3 misunderstandings about WOMAN (TXA in postpartum hemorrhage)

I have addressed the WOMAN trial before on First10EM. (Twice actually). Two mentions of a trial of TXA in postpartum hemorrhage might already seem a bit excessive for an emergency medicine blog. However, this trial has been discussed widely, and there seems to be some persistent confusion about the results of WOMAN. Therefore, seeing as I was updating my post on postpartum hemorrhage this week, I thought I would address the 3 most common misunderstandings about WOMAN. 

The paper

WOMAN trial collaborators . Effect of early tranexamic acid administration on mortality, hysterectomy, and other morbidities in women with post-partum haemorrhage (WOMAN): an international, randomised, double-blind, placebo-controlled trial. Lancet (London, England). 2017; PMID: 28456509 [free full text]

The misunderstandings

#1 That it was a positive trial

It wasn’t. It was a negative trial. I know that will sound strange, because every time someone mentions this trial – from the WHO to NPR – it is touted as a resounding success. The headlines scream “Overlooked Drug Could Save Thousands Of Moms After Childbirth”, or “Cheap lifesaver ‘cuts deaths by a third”. That is simply not true. 

I can’t blame the news outlets for getting this wrong. Even the authors spun this paper, leading with the secondary outcome in both the discussion section and the abstract. (This is incredibly poor editing and peer review on behalf of the Lancet.)

But the trial was negative. The primary outcome was a composite of mortality and hysterectomy. It occurred in 5.3% of the TXA group and 5.5% of the placebo group (p=0.65). If you want to focus on just the most important outcome, all cause mortality was unchanged (2.3% vs 2.6%, p=0.16).

So this was a large, generally well done study, that was clearly negative.  

#2 That disease specific mortality has a clear meaning

Disease specific mortality is among the most confusing and misleading concepts in evidence based medicine. It sounds important. It sounds like a patient oriented outcome. But it isn’t.

All of the crazy headlines about the WOMAN trial hinge on one secondary outcome: the fact that death due to bleeding was statistically reduced by 0.4% (1.5% vs 1.9%; p=0.045; risk ratio 0·81, 95% CI 0·65–1·00). The difference here is small, and far from certain. If you look at the risk ratio, the 95% confidence interval touches 1 (0·65–1·00) which is consistent with non-significant. The fragility index is 0, which means that if 0 patients had a different result, the outcome would be statistically non-significant. (Yes, that is a very weird thing to say. When I calculate the P value myself, it comes out as 0.051, so all the fuss is about an outcome that is actually non-significant. I am not sure why there is a discrepancy.) Considering that this is only a secondary outcome, the incredibly borderline stats increase that likelihood that this is not a real finding, but simply “significant” by chance alone.

However, the bigger problem is with the use of “death due to bleeding” as an outcome at all. Disease specific mortality is frequently used as a surrogate in situations where it is impossible to run trials big enough to show a change in all cause mortality. It is easier to show a change in disease specific mortality – you don’t need as large a trial. The assumption (and it is often a poor assumption) is that if disease specific mortality is improved, all cause mortality will also be improved.

However, WOMAN is not a small trial. They had more than 20,000 patients. It was actually powered to see a change in all cause mortality (after some on the fly adjustments), and there was no difference. We don’t need the surrogate. 

Furthermore, disease specific mortality is a subjective and fallible concept. In most studies, including WOMAN, the cause of death is simply determined by the treating physician. There is no autopsy. Think about your own practice: how often are you absolutely sure of a patient’s cause of death? Now add in an experimental drug. Hemorrhagic shock can look a lot like obstructive shock. How do we know these patients didn’t die from massive PEs? There is a real chance that any difference in disease specific mortality is just an illusion created by the subjective and inaccurate process of completing death certificates. (Now consider the fragility index of 0. How confident are you in this outcome?)

More importantly, disease specific mortality is not a patient oriented outcome. It is a statistician’s outcome. It is worse than meaningless – it is misleading.

Consider a theoretical study of postpartum hemorrhage. I have decided to market “agent X”, and publish a large RCT demonstrating that my new drug decreases “death due to bleeding” from 2% to 0% with a p value <0.000001. Would you buy the drug? 

Before you do, I would suggest that you ask me about all cause mortality. Because the drug I am trying to sell is cyanide. It is incredibly good at decreasing death due to bleeding. If given in the right dose, not a single person dies of bleeding (although 100% die of cyanide poisoning).

Although this example is somewhat over the top, it demonstrates how misleading disease specific mortality can be. Real life examples are much subtler. Cancer screening is often said to decrease cancer specific morality, but a closer look reveals no change in all cause mortality. The discrepancy might just be an accounting error (different things were written on death certificates, with no real difference), but it could also result from real changes. Maybe some people were saved from cancer, but an equal number of people died from complications of surgery or chemotherapy. Either way, at the end of the day, the same number of people are dead. We can tell our patients: with this therapy, in 1 year you are just as likely to die, but I will get to write something different on the death certificate. Do you think many patients will care? (OK, some deaths are worse than others, and some changes might be worth it – but that is never the question being asked).

Disease specific mortality is subjective. It is error prone. It is misleading. It shouldn’t be used in situations where the all cause mortality is known. And WOMAN gives us a very good idea that all cause mortality is unchanged. 

#3 That a negative trial proves TXA doesn’t work

There are many reasons that trials can be negative despite a true treatment effect. In WOMAN, I wonder whether the patients studied were simply too healthy to benefit from TXA. Inclusion was based on blood loss (more than 500 mL after vaginal delivery and 1000 mL after C-section), but more than half of the patients had less than 1000 mL of blood loss. I don’t expect those patients to die (which is reflected in the 2-3% mortality). In fact, more recent definitions, such as the one used by ACOG, don’t even count blood loss of less than 1000 mL without vital sign changes as postpartum hemorrhage. (ACOG 2017) We were looking for a change in all cause mortality, but 97% of the patients in this trial were going to live no matter what we did. TXA could not possibly help. Thus, a real benefit in the sickest patients might have been washed out by the inclusion of a large number of healthy patients. Unfortunately, despite all the attention given to secondary analyses and subgroups, this paper is lacking the one subgroup I really care about: how did TXA fare in the sickest patients?

Maybe WOMAN studied a population that was too healthy. Maybe TXA benefits the sickest of the sick. However, that is pure conjecture, and the current best answer from WOMAN was that TXA did not help. 

So what should we do?

First, we should probably consider other available evidence. The WOMAN trial is not the only trial looking at TXA in postpartum hemorrhage. Ducloy-Bouthors et al (2011) performed an open label RCT of high dose TXA (4 grams over 1 hour, then 1 gram an hour for 6 hours) in 144 women with PPH and more than 800 mL of blood loss. The trial was statistically positive, but the actually outcomes were not impressive, with only a 48 mL difference in total blood loss between the groups. Ayedi et al (2011) randomized 51 patients with postpartum hemorrhage that continued despite oxytocin to TXA or placebo. (The study is only available as an abstract.) There were no deaths. There was a 600 mL difference in blood loss between the groups, but the result wasn’t statistically significant. Another small trial demonstrated no difference between TXA and misoprostol, (Sahar 2014) although the evidence for misoprostol is also somewhat imperfect. (Derman 2006; Tunçalp 2012) So, there really aren’t any other great studies suggesting we use TXA in PPH.

You might want to extrapolate from the use of intravenous TXA in other bleeding populations. We have all become enamoured with TXA since the publication of CRASH-2 demonstrated a 1.5% decrease in absolute mortality. (CRASH-2 2010) However, there were reasons to be skeptical of the results seen in CRASH-2, including the fact that most patients were enrolled in settings without advanced trauma systems, possible selection bias, the fact that there was no change in transfusions (so the biologic plausibility might be questioned), and questions of whether harms would have been adequately captured in this study. So although we all use TXA, we should remember its benefit remains somewhat uncertain. 

The TICH-2 trial looked at TXA in intracranial hemorrhage and showed no benefit. (Sprigg 2018) On the other hand, there may be some evidence of benefit in GI bleed, although the quality of the trials if poor. (Bennet 2014) There is also some evidence of benefit in hemoptysis, epistaxis, and post-tonsillectomy bleeding, although again the trials are all small and have significant limitations. (Wand 2018; Joseph 2018; Schwartz 2019)

So, overall, the evidence for TXA is mixed. I doubt it is the wonder drug that many make it out to be, but it definitely seems to have some physiologic effects, and the harms appear to be minimal.

Overall, my guess is that TXA doesn’t help in postpartum hemorrhage, but I can’t be sure. The purest interpretation of the literature is probably that we shouldn’t be giving TXA. However, TXA appears to be pretty safe, and this data does allow for a small, but potentially important benefit with TXA. It definitely shouldn’t be first line, and it may not work at all, but I might still reach for TXA in the sickest PPH patients.

Other FOAMed

SGEM 214: WOMAN: TXA in postpartum hemorrhage

The Bottom Line

BroomeDocs: Thoughts on the WOMAN Trial

REBEL EM: The WOMAN Trial: Early TXA in Postpartum Hemorrhage

References

ACOG. Practice Bulletin No. 183: Postpartum Hemorrhage. Obstetrics and gynecology. 2017; 130(4):e168-e186. [pubmed]

Ayedi M, Jarraya A, Smaoui M, Zouari J, Smaoui L, Kolsi K. Effect of tranexamic acid on post partum hemorrhage by uterine atony: A preliminary result of a randomized, placebocontrolled trial. European Journal of Anaesthesiology. 2011;28:165.

Bennett C et al. Tranexamic Acid for Upper Gastrointestinal Bleeding (Review). Cochrane Database Syst Rev 2014. PMID: 25414987

CRASH-2 trial contributors, Shakur H, Roberts I, et al. Effects of tranexamic acid on death, vascular occlusive events, and blood transfusion in trauma patients with significant haemorrhage (CRASH-2): a randomised, placebo-controlled trial. Lancet (London, England). 2010; 376(9734):23-32. [pubmed]

Derman RJ, Kodkany BS, Goudar SS, Geller SE, Naik V, Bellad MB, et al. Oral misoprostol in preventing postpartum haemorrhage in resource-poor communities: a randomised controlled trial. Lancet. 2006;368:1248–53.

Ducloy-Bouthors AS, Jude B, Duhamel A, et al. High-dose tranexamic acid reduces blood loss in postpartum haemorrhage. Critical care (London, England). 2011; 15(2):R117. [pubmed]

Joseph J et al. Tranexamic Acid for Patients with Nasal Haemorrhage (epistaxis) (Review). Cochrane Database Syst Rev 2018. PMID: 30596479

Sahhaf F, Abbasalizadeh S, Ghojazadeh M, Velayati A, Khandanloo R, Saleh P, Piri R, Naghavi-Behzad M. Comparison effect of intravenous tranexamic acid and misoprostol for postpartum haemorrhage. Niger Med J 2014;55:348-53

Schwarz W et al. Nebulized Tranexamic Acid Use for Pediatric Secondary Post-Tonsillectomy Hemorrhage. Ann Emerg Med 2019. PMID: 30292524

Sprigg N, Flaherty K, Appleton JP, et al. Tranexamic acid for hyperacute primary IntraCerebral Haemorrhage (TICH-2): an international randomised, placebo-controlled, phase 3 superiority trial. Lancet (London, England). 2018; 391(10135):2107-2115. PMID: 29778325 [free full text]

Tunçalp Ö, Hofmeyr GJ, Gülmezoglu AM. Prostaglandins for preventing postpartum haemorrhage. The Cochrane database of systematic reviews. 2012; [pubmed]

Wand O, Guber E, Guber A, Epstein Shochet G, Israeli-Shani L, Shitrit D. Inhaled Tranexamic Acid TXA for Hemoptysis Treatment: A Randomized Controlled Trial. Chest. 2018; 154(6):1379-1384. PMID: 30321510

WOMAN trial collaborators . Effect of early tranexamic acid administration on mortality, hysterectomy, and other morbidities in women with post-partum haemorrhage (WOMAN): an international, randomised, double-blind, placebo-controlled trial. Lancet (London, England). 2017; PMID: 28456509 [free full text]

TXA in postpartum hemorrhage