Article Text
PDF
Abstract
Objectives HALT-IT was an international, randomised trial which assessed the effects of tranexamic acid (TXA) in 12 009 patients with gastrointestinal (GI) bleeding. The results found no evidence that TXA reduces death. It is widely accepted that results of trials should be interpreted in the context of other relevant evidence. We conducted a systematic review and individual patient data (IPD) meta-analysis to assess if the results of HALT-IT are compatible with evidence for TXA in other bleeding conditions.
Design Systematic review and IPD meta-analysis of randomised trials involving ≥5000 patients assessing TXA for bleeding. We searched our Antifibrinolytics Trials Register on 1 November 2022. Two authors extracted data and assessed risk of bias.
Data synthesis We used a one-stage model to analyse IPD in a regression model stratified by trial. We assessed heterogeneity of the effect of TXA on death within 24 hours and vascular occlusive events (VOEs).
Results We included IPD for 64 724 patients from four trials involving patients with traumatic, obstetric and GI bleeding. Risk of bias was low. There was no evidence for heterogeneity between trials for the effect of TXA on death or for the effect of TXA on VOEs. TXA reduced the odds of death by 16% (OR=0.84, 95% CI: 0.78 to 0.91, p<0.0001; p-heterogeneity=0.40). In patients treated within 3 hours of bleeding onset, TXA reduced the odds of death by 20% (0.80, 0.73 to 0.88, p<0.0001; p-heterogeneity=0.16). TXA did not increase the odds of VOEs (0.94, 0.81 to 1.08, p for effect=0.36; p-heterogeneity=0.27).
Conclusions There is no evidence for statistical heterogeneity between trials assessing the effect of TXA on death or VOEs in different bleeding conditions. When the HALT-IT results are considered in the context of other evidence, a reduction in the risk of death cannot be discounted.
Trial registration number PROSPERO CRD42019128260.Cite Now
- gastroenterology
- trauma management
- obstetrics
- statistics & research methods
- haematology
Data availability statement
Data are available in a public, open access repository. The trial data included in this analysis are available for download from the FreeBIRD data repository (freebird.lshtm.ac.uk).
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
STRENGTHS AND LIMITATIONS OF THIS STUDY
This is an individual patient data (IPD) meta-analysis that assesses how the evidence for the effects of tranexamic acid in patients with gastrointestinal bleeding (compares with the evidence from other acute bleeding conditions.
Allows the results of the HALT-IT trial to be interpreted in the context of other high-quality evidence.
This analysis includes IPD for 64 724 patients from four high-quality randomised trials with minimal missing data.
There are important bleeding conditions that are not represented in the analysis such as surgical bleeding.
Introduction
Tranexamic acid (TXA) reduces bleeding by inhibiting the breakdown of fibrin blood clots. It has been used for many years to reduce heavy menstrual bleeding and bleeding during surgery. More recently, it was shown to improve outcomes in bleeding patients with trauma. The CRASH-2 trial involving 20 211 patients with trauma, showed that TXA reduces the risk of death due to bleeding by one-third when given soon after injury, with no apparent increase in side effects.1 This finding raised the possibility that TXA could be effective for treating patients with other life-threatening bleeding conditions and several large trials were initiated in response, including the HALT-IT (Haemorrhage Alleviation With Tranexamic Acid - Intestinal System) trial.
The HALT-IT trial was a large international randomised controlled trial assessing the effects of TXA in patients with gastrointestinal (GI) bleeding. An earlier Cochrane Systematic Review included seven small trials of TXA in patients with upper GI bleeding, and suggested that TXA could reduce the risk of death from any cause by 40% (RR: 0.60, 95% CI: 0.42 to 0.87).2 However, the estimate was imprecise and the methodological quality of the included trials was poor. There were also questions about the applicability of the evidence to current day patients and the effect on risk of vascular occlusive events (VOEs) was unknown. The HALT-IT trial was initiated with the aim of addressing these uncertainties.
Between 2013 and 2019, 12 009 patients with acute severe GI bleeding were recruited to the HALT-IT trial from 164 hospitals in 15 countries. Patients were randomly assigned to receive TXA (1 g loading dose followed by 3 g maintenance dose over 24 hours) or matching placebo. Outcome data from over 99% patients were included in the final analysis. The results were published in June 2020.3 There was no difference in the risk of death due to bleeding between TXA treated patients and those given placebo (3.7% vs 3.8%; RR: 0.99; 95% CI: 0.82 to 1.18).
HALT-IT was a high-quality trial that was not subject to the methodological weaknesses of the previous trials of TXA in GI bleeding. Moreover, it involved a heterogeneous group of patients who are representative of the patients with GI bleeding seen in current day practice. Based on the HALT-IT trial, we can therefore confidently discount the results of the meta-analysis of the previous small trials of TXA in patients with GI bleeding which are likely to be explained by bias.
It is widely recognised that the results of individual trials should be interpreted in the context of the existing evidence, including an assessment of how similar the results are to those of other trials in the same topic area.4 Rather than consider the results of the HALT-IT trial in the context of the methodologically flawed previous trials in GI bleeding, it is arguably more useful to consider how the results compare with evidence from the other high-quality trials of TXA that have been conducted in different acute bleeding conditions.
Objectives
To assess if the results of the HALT-IT trial are statistically compatible with the evidence for the effects of TXA in other acute bleeding conditions.
Methods
Study design
We analysed individual patient data (IPD) from large randomised trials of TXA in patients with acute severe bleeding. The London School of Hygiene & Tropical Medicine’s CTU Global Health Trials Group maintains a register of completed and ongoing trials assessing the effects of antifibrinolytic drugs. The register contains records of trials identified through searches of the MEDLINE (Ovid MEDLINE(R) ALL 1946 to 1 November 2022), CENTRAL (searched on 1 November 2022), EMBASE (Ovid EMBASE 1980 to 2022 week 44) and WHO ICTRP (searched on 1 November 2022) electronic databases. Each database was searched using a combination of subject headings and keywords (online supplemental material 1). We searched the register for records of trials assessing the effect of TXA on death and VOEs in patients with acute severe bleeding. An initial search of the register was conducted to 31 May 2021 and subsequently updated on 1 November 2022. One author (KK) selected trials involving 5000 patients or more, that had been prospectively registered, and were judged to be at low risk of bias for sequence generation, allocation concealment and blinding of outcome assessment. Two authors (KK and RM) extracted data from the included trials. Study level data on the inclusion criteria, TXA regimen and risk of bias domains were extracted from the study reports. IPD were obtained for the following variables: age, time of bleeding onset, time trial treatment initiated, systolic blood pressure (SBP), treatment allocation, status at 24 hours (dead/alive) and VOEs during follow-up. We checked IPD against the published trials’ results to check for completeness and accuracy. Trials were assessed as being at low, unclear or high risk of bias for random sequence generation (selection bias); allocation concealment (selection bias); blinding of participants, personnel (performance bias); blinding of outcome assessors (detection bias); incomplete outcome data (attrition bias); and selective outcome reporting (reporting bias) according to Cochrane criteria.5 Only trials judged to be at low risk of bias for random sequence generation, allocation concealment and blinding of outcome assessment were included.
Supplemental material
This study was undertaken as part of the work of the Antifibrinolytics Trialists Collaboration. We followed the methods prespecified in the PROSPERO record CRD42019128260, with the exception of limiting this analysis to TXA trials involving 5000 patients or more.
Patient and public involvement
Patient and public representatives were not involved in the design of this IPD analysis.
Outcomes
We assessed the homogeneity of the effect of TXA on death as well as fatal and non-fatal VOEs (myocardial infarction, stroke, pulmonary embolism and deep vein thrombosis). For deaths, we limited the analysis to only those that occurred within 24 hours of bleeding onset. Due to its mechanism of action as an inhibitor of fibrinolysis, we expect TXA to only have an effect on deaths due to bleeding, the large majority of which occur in the first 24 hours. Furthermore, TXA has a short half-life (about 2 hours) and is largely eliminated within 48 hours. Focussing on deaths occurring in the first 24 hours helps us to ensure that we include only those events which are most likely to be affected by the trial treatment, and we avoid the effect of dilution bias that would result from the inclusion of non-bleeding deaths. However, for VOEs we included all events occurring at any point during the follow-up of the trials.
Data analysis
Data on all randomised patients were included regardless of whether or not they received the trial treatment (ie, on an intention-to-treat basis).
We used a one-stage model to conduct the IPD meta-analysis. The one-stage approach analyses IPD in a single meta-analysis based on a regression model stratified by trial. This approach allows for the investigation of within and between trial variances and estimation of the effect of TXA in a single analytical model.
To assess the heterogeneity of treatment effects, we included data from all patients irrespective of the time interval between bleeding onset and randomisation. However, previous analyses of IPD data show an apparent time to treatment effect on risk of death, with TXA most effective when given early (ie, within 3 hours) of bleeding onset, with no effect when given beyond this period.6 Therefore, for deaths, we also assessed the heterogeneity of treatment effects for the subset of patients treated within 3 hours of bleeding onset. To assess the heterogeneity of the treatment effects between trials, we included an interaction term between the treatment and the trial variable. The analyses were controlled for SBP, age and time to treatment. The p values for heterogeneity of the treatment effects between trials were obtained from a likelihood ratio test of the trial treatment effect interaction term in an adjusted logistic regression model. We considered a p value <0.05 to indicate the presence of statistical heterogeneity. We used a logistic regression model to calculate adjusted ORs and 95% CIs. Further details of the statistical methods are given in online supplemental material 2. We used Stata vV.16.1 for all analyses (StataCorp, College Station, Texas, USA).
Quality of the evidence
We used the GRADE (Grading of Recommendations, Assessment, Development and Evaluations) approach to assess the certainty of evidence for each outcome. We considered the impact of the risk of bias of individual trials, precision of pooled effect estimate, heterogeneity between trials, indirectness of the evidence and impact of reporting biases on the pooled effect estimate. The evidence for each outcome was rated as high, moderate, low or very low.
Results
Description of trials
Through the initial search, we identified four completed trials published between 2010 and 2020 that assessed the effects of TXA for the treatment of acute severe haemorrhage involving 5000 or more patients; the CRASH-2,1 7 WOMAN,8 CRASH-39 and HALT-IT3 trials (online supplemental material 3). All IPD for these trials are freely available alongside their data dictionaries from the FreeBIRD data repository (freebird.lshtm.ac.uk).10–13 The subsequent update of the search in November 2022, identified one further trial, the POISE-3 (PeriOperative ISchemic Evaluation-3) trial14 published in May 2022, however due to the time required to obtain and process IPD from trials, it was not feasible to include the IPD in this analysis. We also identified a further seven potentially eligible ongoing trials of TXA involving patients with surgical bleeding,15 16 obstetric bleeding,17–19 mild traumatic brain injury20 and haemorrhagic stroke.21 Details of the included and ongoing trials are presented in online supplemental material 4.
Each of the four included trials for which IPD were available, assessed the effects of intravenous TXA in different acute haemorrhagic conditions. The CRASH-2 (Clinical Randomisation of an Antifibrinolytic in Significant Haemorrhage) trial assessed the effects of TXA in 20 211 adult patients with trauma with, or at risk of, significant bleeding who were recruited in 274 hospitals in 40 countries. Patients were randomly assigned to receive TXA (1 g loading dose then infusion of 1 g over 8 hours) or placebo as soon as possible after, and within 8 hours of, injury. Patients were followed up until death, discharge or until 28 days after injury, whichever occurred first. The outcomes included death, VOEs, blood transfusion and disability. IPD were available for 20 127 of all randomised patients.
The WOMAN (World Maternal Antifibrinolytic) trial involved 20 060 women with postpartum haemorrhage who were recruited in 193 hospitals in 21 countries. Women were randomly assigned to receive 1 g TXA or placebo. If bleeding continued after 30 min or stopped and restarted within 24 hours of the first dose, a second dose of 1 g TXA or placebo could be given. Women were followed up until death, discharge or until 42 days after birth, whichever occurred first. The outcomes included death, hysterectomy, surgical intervention, VOEs, blood transfusion and quality of life. IPD were available for 20 021 of all randomised women.
The CRASH-3 (Clinical Randomisation of an Antifibrinolytic in Significant Head Injury) trial involved 12 737 adults with traumatic brain injury who were recruited in 175 hospitals in 29 countries. Patients were randomly assigned to receive TXA (1 g loading dose hours) or placebo. Patients were followed up until death, discharge or until 28 days after injury, whichever occurred first. The outcomes included death, VOEs, neurosurgery, complications and disability. IPD were available for 12 639 of all randomised patients.
The HALT-IT (Haemorrhage Alleviation With Tranexamic Acid - Intestinal System) trial involved 12 009 adults with upper or lower GI bleeding who were recruited in 164 hospitals in 15 countries. Patients were randomised to receive TXA (1 g loading dose then infusion of 3 g over 24 hours) or placebo as soon as possible after bleeding onset. Patients were followed up until death, discharge or until 28 days after bleeding onset, whichever occurred first. The outcomes included death, VOEs, surgery or radiological intervention, blood transfusion, complications and functional status. IPD were available for 11 937 of all randomised patients.
Together 65 017 patients were included in the four trials; of these, we included individual patient-level data for 64 724 (>99%) participants (table 1). Of the 64 724 patients for who IPD were available, 32 411 were randomly allocated to receive TXA and 32 313 to receive placebo, and 39 466 (61%) patients were treated within 3 hours of bleeding onset. All of the trials collected data on death and VOEs.
Baseline characteristics of patients contributing data to the IPD meta-analysis
There were differences between the characteristics of patients included in the HALT-IT trial and those included in the other three trials (online supplemental material 5). On average, patients in the HALT-IT trial were almost 24 years older and were treated 9 hours later, than patients in the other three trials. Average SBP was also slightly higher in HALT-IT trial patients, although there was little difference in SBP among the subset of patients treated within 3 hours.
The included trials were judged to be at low risk of bias for all domains (online supplemental material 6). The randomisation sequence was computer-generated and allocation was adequately concealed in all trials. Participants, caregivers and outcome assessors were blind to allocation status in all trials and there were minimal missing data (<1%). There were no issues concerning the completeness or integrity of IPD from any trial.
Heterogeneity of the effect of TXA between trials
Outcome data for patients contributing IPD from the included trials are presented in online supplemental material 7.
We found no heterogeneity between trials for the effect of TXA on death within 24 hours (figure 1). Overall, there were 2792 (4%) deaths within 24 hours. The combined estimate from the analysis, including all patients irrespective of time since bleeding onset, suggests that TXA reduces the odds of death by 16% compared with placebo (0.84, 0.78 to 0.91; p<0.0001). The p value for heterogeneity between trials was 0.40. The certainty of evidence was judged to be moderate for this outcome (online supplemental material 8).
Results of IPD meta-analyses of the effect of TXA on death within 24 hours in (A) all patients and (B) patients treated within 3 hours of bleeding onset. The effect estimates are adjusted for age, systolic blood pressure and time to treatment. IPD, individual patient data; TXA, tranexamic acid.
Among patients treated 3 hours or sooner after bleeding onset there were 2085 (5%) deaths within 24 hours. In these patients, TXA reduces the odds of death by 20% compared with placebo (0.80, 0.73 to 0.88; p<0.0001). The p value for heterogeneity was 0.16. The certainty of evidence was judged to be moderate for this outcome (online supplemental material 8).
Overall, 795 (1%) patients suffered a fatal or non-fatal vascular occlusive event.
We found no heterogeneity between trials for the effects of TXA on VOEs (p value for heterogeneity=0.27). The combined estimate suggests that TXA does not increase the odds of a vascular occlusive event compared with placebo (0.94, 0.81 to 1.08; p=0.37).
Table 2 and online supplemental material 9 shows the results for the effect of TXA on VOEs by type. We found no heterogeneity between trials for effect of TXA on VOEs when considered separately. The combined estimates suggest that there is no increase in odds of myocardial infarction, stroke, deep vein thrombosis or pulmonary embolism in patients treated with TXA. The certainty of evidence was judged to be low for all VOEs outcomes (online supplemental material 8).
Effect of TXA on VOEs
Discussion
We found no strong statistical evidence that the effects of TXA on death or VOEs varies between different acute severe bleeding conditions. Our pooled estimate suggests that TXA reduces the odds of death within 24 hours by 16%. The estimate increases to 20% when TXA is given within 3 hours of bleeding onset. The quality of evidence was judged to be moderate for the mortality outcomes. There was no evidence for an increased risk in VOEs associated with TXA, however these estimates are imprecise; being compatible with both an increase and decrease in risk, and the quality of the evidence was judged to be low for these outcomes.
Strengths and limitations
We limited our analysis to large randomised controlled trials at low risk of bias (ie, prospectively registered, adequately concealed and blinded outcome assessment), and we obtained individual patient-level data from four out of five trials meeting our inclusion criteria, which constitutes about 90% of the existing data. All four trials were judged to be at low risk of bias for all domains and together provided data on 64 724 patients with minimal (<1%) missing data. We can therefore be confident that they provide reliable estimates of the effects of TXA and that our results are unlikely to be explained by bias. Each trial assessed the effects of TXA in a different bleeding condition, enabling our analysis to explore heterogeneity across several acute bleeding scenarios. However, there are important bleeding conditions that are not represented in our analysis such as surgical bleeding. The recently completed POISE-3 trial14 assessed the effects of TXA in surgery patients as are several of the potentially eligible ongoing trials we identified. Repeating our analyses when IPD from these trials are available would provide further insight into the effects of TXA across acute severe bleeding conditions.
Implications
We found no statistical evidence that the effect of TXA on death or VOEs varies between bleeding conditions, thus the results of the HALT-IT trial appear to be statistically compatible with the evidence from trials in other acute bleeding conditions. Assuming that there is no true biological or statistical heterogeneity, the pooled estimate from a meta-analysis of data from high-quality randomised trials is the most reliable guide to the approximate effect estimate in patients with GI bleeding. Under this assumption, we cannot discount the possibility that early treatment with TXA reduces the risk of death due to GI bleeding by 20%. This reduction is considerably smaller than that estimated by the meta-analysis of previous trials and is smaller than the difference the HALT-IT trial was powered to detect.
However, absence of evidence does not equate to evidence of absence and we cannot say with certainty that the effects of TXA do not differ between bleeding conditions, only that we found no evidence for it in our analysis. HALT-IT was the only included trial not to observe a statistically significant reduction in the risk of death, indeed there were more deaths among patients receiving TXA than those receiving placebo. There are differences in the clinical presentation of patients with GI bleeding which may be important. Compared with patients with other bleeding, those with GI bleeding tend to be much older with comorbidities and present for treatment much later, often beyond the 3-hour time period during which TXA is most effective. Furthermore, we cannot exclude the possibility that there is variation in the physiology of bleeding which may mean that the effects of TXA vary between patients with different bleeding conditions, and specifically that the effects are different in patients with GI bleeding. Indeed, recent research shows that patient with variceal bleeding due to liver disease have a mixed fibrinolytic profile, with some having increased fibrinolysis but others having profound hypofibrinolysis.22 Further research into the fibrinolytic responses of patients with acute severe bleeding is required to better understand the risk–benefit profile of TXA in patients with GI bleeding. Without such additional insight, TXA should not be recommended as a treatment for GI bleeding.
The HALT-IT trial is the largest trial in patients with GI bleeding conducted to date, yet it was insufficiently powered to detect the modest effect of TXA on death observed by this meta-analysis. Despite the methodological shortcomings of the previous trials in patients with GI bleeding, the large relative reduction in all-cause mortality observed by the meta-analysis of these trials is difficult to ignore and may ultimately have contributed to the underpowering of the HALT-IT trial. When setting the target difference for a sample size calculation, investigators are encouraged to consider evidence from previous trials alongside expert opinion to specify the magnitude of difference in outcome between treatment groups that would be both clinically important and realistic to detect.23 For the CRASH-2, WOMAN and CRASH-3 trials, there were no previous trials of TXA in their respective bleeding conditions, thus the target differences for these trials were based solely on expert clinical opinion of plausible treatment effects. However, for the HALT-IT trial there was evidence from the systematic review of previous trials to also consider. This could have contributed to the underpowering of the HALT-IT trial by influencing clinical opinion about the potential magnitude of the treatment effect, which had been grossly exaggerated by the previous trials. Even more fundamentally, this flawed evidence almost prevented the HALT-IT trial from taking place at all, with many, including the funders and reviewers of the protocol, questioning whether the trial was needed in light of the existing evidence. Based on this response it seems unlikely that prospective funders would have looked favourably on a proposal for a larger trial, powered to detect a more modest, but realistic effect. However, given the potentially important clinical and physiological differences outlined above, it cannot be assumed that an adequately powered trial would observe a beneficial treatment effect.
In 2015, we published an article arguing that systematic reviews of small unreliable trials, increase waste by promoting underpowered trials.24 The experience of the HALT-IT trial seems to support this view. We encourage investigators to employ caution when using results of systematic reviews based on small studies to inform the design of new trials.
Conclusions
In the HALT-IT trial, there was no significant reduction in bleeding deaths with TXA. When the null hypothesis is that TXA has no effect on deaths, the trial results are consistent with it. When we know little about the effects of a drug, a null hypothesis of no effect seems reasonable. However, there is strong evidence that TXA reduces bleeding deaths after trauma and postpartum haemorrhage. In the light of this, it may be reasonable to use the alternative hypothesis that TXA cuts GI bleeding deaths to a similar extent to that seen in trauma and postpartum haemorrhage. When we do this, we find no evidence that the effect of TXA on GI bleeding deaths is different from that in these other bleeding scenarios. Our conclusion after considering the two different perspectives is that while there is no evidence that TXA reduces GI bleeding deaths, we cannot rule out the possibility that it does. However, based on current evidence TXA should not be recommended as a treatment for GI bleeding.
Data availability statement
Data are available in a public, open access repository. The trial data included in this analysis are available for download from the FreeBIRD data repository (freebird.lshtm.ac.uk).
Ethics statements
Patient consent for publication
Ethics approval
We judge that separate institutional review board approval for this analysis is not required. This study involves the analysis of existing trial data. Each trial providing individual patient data received local ethical approval. All patients contributing data to this analysis provided consent for their participation in the original trials. This original consent included approval for the use of their anonymised data in future research studies. Participants gave informed consent to participate in the study before taking part.
Acknowledgments
We are extremely grateful to the many trial collaborators based throughout the world who contributed to the CRASH-2, WOMAN, CRASH-3 and HALT-IT trials. This article was prepared on behalf of the Antifibrinolytics Trialists Collaboration (ATC) which aims to increase knowledge about the effects of antifibrinolytics. We are grateful to all members of the ATC who contribute to its overall objectives.
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
Contributors KK: designed the methods; conducted the bibliographic searches; identified eligible trials; extracted data; assessed risk of bias; interpreted the results; and wrote the final manuscript. Member of the HALT-IT trial Protocol Committee. KK is the study guarantor. RM: designed the methods; conducted the statistical analyses; contributed to the preparation of the final manuscript. HALT-IT trial statistician. HS-S: conceived the study; coordinated the generation of the trial data; contributed to the preparation of the final manuscript. Project Director of the HALT-IT trial. MA: coordinated the generation of the trial data; contributed to the preparation of the final manuscript. Trial Manager of the HALT-IT trial. DB: coordinated the generation of the trial data; contributed to the preparation of the final manuscript. Senior Trial Manager of the HALT-IT trial. IR: conceived the study; designed the methods; interpreted the results; contributed to the preparation of the final manuscript. Chief Investigator of the HALT-IT trial.
Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests None declared.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.