) for dichotomous outcomes and as mean differences (MDs) for continuous outcomes, all with 95% confidence intervals (CIs). We calculated absolute effects using the pooled baseline prevalence from the control arm of included trials.

We assessed heterogeneity between trials using visual inspection of the forest plots, the Chi-squared test for homogeneity (where p < 0.1 indicates important heterogeneity), and the I2 statistic (for which a value of 50% or greater was considered reflective of potentially important heterogeneity) [21]. Although planned, we did not construct funnel plots to assess for publication bias as these are inaccurate when less than ten trials are included in the analysis [22]. We performed a predefined subgroup analysis comparing studies at high RoB compared to those at low RoB. We also performed two post hoc sensitivity analyses, one excluding the results of the largest trial (CRASH-3) [16] and another excluding studies enrolling adolescents [17, 23,24,25]. We performed this sensitivity analysis excluding the results of CRASH-3 as it was the largest trial and because they changed their primary outcome midway through the trial; of note, this was explained by the authors in their statistical plan as an effort to reduce the dilution of effect from non-head-injury-related deaths [26]. We also performed a post hoc subgroup analysis as requested by peer reviewers examining mortality in high-income versus low-to-middle-income countries as defined by the World Bank Classification.

We conducted trial sequential analysis (TSA) [27] using a random-effects model for mortality. For the TSA, we used a statistical significance level of 5%, a power of 80%, and a relative risk reduction of 10%. We used a model variance-based heterogeneity correction and did this analysis using Trial Sequential Analysis v.0.9.5.10 beta software (Copenhagen Trial Unit, Centre for Clinical Intervention Research, Rigshospitalet, Copenhagen, Denmark, https://www.ctu.dk/tsa).

Results

Of the 672 citations identified in the search (see Fig. 1), we excluded 200 duplicates and a further 446 citations after title and abstract screening. We assessed 26 full texts and included 9 RCTs in the review [16, 17, 23,24,25, 28,29,30,31]. There were 14,747 patients included in this study. One trial was initially not published in a peer-reviewed journal [17]; however, we extracted the data from ClinicalTrials.gov and then subsequently updated this data upon its publication [17]. Baseline characteristics of included trials are summarized in Table 1.

Fig. 1
figure 1

Study flowchart

Full size image
Table 1 Baseline characteristics of included studies
Full size table

Description of included studies

Three RCTs were multicenter [16, 17, 30], while six were conducted at a single site [23,24,25, 28, 29, 31]. The mean age of participants ranged from 35 to 55 years. All trials included adults; however, four trials also included adolescents [17, 23,24,25]. One trial excluded severe TBI, defined as Glasgow Coma Scale (GCS) < 8 at presentation [25], while two trials excluded mild TBI (GCS > 12) [17, 24]. The other trials enrolled patients with any TBI severity; however, the majority of included patients had moderate-to-severe TBI (see Table 1 for more details regarding TBI severity from included trials). Although all included trials focused mainly on patients with TBI, three trials explicitly excluded patients with extracranial injuries [16, 23, 31], two trials excluded patients who required immediate surgery [24, 25], and two trials excluded patients who required either massive transfusions or transfusion of fresh frozen plasma [28, 29]. The timing of TXA administration varied among studies: within 2–3 h in 3 trials [16, 17, 25] and within 8 h in 5 trials [23, 24, 28,29,30]. One trial allowed for TXA administration up to 24 h from initial presentation [31]. The dosage of TXA was similar across included trials with the most common regimen being a loading dose of 1 g, followed by a maintenance dose of 1 g over 8 h. One trial compared two different TXA dosing regimens (1 g and 2 g loading dose) versus placebo, and we grouped both TXA arms together for the purposes of analyses [17]. Two of the included trials [23, 31] were judged to be at high RoB, four trials [17, 25, 28, 29] at probably high RoB, and one trial [16] at probably low RoB, while two trials [8, 24] were judged to be at low RoB (see Table 2 for all RoB judgements).

Table 2 Risk of bias assessment
Full size table

Efficacy outcomes

Table 3 shows the summary of findings for all outcomes including the certainty of evidence. Pooled analysis found that TXA likely had no effect on mortality [RR 0.95; 95% CI 0.88–1.02; risk difference (RD) 1.0% reduction; 95% CI 2.5% reduction to 0.4% increase; moderate certainty] (Fig. 2 and Table 3), or disability as assessed with the DRS (MD − 0.18 points; 95% CI − 0.43 to 0.08; moderate certainty), and an uncertain effect on disability based on the proportion of patients with a GOS score less than 4 or a GOS-E score less than or equal to 4 [RR 0.9; 95% CI 0.69–1.17; 0.3% risk difference (RD); 95% CI − 1.1% to 0.6%; very low certainty] (Figs. 3, 4). Of note, one of the studies did not report the standard deviation of the DRS in their published manuscript, but we were able to acquire these data from their results presented on ClinicalTrials.gov [17]. As per the TSA analysis, the optimal information size was not reached for mortality, contributing to the assessment of imprecision and overall moderate certainty (See supplementary appendix, appendix 9, supplement Fig. 12).

Table 3 GRADE summary of findings
Full size table
Fig. 2
figure 2

Forest plot comparing TXA and placebo for the outcome of all-cause mortality at the longest point of follow-up

Full size image
Fig. 3
figure 3

Forest plot comparing TXA and placebo for the continuous outcome of disability using the Disability Rating Scale

Full size image
Fig. 4
figure 4

Forest plot comparing TXA and placebo for the dichotomous outcome of disability using the Glasgow Outcome Scale with a score of < 4 indicating a poor outcome and the Glasgow outcome scale-extended ≤ 4 indicating a poor outcome

Full size image

TXA administration may reduce hematoma expansion on subsequent neuroimaging (see supplementary appendix, appendix 8, supplement Fig. 4b) (RR 0.77; 95% CI 0.58–1.03; RD 3.6% reduction; 95% CI 6.6% reduction to 0.5% increase); however, this was based on low certainty evidence, limited by imprecision. Hematoma expansion as assessed by volume of blood in millilitres (mL) seen on subsequent neuroimaging may also be reduced in patients who received TXA (MD − 2.46 mL; 95% CI − 6.46 mL to 1.55 mL; moderate certainty), although the absolute difference was small (See supplementary appendix, appendix 8, supplement Fig. 4a).

TXA administration had an uncertain effect on hospital length of stay [MD 0.19 days (d); 95% CI − 1.11d to 1.49d; low certainty] and ICU length of stay (MD 1.33d; 95% CI − 0.99d to 3.65d; very low certainty) (supplementary Appendix, appendix 8, supplement Figs. 5–6). We found an uncertain effect on the need for neurosurgical intervention in those receiving as compared to those not receiving TXA (RR 1.11, 95% CI 0.89–1.39; RD 1.7% increase; 95% CI 1.7% reduction to 5.9% increase; low certainty) (supplementary Appendix, appendix 8, supplement Fig. 2). A post hoc subgroup analysis found that TXA administration had a similar effect in low-to-middle-income countries (RR 0.94, 95% CI 0.74–1.18) as it did in high-income countries (p value for subgroup effect > 0.10) (supplementary Appendix, appendix 8, supplement Fig. 11). CRASH-2 and CRASH-3 were not a part of this analysis, because these trials included patients from both high- and low-to-middle-income countries [8, 16].

Safety

We found similar rates of adverse events (a composite outcome variably defined by individual study authors) between those receiving and those not receiving TXA (RR 0.97, 95% CI 0.85–1.11, RD 0%, 95% CI 0.2% lower to 0.1% higher, moderate certainty). Pooled results demonstrated probably no increased risk of deep vein thrombosis (RR 0.94, 95% CI 0.57–1.55, low certainty), vascular occlusive events (RR 0.86, 0.62–1.2, moderate certainty), stroke (RR 0.83, 95% CI 0.53–1.29, moderate certainty), or seizure (RR 1.11, 95% CI 0.92–1.34, moderate certainty) in patients receiving, as compared to those not receiving TXA although confidence intervals for all harm outcomes were wide, and did not rule out the potential for harm (Table 3). There was an uncertain effect of TXA on pulmonary embolism (RR 1.19, 95% CI 0.46–3.06, very low certainty). Of note, the studies which reported on deep vein thrombosis and pulmonary embolism did not comment whether patients were routinely screened for VTE (identifying asymptomatic events) or only imaged if symptomatic [17, 26, 31].

Sensitivity and subgroup analysis

Neither post hoc sensitivity analyses, one excluding the largest trial (CRASH-3) and the other excluding trials enrolling adolescents, showed differences in estimates or conclusions for any of the outcomes of interest (See supplementary appendix for forest plots, appendix 8, supplement Figs. 8–10).

A prespecified subgroup analysis comparing mortality on high RoB studies-to-low RoB studies did not find RoB to be an effect modifier (p value for subgroup interaction = 0.50) (see supplementary appendix, appendix 8, supplement Fig. 7). Although we planned additional subgroups based on severity of TBI and timing of TXA administration, the number of trials reporting separate outcome data for these subgroups of interest did not allow for this analysis.

Discussion

This systematic review and meta-analysis demonstrates that TXA probably does not have an important effect on mortality or disability, and an uncertain effect on need for neurosurgical intervention and length of stay when administered to patients with TBI. TXA probably does not increase the risk of adverse events.

This review was prompted by the publication of the CRASH-3 trial, which concluded that TXA is safe in patients with TBI and that treatment within 3 h of injury reduces head-injury-related death. The primary outcome for CRASH-3 was revised mid-trial from all-cause mortality to head-injury-related mortality at 28 days following injury. In the published statistical analysis plan, the authors explain this change which was made to limit the analysis to causes of death that might be affected by TXA (i.e., head-injury-related death), thereby avoiding dilution of effect from non-head-injury-related deaths [26]. As there may be subjectivity in classifying the cause of death, misclassification is a concern with this approach and could introduce bias in an otherwise objective outcome [26]. We chose to analyze all-cause mortality, as opposed to head-injury-related death, due to the classification issues raised above, and because we anticipated this outcome would be more widely reported across the included trials. In fact, no other trial reported head-injury-related death. Using all-cause mortality, our pooled analysis demonstrates no effect of TXA on all-cause mortality. Given these concerns and because CRASH-3 provided the greatest weight to the pooled analysis, we performed a post hoc sensitivity analysis excluding this trial which did not change the results or conclusions for any of the outcomes of interest.

A previously published meta-analysis examining patients with TBI demonstrated a reduction in mortality with TXA [15]; however, it did not include the latest data, and analyzed all patients enrolled in the CRASH-2 trial, including those with TBI and extracranial traumatic injuries. To limit clinical heterogeneity, we only included the subset of CRASH-2 patients who also had a TBI [30]. Although beneficial in other populations, there are a number of possible explanations for TXA’s lack of efficacy in patients with TBI. Mortality in TBI is best predicted by durations of hypotension, hypoxemia, and pyrexia insults [32,33,34]. As such, the emphasis of management in patients with TBI is limiting secondary brain injury. TXA does not physiologically or mechanistically address these features, which may explain the lack of benefit. In the absence of an effect on survival or disability, it is unclear how important the difference seen in hematoma expansion would be to patients and clinicians in the setting of TBI. The mean difference of -2.46 mL (95% CI − 6.46 to 1.55 mL) is likely of very limited clinical significance, especially without improvements in other more patient-important outcomes. However, even a small difference in hematoma size in a critical location may be relevant.

Although of limited efficacy, these results demonstrate no increased risk of adverse events with the administration of TXA. This finding is consistent with prior large RCTs examining TXA in postpartum hemorrhage, trauma, and intracerebral hemorrhage [8, 11, 12]. In addition to being safe, TXA has been shown to be cost-effective when given to heterogenous trauma patients in low-, middle-, and high-income settings [35]. Of note, our post-hoc subgroup analysis comparing high-income versus low-to-middle-income countries did not show a benefit with the use of TXA in either group. Although TXA is a cheap drug, given its uncertain effects and the subgroup findings, these results do not support the routine use of TXA in high-income or low-to-middle-income countries [35]. It is unlikely that another study as large as CRASH-3 or the out-of-hospital TXA study by Rowell et al. [17] will be conducted over the short term; and as such, these findings likely represent the best summary of evidence on which clinicians have to guide their practice. These results do not support strong directives (either for giving TXA or against giving TXA) to clinicians caring for patients with TBI. Some clinicians may rationalize not giving TXA to these patients given the costs and lack of clear benefit, while others may choose to administer TXA given the lack of demonstrable harm and potential reduction in hematoma size. The findings of this review will be of interest to future guidelines addressing the topic of TXA in TBI who will be able to more carefully consider these aspects of balancing benefits, harms, values, preferences, and costs [9, 36].

It is possible certain subgroups of patients may benefit more or less from TXA; unfortunately, we did not have sufficient trial-level data to perform a number of planned subgroup analyses. The TSA found that the information size was not enough to exclude an important effect with the intervention. It is possible that TXA could be more efficacious in those that receive the drug earlier (for example within 3 h of injury) or in those with differing severity of TBI; however, this analysis is not able to address these questions. Future studies need to focus on these specific populations, with a large enough sample size, including TBI patients with a concomitant hemorrhagic brain injury stratified by subtype of hemorrhage (i.e., subdural, epidural, subarachnoid, and intraparenchymal). Further data are also needed examining the role of TXA in TBI patients also taking oral anticoagulants and antiplatelet agents.

This systematic review and meta-analysis has several strengths including a pre-registered protocol, a comprehensive literature search including unpublished sources, duplicate and independent screening and data abstraction, and GRADE assessment of certainty of evidence. There are also limitations. First, we were unable to perform a number of pre-planned subgroup analyses due to lack of sufficient granularity in published data. Second, the included studies were heterogenous in regards to patients enrolled, specifically severity of TBI and presence of extracranial injuries. Fortunately, this clinical heterogeneity did not translate into important inconsistency (statistical heterogeneity) amongst any of the outcomes of interest. Although we tried to limit analysis to studies that enrolled patients with isolated TBI, some included a small number of patients with TBI and extracranial injuries; however, even if included, these extracranial injuries were not severe with clear exclusions for major injuries requiring massive transfusion.

Conclusion

In patients with acute TBI, TXA probably has no effect on mortality or disability. TXA may decrease hematoma expansion on subsequent imaging; however, this outcome is probably of less importance to patients. The use of TXA probably does not increase the risk of adverse events.