Introduction

Gastrointestinal (GI) bleeding is a common presenting complaint of patients arriving to the Emergency Department (ED). Upper gastrointestinal bleeding (UGIB) incidence estimates range between 50 and 100 persons per 100,000 per year. Lower GI bleeding (LGIB) presents less frequently, with approximately 20 per 100,000 per year requiring admission in the USA [1].

UGIB has a mortality rate of between 2 and 14% [2]. Although UGIB presents 2–5 times more frequently, LGIB has a higher mortality rate. One US study showed a mortality rate of 3.12% in LGIB compared to 2.33% in UGIB [3]. Overall, the mortality rate in gastrointestinal bleeding has remained static in the last 50 years, as per NICE guidelines report in 2014, despite medical and surgical advances [4]. This is a concern.

Tranexamic acid (TXA) was developed by Shosuke and Utako Okamoto in Japan in 1965 [5]. TXA is a well-defined synthetic analogue of the amino acid lysine. It reversibly binds lysine receptor sites on plasminogen, the precursor to plasmin formation, an enzyme responsible for clot breakdown. The binding of TXA to plasminogen prevents its conversion to plasmin and therefore reduces the fibrinolytic effects of plasmin.

Both trauma and surgery trigger similar haemostatic responses including fibrinolysis (clot degradation), and in extreme cases, hyperfibrinolysis. Antifibrinolytic agents counteract this by inhibiting clot breakdown by targeting plasmin [6]. Additionally TXA is hypothesised to have an anti-inflammatory effect to counteract the pro-inflammatory properties of plasmin [6].

In 1967, Cox et al. hypothesised that gastric hyperfibrinolysis may be a key factor in inducing rebleeding in gastric and duodenal ulcers. They suggested that antifibrinolytic agents (epsilon-aminocaproic acid (EACA) and TXA) may be theoretically useful in preventing rebleeding and therefore mortality [7]. Nilsson et al. further supported this with a study on the gastric secretions of 39 patients in 1975. It showed that patients with peptic ulcer disease and haemorrhagic gastro-duodenitis had higher levels of fibrinolytic activity [8].

A number of large randomised control trials (RCTs), including CRASH-2 [6], WOMAN [9] and CRASH-3 [10], assessing the use of TXA have emerged as safe and beneficial in reducing mortality in bleeding trauma, obstetric and head injury patients, respectively. TXA is also commonly used in surgery, particularly orthopaedic surgery to reduce operative blood loss [11].

However, the literature in relation to the use of TXA in gastrointestinal bleeding has not conclusively shown any benefit. A 2014 Cochrane review by Bennett et al. found that TXA appeared to have a beneficial effect on mortality in patients with UGIB. However, in 2021, a large (n = 12,009) multicentre international double-blinded randomised control trial (HALT-IT) found that TXA did not reduce mortality and conferred an increased risk of VTE events. The recommendation from the HALT-IT trial was that TXA should not be used in management of GI bleeding [12].

Since the publication of HALT-IT, two notable systematic reviews and meta-analyses looking into TXA in UGIB were performed by Burke et al. and Kamal et al. They included data from HALT-IT [13, 14]. Burke et al. included 8 RCTs and concluded that while TXA did not have a beneficial effect on mortality, it reduces rebleeding and need for surgery [13]. Kamal et al. included 12 RCTs and concluded that TXA did not improve mortality or other secondary outcomes and conferred an increased risk of venous thromboembolism (VTE). However, this increased risk of VTE only became statistically significant when assessing the subgroup of high-dose TXA vs placebo and the association between TXA at any dose and VTE was not statistically significant [14].

When it comes to complications associated with TXA, specifically increased risk of VTE, there is contradicting information in the literature. From a theoretical perspective, studies in animal models have shown increased thrombus formation in a dose-dependent manner with administration of TXA [15]. However, large RCTs have shown that TXA does not increase the risk of VTE [6], while others show new evidence to dispute this [12].

The aim of this research was to systematically compile and evaluate all of the available RCT data on TXA in GI bleeding and ascertain whether it is safe and/or beneficial in reducing mortality and to review other outcomes. Specifically the aim of this review was to build on previous systematic reviews and meta-analyses by Burke et al. and Kamal et al. [13, 14] by including trials assessing TXA in lower gastrointestinal bleeding along with UGIB, similar to the approach taken in the HALT-IT trial. This review is the first to incorporate both lower and upper gastrointestinal bleeding into the same systematic review and meta-analysis. We also wanted to evaluate whether the increased risk of VTE seen in the HALT-IT trial would be confirmed by meta-analysis including other trials assessing TXA in GI bleeding [12].

Methods

The systematic review and meta-analysis of published RCTs was conducted in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [16]. As the research review pertains to previously published clinical trials, it does not require further consent of patients involved, and, in-keeping with local institution review board, does not require ethical approval. All of the information in the trials was anonymised and, as such, there will be no sensitive patient information requiring storage and is therefore in compliance with General Data Protection Regulation (GDPR) guidelines. All of the aforementioned study authors have contributed to the formulation of the study protocol which was subsequently registered on PROSPERO on 8th March 2022, Reference No. CRD42022308878 [17].

PICO (Population, Intervention, Comparison and Outcome) tool

The PICO tool [18] was utilised to formulate the clinical research question, see Table 1.

Table 1 Population, Intervention, Control and Outcome (PICO) table
Full size table

Aim and outcomes

The aim of this review was to assess the efficacy and safety of the use of TXA in acute gastrointestinal bleeding. The primary outcome was the effect of TXA on mortality in patients with gastrointestinal haemorrhage when compared with placebo. The secondary outcomes are detailed in Table 2.

Table 2 Table of secondary outcomes
Full size table

Search strategy

The detailed search criteria for the review was composed by converting key words into Medical Subject Headings (MeSH) terms. The search terms “Gastrointestinal Hemorrhage” [Mesh] AND “Tranexamic Acid” [Mesh] were then used to perform an electronic search of MEDLINE (PubMed), EMBASE, Cochrane (CENTRAL) and Scopus databases on 31st March 2022. Searches of clinical trial registries, EU Clinical Trials Register and ClinicalTrials.gov, were also performed. All published studies regardless of publishing language were included, provided an English translation was available. There were no restrictions based on the year of publication. The retrieved studies had duplicates removed and then the titles, abstracts and full texts were screened to identify the studies for review.

Inclusion and exclusion criteria

The identified studies were considered if the inclusion criteria, see Table 3, was met. Studies were omitted based on the exclusion criteria, see Table 3.

Table 3 Inclusion and exclusion criteria table
Full size table

Quality assurance and quality control

The literature search was independently conducted by two reviewers using the predetermined search strategy to identify the list of trials. Duplicates were excluded, then screening of titles, abstracts and full texts was performed and application of the inclusion and exclusion criteria was assessed independently. This helped to ensure compliance with systematic review process and independent concordance of the relevant studies prior to data extraction.

Data extraction

After independent identification of studies to be included, 14 suitable randomised control trials were identified. The following data points were extracted, where available, from the individual studies: 1. Manuscript title, 2. Lead author, 3. Year of publication, 4. Country of research facility, 5. Single/multicentre study, 6. Blinding implemented, 7. Study methodology, 8. Number of participants, 9. Participants demographics, 10. Intervention protocol, 11. Primary outcome: mortality, 12. Secondary outcomes, see Table 2. The data was input into a table on Microsoft Excel (version 16.40).

Risk of bias analysis

The risk of bias analysis of included studies was performed using the Cochrane risk-of-bias tool for randomised trials (RoB 2) [19]. This tool is structured into a fixed set of types of bias, each assessing different aspects of trial design, procedures and reporting. It is represented graphically with each study classified as low, unclear and high risk of bias, for each of the different types of biases. The risk of bias analysis was performed independently by reviewers. The modelling and graph was performed using Microsoft Excel (version 16.40).

Data analysis

The primary and secondary outcomes were expressed as dichotomous and continuous outcomes; the dichotomous outcomes were reported as risk ratios (RR) with 95% confidence intervals (CI) following estimation using the Mantel–Haenszel method. Continuous variables were expressed as mean differences (MD) with 95% confidence intervals (CI) using the inverse variance method. Random effects modelling were applied for all primary and secondary outcomes, regardless of statistical heterogeneity (I2) value due to significant qualitative heterogeneity of the studies. All tests of significance were two tailed with P < 0.05 utilised to determine statistical significance. Meta-analysis was performed using Review Manager (RevMan) version 5.4.1 software, the Cochrane Collaboration 2020.

Results

Literature search

The systematic search strategy identified a total of 1497 studies; 576 duplicate records were subsequently removed. The remaining studies initially underwent title review, abstract review and full text review with 921, 256 and 45 reviewed respectively at each stage. Post full text review 19 studies were assessed for eligibility. After further exclusion of 5 studies, 14 remaining studies were determined to have satisfied the inclusion and exclusion criteria and were included in the systematic review and meta-analysis [12, 20,21,22,23,24,25,26,27,28,29,30,31,32,33]. The details of this are outlined in the PRISMA 2020 flowchart, see Table 4 [34].

Table 4 PRISMA 2020 flow diagram [34]
Full size table

Included study characteristics

Of the included 14 studies, there was marked geographical and chronological variation. They include 7 European, 4 Asian, 2 Australian and one large multinational trial (HALT-IT) with participants from 15 countries across the developed and developing world. The 14 studies included data from 14,338 patients, 34.93% were female (n = 5008), and 65.07% were male (n = 9330). The weighted mean age overall was 58.4 years with the mean of the TXA group being 58.34 years and the placebo group being 58.46 years. The year of publication of the trials spanned 48 years from Cormack et al. in 1973 to HALT-IT and Bashiri et al. in 2021 [12, 20,21,22,23,24,25,26,27,28,29,30,31,32,33]. While all 14 trials included placebo groups, there were significant variation in the intervention groups. TXA was administered through a variety of routes, at different doses and for different durations. See Appendix for details of individual included study characteristics. From the time of the first trial in 1973 to the last in 2021, there have been a number of notable changes to management of gastrointestinal bleeding, with the routine use of proton pump inhibitors (PPI) and the emergence of therapeutic endoscopic methods to control gastrointestinal bleeding [35]. For these reasons, it was decided that there was qualitatively significant heterogeneity between the trials and that random effect modelling should be used for meta-analysis calculation.

Risk of bias assessment

The risk of bias for the included studies was evaluated using the Cochrane risk-of-bias tool for randomised trials (RoB 2) [19]. The evaluation of bias was independently carried out by two reviewers. The results of the evaluation are graphically represented in Table 5.

Table 5 Risk of bias assessment table [19]
Full size table

Tabulated meta-analysis outcomes (Table 6)

Table 6 Tabulated meta-analysis outcomes
Full size table

Primary outcome: mortality

All 14 trials reported outcomes for analysis in relation to mortality (100%). The meta-analysis included data from 14,338 patients. However, the results from Bashiri et al. (n = 70) were not estimable as there were no incidences of mortality in either the TXA or placebo groups [22]. Results of the meta-analysis (Fig. 1) revealed a risk ratio of 0.86 with 95% confidence interval ranged from 0.74 to 1.00 and P 0.05; as such the result did not reach statistical significance. The I2 value was 0%; however, random effects modelling was used as described above, see “Included study characteristics”.

Fig. 1
figure 1

Meta-analysis forest plot assessing the effect of TXA versus placebo on mortality

Full size image

Secondary outcomes

Rebleeding rate

Twelve of the 14 trials reported outcomes for analysis in relation to rebleeding rate (85.7%). The meta-analysis included data from 13,968 patients. Results of the meta-analysis (Fig. 2) revealed a risk of ratio of 0.85 with 95% confidence interval from 0.71 to 1.03 and P 0.09; as such the result did not reach statistical significance. The I2 value was 21%; however, random effects modelling was used as described above, see “Included study characteristics”.

Fig. 2
figure 2

Meta-analysis forest plot assessing the effect of TXA versus placebo on rebleeding

Full size image

Need for intervention

Eight of the 14 trials reported outcomes in relation to intervention (57.1%). Where specific intervention data was provided (therapeutic endoscopic intervention and surgical intervention), the data was analysed separately, see below meta-analysis for need for therapeutic endoscopic intervention and need for surgery. This meta-analysis includes data from 13,195 patients. Results of the meta-analysis (Fig. 3) reveal a risk ratio of 0.92 with a 95% confidence interval from 0.76 to 1.11 and P 0.38; as such the result did not reach statistical significance. The I2 value was 22%; however, random effects modelling was used as described above, see “Included study characteristics”.

Fig. 3
figure 3

Meta-analysis forest plot assessing the effect of TXA versus placebo on the need for intervention

Full size image

Need for therapeutic endoscopic intervention

Six of the 14 trials reported outcomes in relation to therapeutic endoscopic intervention (42.9%). The meta-analysis included data from 12,905 patients. Results of the meta-analysis (Fig. 4) reveal a risk ratio of 0.92 with a 95% confidence interval from 0.7 to 1.22 and P 0.57; as such the result did not reach statistical significance. The I2 value was 37%; however, random effects modelling was used as described above, see “Included study characteristics”.

Fig. 4
figure 4

Meta-analysis forest plot assessing the effect of TXA versus placebo on the need for therapeutic endoscopic intervention

Full size image

Need for surgical intervention

Twelve of the 14 trials reported outcomes in relation to surgical intervention (85.7%). The meta-analysis included data from 13,944 patients. However, Saidi et al. (n = 131) and Bashiri et al. (n = 70) were not estimable as there were no incidences of surgical intervention in either the TXA or placebo groups in either trial [22, 36]. Results of the meta-analysis (Fig. 5) revealed a risk ratio of 0.75 with a 95% confidence interval from 0.53 to 1.07 and P 0.11; as such the result did not reach statistical significance. The I2 value was 52% and as such random effects modelling was used.

Fig. 5
figure 5

Meta-analysis forest plot assessing the effect of TXA versus placebo on the need for surgical intervention

Full size image

Transfusion requirement

Ten of the 14 trials reported outcomes in relation to transfusion requirement (71.4%). The meta-analysis included data from 13,343 patients. Results of the meta-analysis (Fig. 6) revealed a risk ratio of 1.01 with a 95% confidence interval from 0.94 to 1.07 and P 0.86; as such the result did not reach statistical significance. The I2 value was 15%; however, random effects modelling was used as described above, see “Included study characteristics”.

Fig. 6
figure 6

Meta-analysis forest plot assessing the effect of TXA versus placebo on transfusion requirement

Full size image

Transfusion volume

Five of the 14 trials reported outcomes in relation to mean transfusion volume (35.7%), expressed as units of packed red blood cells. The meta-analysis included data from 12,352 patients. However, data from Bergqvist et al. was not estimable as details of standard deviation or other variances metrics were not provided [20]. Results of the meta-analysis (Fig. 7) revealed a mean difference of − 0.68 with 95% confidence interval from − 1.51 to 0.14 and P 0.10; as such the result did not reach statistical significance. The I2 value was 84%; however, random effects modelling was used as described above, see “Included study characteristics”.

Fig. 7
figure 7

Meta-analysis forest plot assessing the effect of TXA versus placebo on transfusion volume

Full size image

Overall length of stay

Four of the 14 trials reported outcomes in relation to mean length of stay as hospital inpatient (33.3%). The meta-analysis included data from 12,595 patients. Results of the meta-analysis (Fig. 8) reveal a mean difference of 0.03 with 95% confidence interval from − 0.03 to 0.08 and P 0.31; as such the result did not reach statistical significance. The I2 value was 0%; however, random effects modelling was used as described above, see “Included study characteristics”.

Fig. 8
figure 8

Meta-analysis forest plot assessing the effect of TXA versus placebo on overall length of stay

Full size image

ICU length of stay

Two of the 14 trials reported outcomes in relation to mean length of ICU stay (16.7%). Hawkey et al. reported proportion of patients requiring ICU stay but not the mean length and as such have been omitted from this meta-analysis [27]. Results of the meta-analysis (Fig. 9) reveal a mean difference of − 0.92 with 95% confidence interval from − 2.26 to 0.43 and P 0.18; as such the result did not reach statistical significance. The I2 value was 100%, and as such random effects modelling was used.

Fig. 9
figure 9

Meta-analysis forest plot assessing the effect of TXA versus placebo on ICU length of stay

Full size image

Venous thromboembolic events

Seven of the 14 trials reported outcomes in relation to venous thromboembolic events (50%), including deep venous thrombosis (DVT) and pulmonary embolism (PE). The meta-analysis included data from 13,209 patients; however, Saidi et al. (n = 131) were not estimable as there were no incidences of VTE in either TXA or placebo groups [36]. Results of the meta-analysis (Fig. 10) revealed a risk ratio of 1.29 with a 95% confidence interval from 0.53 to 3.16 and P 015; as such the result did not reach statistical significance. The I2 value was 39%; however, random effects modelling was used as described above, see “Included study characteristics”.

Fig. 10
figure 10

Meta-analysis forest plot assessing the effect of TXA versus placebo on venous thromboembolic events

Full size image

Arterial thromboembolic events

Five of the 14 trials reported outcomes in relation to arterial thromboembolic events (35.7%), including cerebrovascular accidents (CVA), transient ischaemic attacks (TIA) and myocardial infarctions (MI). The meta-analysis included data from 12,777 patients. Results of the meta-analysis (Fig. 11) revealed a risk ratio of 0.94 with a 95% confidence interval from 0.63 to 1.4 and P 0.75; as such the result did not reach statistical significance. The I2 value was 0%; however, random effects modelling was used as described above, see “Included study characteristics”.

Fig. 11
figure 11

Meta-analysis forest plot assessing the effect of tranexamic acid versus placebo on arterial thromboembolic events

Full size image

Other adverse effects

There were a number of other adverse effects discussed in the 14 trials; however, there was insufficient data in order to perform a meta-analysis. The most commonly described adverse effect in the TXA group was GI upset. The other finding of note comes from the large HALT-IT trial which described a statistically significant increased risk of seizure in the TXA group when compared with the placebo group. HALT-IT detailed seizures in 38 patients (0.6%) in the TXA group versus 22 patients (0.4%) in the placebo group, with a risk ratio of 1.73 with 95% confidence interval of 1.03 to 2.93 [12]. It was not possible to perform a meta-analysis as the other trials had insufficient data on seizures reported.

Subgroup analysis

TXA on mortality in upper GI bleeding versus lower GI bleeding

The majority of trials exclusively studied the use of TXA in upper GI bleeding. HALT-IT and Smith et al. were the only 2 trials that included patients with lower GI bleeding. Subgroup analysis looking at mortality in patients with lower GI bleeding showed no statistically significant difference between TXA and placebo with a risk ratio of 1.66 with 95% confidence interval of 0.66 to 4.19 and 0.28. However, the subgroup meta-analysis of mortality in upper GI bleeding demonstrated a statistically significant difference between TXA and placebo with a risk ratio of 0.85 with 95% confidence interval of 0.72 to 0.99 and P 0.03 (Fig. 12).

Fig. 12
figure 12

Subgroup meta-analysis of TXA versus placebo on mortality in upper and lower GI bleeding

Full size image

TXA on mortality in variceal versus non-variceal bleeding

Variceal bleeding data was not included in enough detail in many of the RCTs to facilitate adequate inclusion in the meta-analysis. Trials such as Bashiri et al., Karadas et al. and Saidi et al. excluded patients with variceal bleeding during initial patient selection. A number of studies (Bergqvist et al. and Hawkey et al.) included sufficient detail to be included but there were insufficient numbers in the variceal bleeding groups to be included in the meta-analysis. The subgroup meta-analysis of TXA versus placebo of non-variceal bleeding included data from 13 RCTs and it demonstrated a statistically significant result with a risk ratio of 0.75 with 95% confidence interval of 0.6 to 0.94 and P 0.01. Only 4 trials included sufficient data to be included in the subgroup meta-analysis of TXA versus placebo in variceal bleeding which did not show a statistically significant result; risk ratio 0.98 with 95% confidence interval of 0.79 to 1.21 and P 0.84 (Fig. 13).

Fig. 13
figure 13

Subgroup meta-analysis of TXA versus placebo on mortality in non-variceal and variceal bleeding

Full size image

Discussion

The primary outcome of this systematic review and meta-analysis showed that tranexamic acid did not reduce mortality in patients with gastrointestinal bleeding when compared with placebo (RR: 0.86 95% CI (0.74–1.00), P = 0.05). This builds on the evidence detailed in previous meta-analyses by Burke et al. and Kamal et al. [13, 14]. Similar to these reviews, this meta-analysis was heavily influenced by the large HALT-IT trial (n = 12,009) which found no reduction in mortality with TXA when compared with placebo. Prior to HALT-IT, a Cochrane review by Bennett et al. in 2014 had concluded that TXA appears to have a beneficial effect on mortality but this was uncertain due to a high dropout rate in a number of the trials [37].

Rebleeding is important in the pathogenesis and natural history of gastrointestinal bleeding. It often marks the transition from a stable to an unstable patient and precedes terminal bleeding events resulting in mortality [38]. As such, a medication that could prevent rebleeding could be useful in reducing mortality, ICU admission and transfusion requirements. However, despite this theoretical support for TXA in preventing rebleeding, this meta-analysis found no statistically significant reduction in rebleeding rates when TXA was compared with placebo (RR: 0.83 95% CI (0.71–1.03), P = 0.09).

This meta-analysis also assessed the role of TXA in reducing the need for intervention in gastrointestinal bleeding. Where segregated data was available within the trials, we divided intervention into therapeutic endoscopic interventions and surgical intervention. Looking firstly at overall intervention, there was no statistically significant difference between TXA and placebo (RR: 0.92 95% CI (0.76–1.11), P = 0.038). Similarly there was no reduction in therapeutic endoscopic interventions (RR: 0.92 95% CI (0.7–1.22), P = 0.57) or surgical intervention (RR: 0.75 95% CI (0.53–1.07), P = 0.11) with TXA when compared with placebo. We also looked at interventional radiology (IR) guided embolisation but data on this was only reported in the HALT-IT trial. Similar to other interventions, there was no statistically significant risk reduction with TXA when compared with placebo (RR: 0.83 95% CI (0.61–1.13)).

From a theoretical perspective, TXA, as an antifibrinolytic agent, may prevent fibrinolysis and increase the risk of propagation and embolisation of clots and therefore increase the incidence of thromboembolic events (TE) [15]. In 2021, Taeuber et al. [39] performed a more extensive systematic review and meta-analysis which incorporated 216 trials and 125,550 patients. They found that thromboembolic events occurred in 2.1% (n = 1020) of the intravenous TXA group versus 2.0% (n = 900) in the control group. They found no association between intravenous TXA and thromboembolic events (risk difference: 0.001 95% CI (− 0.001–0.002), P = 0.49). Our meta-analysis looked at both arterial thromboembolic events (MI, CVA, TIA) and venous thromboembolic events (DVT, PE). Only 6 trials reported data on VTE, and 5 trials reported data in relation to arterial thromboembolic events. Despite the HALT-IT trial reporting increased risk of VTE [12], our meta-analysis found that there was no statistical significance between TXA with venous thromboembolic events (RR: 1.20 95% CI (0.90–1.60), P = 0.22) or arterial thromboembolic events (RR: 0.94 95% CI (0.63–1.40), P = 0.75) when compared with placebo. Our meta-analysis finding is consistent with the findings of Taeuber et al. in showing no association between TXA and thromboembolic events.

Of other adverse effects mentioned throughout the trials, nausea and vomiting was the most common. However, data in relation to this outcome was not provided in sufficient detail to facilitate a meta-analysis. Many trials included incidences of nausea and vomiting but either did not specify the breakdown of cases between TXA and placebo [20,21,22,23, 26, 27, 29, 31, 32, 36] groups or provided data on events in the TXA group with no comparative figures from the placebo group. The other adverse effect of note was seizures. While the majority of trials did not report incidence of seizures, HALT-IT reported a statistically significant increase risk of seizures in the TXA group when compared with placebo, with 38 events occurring in the TXA group compared with 22 in the placebo (0.6% vs 0.4%, RR 1.73 95% CI (1.03–2.93)) [12]. However, as highlighted by Murao et al. in their systematic review, the risk of seizures from TXA is dose dependent [40]. They noted that specifically doses of above 2 g of TXA per 24 h confer an increased risk of seizure. The TXA doses used for GI bleeding often exceeded this 2 g per 24 h, for example HALT-IT used 1 g stat and a further 3 g over 24 h in a continuous 125 mg/h infusion. The doses used in TXA for GI bleeding often exceeded the doses used in other large TXA trials, such as CRASH-2 (1 g stat and 1 g infused over 24 h) and WOMAN (1 g stat and further 1 g if recurrent bleeding). Therefore, the contraindication of TXA in patients with known seizure disorders may only be applicable when using higher doses of TXA, specifically when using fixed dose in patients suffering with renal impairment or liver disease.

The aim of this systematic review was to build on previous reviews performed by Burke et al. and Kamal et al. [13, 14] and to specifically include studies investigating lower gastrointestinal bleeding, similar to the HALT-IT trial [12]. However, this strategy only yielded 2 additional trials Smith et al. (n = 96) and Bashiri et al. (n = 70) [22, 31]. The meta-analysis outcomes of this review have not revealed any additional findings when compared with Kamal et al. [14].

Subgroup analysis was performed looking into whether TXA was beneficial in patients with upper GI bleeding, lower GI bleeding and whether the bleed aetiology was non-variceal or variceal bleeding. Subgroup analysis demonstrated that when upper GI bleed cases were isolated there was a statistically significant mortality risk reduction with TXA when compared with placebo, RR 0.85 95% CI (0.72–0.99). However, the weight of the HALT-IT RCT is diluted in this subgroup meta-analysis. HALT-IT is responsible for 82.7% of patients in the overall meta-analysis (10,681/12,914); however, its weighting in the meta-analysis is 68.7%.

It must also be noted that the trial HALT-IT is more rigorous in its methodology than the smaller, more biased trials. These two facts combined and the proximity of the upper margin of the confidence interval to 1 would lend doubt to the true significance of the subgroup meta-analysis findings. HALT-IT individually found no statistically significant difference in mortality in the upper GI bleeding subgroup. It is possible the segregation of upper and lower GI diluted the statistical power of HALT-IT allowing smaller, less methodologically rigorous trials to push the meta-analysis calculation towards statistical significance. It must also be noted that the well designed and conducted HALT-IT trial subgroup analysis clearly showed no statistical difference between the TXA and placebo groups both with a risk ratio close to 1 (0.97) and a confidence interval sitting equally either side of 1 from 0.81 to 1.17.

Subgroup analysis assessing the effect of TXA versus placebo on mortality in patients with non-variceal bleeding demonstrated a statistically significant risk reduction, RR 0.75 95% CI (0.60–0.94). Similar to the above criticism of the subgroup analysis for UGIB, the non-variceal bleeding subgroup meta-analysis dilutes the impact of the HALT-IT RCT. HALT-IT accounts for 74% (6576/8886) of patients but was only weighted at 19%. Also when looking at the subgroup analysis from the HALT-IT trial individually, it clearly shows no difference between the subgroups with risk reduction of 1.01, 95% CI (0.71–1.43).

Subgroup analysis looking into dosing or route of administration was not performed as part of this review. However, a systematic review and meta-analysis by Dionne et al. assessed the difference between high-dose and low-dose TXA for both upper and lower GI bleeding [41]. They found that high-dose IV TXA did not improve outcomes and conferred an increased risk of VTE and seizures. When looking at low-dose TXA, their meta-analysis showed no effect on mortality but did show a statistically significant risk reduction in rebleeding and need for surgery. Their findings contradict the expected dose–response relationship and, coupled with the somewhat arbitrary delineation of “low” versus “high” dose, it is uncertain whether there findings are of true significance.

Overall looking into the different subgroups, it is uncertain whether the statistically significant meta-analysis results hold up as truly significant on closer analysis. If anything can be taken from the subgroup analysis, it may be that future research or trials should focus more on the use of TXA in cases of non-variceal upper GI bleeding. This also has some pathophysiological support, through inhibition of gastric hyperfibrinolysis described by Cox et al. [7]. Future research should potentially focus on thromboelastography (TEG) to identify acute gastrointestinal bleeding patients with evidence of hyperfibrinolysis. TXA is likely to be more beneficial in this subset of patients.

This systematic review and meta-analysis has a number of limitations. These include the high risk of bias and high numbers of dropouts in a number of the older, smaller RCTs. There are also significant differences in the number of participants. The majority of the older, smaller trials are eclipsed in terms of participants by HALT-IT which accounts for 82.7% of the overall participants in the review. The chronology of the trials is also a limiting factor. The first trial included is Cormack et al. which was performed in 1973 with the latest trial, HALT-IT being performed in 2021. This duration, spanning 48 years, results in bias due to the significant changes in medical practice and the management of acute gastrointestinal bleeding over the time period.

There are a significant amount of non-randomised cohort studies, audits and case reports that looked at the use of TXA in gastrointestinal bleeding. Many of these, due to poor or biased study methodology or low participant numbers, do not provide robust outcomes that challenge the RCTs discussed in this systematic review. However, two studies merit mention.

The first is a large retrospective cohort study (n = 10,254) from Taiwan by Ting et al. published in 2022 [42]. The retrospective study used data from Taiwan’s National Health Insurance Research Database (NHIRD) to compare early administration of TXA versus late administration of TXA for gastrointestinal bleeding, with “early” being determined as within the ED and “late” being after being admitted from the ED. Multiple Cox regression analysis revealed significantly lower mortality in the early TXA versus the late TXA group (adjusted hazard ratio 0.64 95% CI (0.57–0.73)). It also reported no significant increase in thromboembolic events in early versus late TXA groups (adjusted hazard ratio 1.03 95% CI (0.94–1.12)). However, it must be noted that Ting et al. compared early versus late TXA administration instead of TXA versus control or TXA versus no TXA groups. Despite the number of participants in this cohort studies, due to the retrospective approach and cohort study methodology, it is more susceptible to biases than an RCT. Given the limitations and inherent biases of this type of study, it is uncertain that the results are sufficient to contradict the findings of the more methodologically rigorous randomised control trials, such as HALT-IT [12].

Another large Japanese retrospective observational study (n = 61,052) by Miyamoto et al. assessed the use of TXA for diverticular lower gastrointestinal bleeding and determined that there was no statistically significant difference in in-hospital mortality between the TXA and control group (OR: 1.07 95% CI (0.88–1.30)) [43]. They did find that the TXA group were less likely to develop severe bleeding (OR: 0.93 95% CI (0.89–0.99)) or require blood transfusions (OR: 0.88 95% CI (0.84–0.92)). The TXA group had shorter hospital stays − 0.23 days 95% CI (− 0.01–0.44) and cost of admission was also reduced (− 233 USD 95% CI (− 153 to − 314)). This study does not show a benefit in relation to mortality. When interpreting the secondary outcomes (decreases in severe bleeding and need for transfusion), again similar to Ting et al., it is important to note this trial’s limitations and biases when trying to compare with well conducted RCTs.

There are 6 ongoing trials assessing TXA in gastrointestinal bleeding with statuses of either unknown or recruiting on the EU Clinical Trials Registry and ClinicalTrials.gov registry, see Table 7. Two of these, EXARHOSE [44], a French study with an estimated enrolment of 500, and an Indian RCT with an enrolment target of 600, are specifically looking at the use of TXA in liver cirrhosis-related upper gastrointestinal bleeding. The total estimated number of patients to be enrolled in these upcoming trials is 2034. The status of some trials is unknown on the trial registers and therefore their outcomes may not be published. This includes HEXUGI, an Israeli trial commenced in 2014 (est. n = 300), and TAUGI, a South Korean trial commenced in 2012 (est. n = 414). It is difficult to ascertain the impact these ongoing trials may have in the debate on the use of TXA in gastrointestinal bleeding. It is unlikely that these trial results would greatly influence the field given the low enrolment numbers when compared with the already published trials, especially that of the large HALT-IT trial (n = 12,009).

Table 7 Recruiting or unknown status RCTs from clinical trial registers
Full size table

Conclusion

The primary outcome of this systematic review and meta-analysis showed that tranexamic acid did not reduce mortality in patients with upper or lower gastrointestinal bleeding when compared with placebo. Our study, the first to incorporate both upper and lower gastrointestinal bleeding together, also found no improvement in secondary outcomes (rebleeding, transfusion requirement, need for intervention or length of stay) when compared with placebo. We found no significant association between tranexamic acid and venous thromboembolic events. However, there appears to be an increased risk of seizures with tranexamic acid. Given the results of our meta-analysis, we would not recommend the routine use of tranexamic acid for patients presenting with acute gastrointestinal bleeding.