St Emlyn’s Lesson Plans
Journal club: HALT-IT and treatment transfer
Appraise HALT-IT’s methods, benefits and harms, then discuss what its results mean for GI bleeding care.
Higher Specialty Training · 60-minute session
For learners
Before: Allow about 30 minutes in total.
When we meet: We will discuss the trial’s results, subgroup claims and how to explain a treatment decision to a colleague.
For facilitators
Send: the learner preparation link before teaching.
Bring: The final paper's outcome tables and Figure 3, the linked analysis plan and amendment history, calculator, whiteboard and the local GI-bleed pathway for context.
Before the session
If you are attending this session, complete the core preparation below before you come. Allow about 30 minutes in total.
Core preparation
HALT-IT trial, Lancet 2020 · About 25 minutes
Read the completed trial's methods, participant-flow diagram and main outcome tables. The second case supplies the outcome-change timeline for discussion.
https://doi.org/10.1016/S0140-6736(20)30848-5
Prepare (allow 5 minutes): Write the population, TXA regimen, comparator and primary outcome in four short lines. Add one question about validity or applicability to bring to the group.
Think about: What evidence would you need before using a treatment that works in trauma for GI bleeding?
Picture the scene
A patient has major haematemesis and blood transfusion is underway. A colleague asks whether tranexamic acid should be given because it helps in trauma.
By the end of the session
- Frame the trial’s population, intervention, comparator and outcomes before judging the result.
- Interpret relative effects, uncertainty and adverse events without equating a non-significant result with proof of equivalence.
- Explain why evidence from one bleeding context does not automatically transfer to another.
After the session
- Identify a treatment you use because it works in a related disease; check whether the transfer is supported by direct evidence.
Optional further learning
You do not need to complete this before attending. The listening time is additional to the core preparation above.
Optional listening · 5 min 11 sec
St Emlyn’s Podcast: Ep 91 - Critical Appraisal Nugget 7: Relative risk, Absolute risk and NNT in 5 minutes
Refresh absolute and relative treatment effects before judging the size and uncertainty of a trial result.
Search for this episode by title in MedPod Learn if you want to record your listening there.
Use the HALT-IT paper to express one outcome as an absolute difference and a relative effect, including uncertainty. Explain why a non-significant benefit estimate cannot establish an NNT benefit.
https://www.stemlynspodcast.org/e/can-7-relative-risk-absolute-risk-and-nnt-in-5-minutes/
Facilitator guide
Send learners the preparation link several days before teaching. Allow about 30 minutes in total.
Before learners arrive
- Start with the original published protocol's Primary outcome and Analysis sections. Then read the final paper's Statistical analysis section and Figure 3, the published analysis plan's Primary outcome and Subgroup analyses sections, and the NIHR Methods chapter's November 2018 amendment entry. All four sources are linked below. Check the event-rate calculations and the factual timeline in case two before teaching.
- Ask a gastroenterology or critical-care colleague to clarify current local GI-bleeding care so that journal-club discussion does not substitute for resuscitation training.
Equipment: The final paper's outcome tables and Figure 3, the linked analysis plan and amendment history, calculator, whiteboard and the local GI-bleed pathway for context.
Setup: Give pairs different appraisal tasks. Ask everyone to write a provisional conclusion before discussion and revisit it after the results and harms have been considered.
If the preparation has not been completed
Within the ten-minute methods slot, use eight minutes for pairs to read the abstract, methods summary and outcome tables, followed by two minutes to report the population, comparison and one result. Assign different pairs to mortality and harms. Keep the factual timeline in the second case available. In the 30-minute version, use two minutes for the clinical question and eight for this shared preparation; retain 12 minutes for results, six for application and two for closing.
Suggested session plan
- 0–5 min — State the bedside decision. Keep the patient with haematemesis in mind as the group evaluates the trial.
- 5–15 min — How the trial was done. Ask pairs for PICO, allocation, masking and outcome definitions.
- 15–33 min — Benefits and harms. Calculate or explain absolute and relative effects. Spend most time connecting uncertainty to a decision.
- 33–45 min — Subgroups and outcome history. Use the linked sections and supplied timeline to examine subgroup claims and the outcome amendment; ask which evidence supports each interpretation.
- 45–55 min — Apply and explain. Rehearse the colleague conversation and identify what remains outside this trial’s scope.
- 55–60 min — Close. Each person states what would change their conclusion and what would not.
Start the discussion
Read the opening scenario together. Give everyone a moment to think, then ask what concerns them and what they need to know next.
Cases and discussion
Use these cases, or invite an anonymised case from the group that addresses the same questions.
Should we give TXA now?
A person with severe upper-GI bleeding is receiving active resuscitation. An experienced colleague favours TXA because they have seen it used in other haemorrhage. The group must explain a decision while the team continues standard care.
Add this information as the discussion develops:
- Ask the group to state the clinical question before revealing the study findings.
- Provide the trial’s five-day bleeding-death result: 222/5956 with TXA versus 226/5981 with placebo; RR 0.99, 95% CI 0.82–1.18.
- Provide the venous-thromboembolism result: 48/5952 versus 26/5977; RR 1.85, 95% CI 1.15–2.98.
Ask the group:
- Is this a superiority, non-inferiority or equivalence question?
- What effects remain compatible with the primary confidence interval?
- How does a potential harm signal alter enthusiasm for an unproven benefit?
- How would you explain the decision to a colleague without claiming TXA is harmful in every context?
Discussion notes
Points to draw out:
- HALT-IT randomised 12,009 people with significant GI bleeding. Its tested regimen was a high-dose 24-hour strategy, and it did not show reduced five-day bleeding mortality; venous thromboembolism was more frequent.
- The clinical inference is to avoid routine TXA for GI bleeding based on this evidence. It does not negate indication-specific evidence for trauma or obstetric haemorrhage.
- Distinguish the point estimate from the range of plausible effects. A non-significant primary result does not prove two treatments identical; the practice decision weighs uncertainty, baseline risk, harms and alternatives.
Common misconceptions:
- Saying “TXA works for bleeding” as though haemorrhage is one uniform condition.
- Quoting only the p value and omitting the effect estimate, interval or harms.
For a more experienced group: Calculate approximate absolute event differences from the supplied counts and discuss why an NNT estimate is unstable around a null effect.
Subgroups and an outcome change
One colleague asks whether earlier treatment or a different bleeding population might produce a different result. Another asks how the change in primary outcome affects confidence in the trial. Use the analysis plan and results to examine both questions.
Add this information as the discussion develops:
- Open the published analysis plan's Primary outcome and Subgroup analyses sections, then Figure 3 in the completed trial. The links are in Evidence and guidance.
- Compare the decision, approval and publication dates in the notes below. Ask each pair to identify what the timeline answers and what it leaves open.
Ask the group:
- Was the subgroup specified before analysis, and is there evidence of interaction?
- How many subgroup comparisons were examined?
- When and why was an outcome changed, and was this before unblinded results?
- What additional study would answer the remaining question?
Discussion notes
Points to draw out:
- The trial began with 28-day all-cause mortality as the primary outcome. The completed report dates the decision to use five-day bleeding mortality to 21 November 2018, with 10,190 participants recruited, before unblinding. The NIHR amendment history records ethics approval on 10 April 2019 and MHRA approval on 25 April 2019. The analysis plan was published on 30 July 2019 before unblinding; its publication did not precede completion of recruitment.
- The investigators' rationale was that many deaths were unrelated to bleeding, and treatment of the initial bleed was unlikely to affect much later deaths. Examine the consequences of a narrower outcome, cause-of-death classification and continued reporting of all-cause mortality. A documented pre-unblinding change addresses one concern; it does not end the appraisal.
- Four primary-outcome subgroup factors were prespecified: time to treatment, bleeding site, suspected variceal bleeding/comorbid liver disease, and clinical Rockall score. The final report found no evidence of heterogeneity for these factors (p=0.53, 0.34, 0.94 and 0.32 respectively). This does not prove identical effects or establish a benefit in an early-treatment subgroup.
- Compare each effect estimate and interval in Figure 3, rather than comparing whether separate subgroup p values cross 0.05. Keep the prespecified efficacy analyses distinct from exploratory analyses of thromboembolism. State what further evidence would change the clinical conclusion.
Common misconceptions:
- Treating significance in one subgroup and non-significance in another as proof of different effects.
- Assuming that any outcome amendment either invalidates a trial or requires no further scrutiny.
For a more experienced group: Draft a short proposal for a future study with a defined population and clinically meaningful outcome, and explain why equipoise would exist.
Important clinical points
- Journal club must not delay standard resuscitation, endoscopy planning or other indicated GI-bleeding care.
- Do not extrapolate the GI-bleed result into a blanket prohibition of TXA in other indications.
Closing the session
Use these questions to check what the group will take away. Go back over any clinical points that remain unclear.
- What does RR 0.99 fail to prove?
- Why do absolute event rates matter?
- What makes a subgroup claim credible?
- How would you justify the bedside decision in two sentences?
Key teaching points:
- Keep the trial’s population, regimen and outcomes clear when interpreting its conclusion.
- Consider adverse events and confidence intervals alongside the main effect estimate.
- Evidence for TXA in another condition does not establish a benefit in GI bleeding.
Before leaving, ask each person to choose one action from After the session and say when they will try it. They complete that follow-up in practice, with supervision where needed.
Adapting the discussion
Uncertainty and local practice
The trial answers its tested strategy in its population. Future hypotheses about different regimens or timing should be labelled hypotheses rather than used as current recommendations.
If the discussion drifts
Which result in this trial supports that claim, and would it change what we do for the patient in front of us?
For a more experienced group
Compare the original protocol and final report, then evaluate how outcome selection, ascertainment and competing risks influence interpretation.
A 30-minute version
30-minute option: question 3, methods 7, results and harms 12, application 6, closing 2. Omit the extended subgroup debate.
Evidence and guidance
- HALT-IT trial, Lancet 2020 — Randomised trial of tranexamic acid in significant gastrointestinal bleeding, including mortality and adverse events.https://doi.org/10.1016/S0140-6736(20)30848-5
- Roberts et al. HALT-IT original published protocol (2014) — Read Primary outcome and Analysis for the original in-hospital mortality outcome within 28 days and planned subgroup analyses. Compare with the later analysis plan and final report.https://link.springer.com/article/10.1186/1745-6215-15-450
- Brenner et al. HALT-IT statistical analysis plan (2019) — Read Primary outcome and Subgroup analyses. Published 30 July 2019; the trial reports publication before unblinding.https://link.springer.com/article/10.1186/s13063-019-3561-7
- HALT-IT investigators: NIHR Methods and protocol amendments (2021) — See the November 2017 and November 2018 substantial-amendment entries. Distinguishes the sample-size amendment, outcome-change decision and regulatory approval dates.https://www.ncbi.nlm.nih.gov/books/NBK574635/?report=printable
Suggested curriculum links
RCEM 2021 curriculum (2025 update): SLO10 Research and data
Learners appraise a randomised treatment trial, uncertainty, harms and applicability.
Check these links against the learner’s stage and programme. Attendance at a teaching session does not establish competence.
First published 2020-06-27 · Iain Beardsell.