Effect of tranexamic acid on thrombotic events and seizures in bleeding patients: a systematic review and meta-analysis

We included trials with the following characteristics (1) study types: randomized trials, (2) population: patients with traumatic, surgical, obstetric, intracranial or gastrointestinal bleeding, (3) intervention: intravenous tranexamic acid at any dose, (4) control: placebo or no intervention (without TXA administration) and (5) outcomes: any thrombotic events (venous thromboembolism, acute coronary syndrome and stroke) and seizures.

We conducted a comprehensive search of MEDLINE (via PubMed), EMBASE and the Cochrane Central Register of Controlled Trials from inception until, June 1, 2020, for randomized trials comparing intravenous TXA and placebo or no intervention in bleeding patients (trauma, surgery, post-partum hemorrhage, spontaneous intracranial hemorrhage, gastrointestinal hemorrhage). We did not apply any language or time restrictions to the electronic searches. The search strategy in MEDLINE was as follows: (Tranexamic acid [mesh] OR tranexamic acid [tiab] OR TXA [tiab]) AND ((randomized controlled trial [pt] OR controlled clinical trial [pt] OR randomized [tiab] OR clinical trials as topic [mesh: noexp] OR randomly [tiab] OR trial [ti]) NOT (animals [mh] NOT humans [mh])). Our MEDLINE search strategy was adapted for searches in the other two databases (Additional file 1: Table S1). Reference lists of eligible studies were hand-searched to identify further relevant studies.

Two authors (HN, SM) independently screened articles for inclusion on the basis of title and abstracts and reviewed relevant articles as full text. Disagreements were resolved by discussion and referral to a third author (KY) if necessary. Two authors (HN, SM) extracted the study characteristics from each included study, including year of publications, study population, number of participants, name of comparators, dose of treatment, indication for treatment, name and types of adverse events and information assessing risk of bias in the studies.

Two authors (HN, SM) independently assessed the risk of systematic errors (bias) of the trials included in the meta-analysis according to the Cochrane Handbook, version 6.1 [16]. To evaluate the risk of bias in the individual RCTs, we will use the revised uniform criteria of the Cochrane risk-of-bias tool for randomized trials ver. 2 (RoB 2).

Risk of bias was rated according to the following domains (1) bias due to randomization, (2) bias due to deviations from indented interventions, (3) bias due to missing data, (4) bias due to outcome measurement, (5) bias due to selection of reported result. Trials adjudicated as having concerns or having high risk of bias for one or more domains were classified as having an overall high risk of bias.

The primary outcome measure was the number of patients with any thrombotic events, including acute coronary syndrome (ACS), stroke/transient ischemic attack (TIA) and venous thromboembolism (VTE). The number of patients with ACS, stroke/TIA and VTE was also assessed separately. The number of patients with seizures was also measured as primary outcome.

Statistical analyses were performed using Review Manager Software 5 (Review Manager [RevMan] Version 5.4. Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2020) and STATA software, V.15.0 (STATA Corporation, College Station, TX, USA). For each included trial, we calculated the relative risks with 95% confidence intervals for all outcome measures. Heterogeneity among the trials was explored by inspection of forest plots and calculation of I² statistics [16]. We performed both fixed effects analysis using the Mantel–Haenszel method and random effects analyses using the DerSimonian and Laird estimator, reporting the most conservative summary estimate with the broadest confidence interval [17]. Pre-specified subgroup analyses were performed as standard using RevMan 5 software.

We conducted analysis to establish, first, the effects of TXA on any thrombotic events and on seizures in bleeding patients. We also assessed the effect of TXA on venous thromboembolism, acute coronary syndrome and stroke separately. We examined the following pre-planned sensitivity analyses: type of underlying disease (trauma, cardiac surgery, orthopedic surgery, obstetrics and gynecology, spontaneous intracranial hemorrhage, gastrointestinal bleeding), TXA dose (≤ 2 g or > 2 g per day), sample size (< 500 or ≥ 500), risk of bias (low risk of bias or high risk of bias). Additionally, we conducted subgroup analysis in children. Meta-regression analysis was carried out to investigate the association between the incidence of adverse events and the dose of TXA.

Statistical heterogeneity will be evaluated informally from forest plots of the study estimates and more formally using the chi-squared test (P value < 0.1 = significant heterogeneity) and I² statistic (I² > 50% = significant heterogeneity).

To investigate publication biases, we created funnel plot in which the log RRs were plotted against their standard errors and tested the symmetry of the funnel plots with Begg’s rank correlation test and Egger’s linear regression test.

We also assessed the overall certainty of evidence for each outcome using the grading of recommendations assessment, development and evaluation (GRADE) approach.

In total, we screened 6761 abstracts, of which 377 were eligible for full-text review. 234 trials with 102,681 participants were eligible for inclusion in the meta-analysis (Fig. 1).

Table 1 summarizes characteristics of included trials. 124 (53%) trials were in orthopedic surgery, 47 trials (20%) were in cardiac surgery, 22 trials (9%) were in obstetric and gynecological surgery, 7 trials (3%) were in trauma (including traumatic brain injury), 9 trials (4%) were in spontaneous (non-traumatic) intracranial hemorrhage, 3 trials (1%) were in gastrointestinal hemorrhage. 13 trials had sample sizes over 500. [5, 6, 12, 18‐27] 9 trials involved children. Among 234 trials included in this meta-analysis, data were available on any thrombotic events for 230 trials, on venous thromboembolism for 208 trials, on acute coronary syndrome for 107 trials, on stroke for 106 trials and on seizures for 40 trials.

Overall, 109 trials were deemed to be at low risk of bias. 182 trials were judged at low risk of bias for randomization process, 158 trials for deviations from intended interventions, 224 trials for missing outcome data and 125 trials for measurement of outcome result (Additional file 2: Figure S1). Supplementary material is attached showing the characteristics and risk of bias of the included trials (Additional file 1: Table S2; Additional file 2: Figure S2).

We found no evidence that TXA administration increased the risk of thrombotic events (RR = 1.00 [95% CI 0.93–1.08], I² = 0%, high certainty of evidence), seizures (1.18 [0.91–1.53], I² = 34%, moderate certainty), venous thromboembolism (1.04 [0.92–1.17], I² = 0%), acute coronary syndrome (0.88 [0.78–1.00], I² = 0%) or stroke (1.12 [0.98–1.27], I² = 0%) (Fig. 2). These results were consistent for both random effects and fixed effect statistical model and we observed no evidence of publication bias, either when evaluating the funnel plot (Additional file 2: Figures S3 and S4) or statistically for thrombotic events (Begg's rank correlation test: p = 0.12 and Egger's linear regression test: 0.424) and seizures (Begg's rank correlation test: p = 0.63 and Egger's linear regression test: 0.199). The certainty of evidence was judged as high for thrombotic events but moderate for seizures due to imprecision (Table 2).

In sensitivity and subgroup analysis, we did not find any increase in thrombotic events in trauma (0.89 [0.69–1.17]), obstetrics and gynecology (0.81 [0.53–1.26]), cardiac surgery (0.92 [0.81–1.05]), orthopedic surgery (0.98 [0.80–1.21]) (Fig. 3A; Additional file 2: Figure S5). However, the risk of thrombotic events appeared higher in the TXA group than in the control group in spontaneous (non-traumatic) intracranial hemorrhage (1.33 [1.09–1.63]) (Fig. 3A). In a dose-by-dose sensitivity analysis, thrombotic events were not increased in patients receiving 2 g/day of TXA or less (≤ 2 g/day: 0.94 [0.84–1.05]) (Fig. 3A; Additional file 2: Figure S6). There was uncertainty on thrombotic events in patients receiving more than 2 g/day of TXA (> 2 g/day: 1.18 [0.98–1.43]). Meta-regression analysis showed there was no statistically significant association between higher dose of TXA and increased risk of thrombotic events (p = 0.122) (Fig. 4A). The results were similar to primary analysis, when the analyses were restricted to trials with low risk of bias (0.98 [0.90–1.06]) and sample size over 500 (0.96 [0.85–1.07]) (Fig. 3A). Of trials conducted in children, one patient had a thrombotic event in the TXA group, compared to none in the placebo group (Additional file 2: Figure S7).

In a per disease analysis, the risk of seizures was higher in the TXA group than in the control group in cardiac surgery (4.20 [1.56–11.32]) and gastrointestinal hemorrhage (1.73 [1.03–2.93]) (Fig. 3B; Additional file 2: Figure S8). In a dose-by-dose analysis, there was an observed increase in seizures in patients receiving more than 2 g/day of TXA (> 2 g/day: 3.05 [1.01–9.20]). Meanwhile, we found no increase in seizures in patients receiving 2 g/day of TXA or less (≤ 2 g/day: 1.02 [0.88–1.19]) (Fig. 3B; Additional file 2: Figure S9). Meta-regression analysis showed relative increase for seizures to be proportional to the magnitude of the TXA dose increase (p = 0.011) (Fig. 4B). There were no seizures reported in trials in children (Additional file 2: Figure S10).

We also performed sensitivity analysis for separately evaluating venous thromboembolism, acute coronary syndrome and stroke stratified by underlying disease (Fig. 5). The risk of venous thromboembolism was higher in TXA group than control group in gastrointestinal hemorrhage (1.89 [1.21–2.96]) (Fig. 5; Additional file 2: Figure S11). In other study populations, we found no increase in venous thromboembolism. We found no evidence TXA increased the risk of acute coronary syndrome in trauma, obstetrics and gynecology, cardiac surgery, orthopedic surgery and gastrointestinal hemorrhage (Fig. 5; Additional file 2: Figure S12). The risk of stroke appeared higher in the TXA group than in the control group in patients with intracranial hemorrhage (1.40 [1.05–1.86]) (Fig. 5; Additional file 2: Figure S13).

In this systematic review and meta-analysis, intravenous tranexamic acid administration did not appear to increase the risk of thrombotic events and seizures in bleeding patients. In per disease analysis, the risk of thrombotic events and stroke was higher in the TXA group than control in intracranial hemorrhage. With regard to seizures, its risk was increased in cardiac surgery and gastrointestinal hemorrhage, and those increase probably came from higher dose of TXA use. In view of the evidence, tranexamic can be safely used in bleeding patients, especially in trauma, obstetrics and gynecology and orthopedic surgery. There may be a dose-dependent increase in the risk of seizures.

Although theoretically, TXA might be expected to increase the risk of thrombotic events due to its antifibrinolytic activity, the present analysis found no evidence in support of this hypothesis. In the early phase of bleeding in trauma or surgery, early fibrinolysis is triggered by the release of tissue plasminogen activator and causes excessive bleeding [28‐31]. Subsequently, plasminogen activator inhibitor-1 level increases and inhibit fibrinolysis [29, 32]. TXA is targeted to modulate early fibrinolysis, and thus it is rational we could not find any increase in thrombotic events in bleeding patients with hyperfibrinolysis even for TXA use.

TXA-associated seizures are caused by another mechanism. TXA is a competitive antagonist of GABA_A in brain. In in vitro and animal experiment, higher TXA level in the cerebral spinal fluid, correlated with serum concentration, was associated with the incidence of seizures [11, 33]. In fact, the results of this meta-analysis were consistent with the basic research reporting dose-dependent increase in the risk of seizures.

A major strength of this analysis is that the present meta-analysis included recently published all the available large-scale randomized controlled trials [5, 6, 12, 18‐22]. Pooling data from 102,681 patients enrolled in 234 randomized controlled trials of tranexamic acid, our study included by far more information than any previous meta-analysis that have addressed this issue [4, 7, 13]. Therefore, this meta-analysis would provide greater evidence regarding safety of tranexamic acid in bleeding populations. In fact, the certainty of evidence evaluated by the GRADE approach was high in thrombotic events with narrow confidence intervals.

However, this meta-analysis also has some limitations. First, all bleeding patients were included in this meta-analysis. Although this significantly increased the pooled information size, it also introduced clinical heterogeneity. However, we performed detailed pre-planned subgroup analysis according to disease type and dose of TXA use and some implication was attained with subgroup analysis. Second, although thrombotic events and seizures did not show any funnel plot asymmetry, there remains a possibility of reporting bias or publication bias due to inadequate information of adverse events in some trials. However, recently published large-scale randomized clinical trials that accounted for most of the weight in this meta-analysis provided detailed data of adverse events. Third, there might be heterogeneity in the diagnostic accuracy for detecting thrombosis adverse in the included trials and any inaccuracy may have biased toward the null of the TXA effects on the outcomes. Fourth, survivor bias may have influenced the results because patients who die are not ‘available’ to suffer a thrombotic adverse event. Although this possibility has not been quantified in individual randomized trials, we can consider the direction of any such bias. Because there were fewer deaths with TXA [5, 6], there would be less opportunity to have a thrombotic event among TXA treated patients. Therefore, our conclusions about safety would not be threatened by survivor bias. Fifth, since there were no available data regarding thrombotic events by anticoagulated status, we could not assess the use of antiplatelet/anticoagulant drugs, which are commonly used in perioperative period and ICU settings, and that could have affected the incidence of thrombosis.

In the previous cumulative systematic review and meta-analysis including surgical bleeding, there was no observed increase in thromboembolic events such as stroke, ACS and VTE [4], although harmful effects were undeniable due to imprecision. Moreover, recent individual patient-data meta-analysis including two large-scale RCTs [5, 6] reported that vascular occlusive events did not increase in acute severe bleeding [7], considered as beneficial population for TXA treatment [34]. As well as the previous reports, the present study showed that TXA did not appear to increase thrombotic events in bleeding populations with high certainty of evidence. On the other hand, the risk of thrombotic events, especially stroke, were higher in TXA group than control group in intracranial (non-traumatic) hemorrhage, which was consistent with previous systematic review and meta-analysis [35, 36]. With regard to the thrombotic risk in females, tranexamic acid did not appear to increase the risk of thrombotic events in patients in obstetrics and gynecology trials.

As for seizures, administration of high dose TXA was associated with increased risk of seizures in cardiac surgery according to previous studies [37, 38]. A recently published large-scale randomized controlled trial (HALT-IT) comparing high dose of TXA (4 g/day) and placebo in gastrointestinal hemorrhage also showed increased risk of seizures. With totaling this evidence, the present study demonstrated dose-dependent increase in seizures with TXA use in bleeding patients. However, TXA did not appear to increase the risk of seizures when administered at a dose of 2 g/day or less.