Hopeless.  Not the drug – the research

The clue is in the journal. Critical Care, one where authors pay to get their articles in print.   No, there’s no mistake – the author, who does the work, formats the article and hands over copyright, pays $1,825 to the publisher, who makes money from it.  It’s not even printed on paper.  Just published on the web.  We used to call that sort of thing “Vanity Publishing”.   The modern label is “author pays”, or for enthusiasts “open access”.  I’ll be writing more about this.

No-one should be surprised that the quality of articles published this way is poor.  Why would anyone pay to get their good stuff published, if a decent journal like the BMJ or Lancet would take it for free.  But it got published.  Read the full paper here.

It is a randomised trial.  Two groups of patients who had just delivered their baby and were bleeding heavily, were randomly allocated, half to tranexamic acid (TXA) a drug which slows down clot breakdown, and half not.  But there were problems.

The trial was open; the doctors knew who got the TXA, and the primary endpoint was blood loss, which is notoriously difficult to measure.  However carefully you collect it in bags and weigh the swabs, you still have to estimate how much is blood and how much amniotic fluid. The authors claim that the staff measuring the blood loss were unaware whether the patient had received TXA, but this is implausible.  The doctors who recruited and randomised the patients and administered the drug to half of them were in the same room.  How could they not know what was going on?  The authors admit that the blood loss measurements are susceptible to bias.

The time period over which blood loss was measured is also a source of potential bias? The authors state that it was between what they call T1 (time point 1) and two hours later. T1 is reported as time of delivery in figure 1, but as time of inclusion in the text.  Does time of inclusion mean time of randomisation, or of drug administration? Presumably the former, but the authors have just told us that the people measuring the blood loss were unaware of the treatment allocation.  How could they know that randomisation had happened and it’s timing, but not then be able to see the drug being administered.  They need to know it in real time (see below).

I know what it is like when a patient bleeds heavily immediately after delivery – controlled chaos as the baby is passed to anxious relatives, extra staff are called, intravenous lines are sited, drugs to contract the uterus administered, the placenta checked to ensure it’s all out, and the vagina inspected for obvious bleeding vessels.  It is inconceivable that T1 is recorded, and the staff stop to measure the blood loss up to that point, and subsequently separately.  They must usually just make a guess when its all over.   If they occasionally timed their guess at T1 at the drug administration the error would be understandable, but it would have the effect of recording blood loss over a shorter time period in the treatment group because the drug was only administered in that group.

One obvious solution would be to report the total blood loss after the birth.  All the swabs could be carefully weighed, and all the blood measured, with no subjective judgement needed as to which came before, and which after T1/enrolment.  And since bleeding prior to T1 was greater in the TXA group the effect would be to reduce the apparent effect of TXA.   I wonder why it wasn’t reported.

Even if T1 was defined unambiguously, why did the authors not choose the blood lost between it and 30 minutes, or six hours later.  Or between any other two time points.  Or the drop in haemoglobin, or the number of units transfused.   If you measure something 20 different ways and then perform a statistical test to see if the difference you observe could have occurred by chance only one in 20 times, you’ll get an apparently statistically significant result nearly every time. Even with a drug that does nothing at all.

The way to get round this is to publish your primary endpoint, sample size, and statistical tests before you start, or at the very latest before you take a peep at the data.  It’s called trial registration.  Reputable journals insist on it.  Even “author pays” ones usually do.  This trial’s registration number ISRCTN09968140 appears at the start of the paper, so sad sacks like me can check the details here.  The primary endpoint is indeed blood loss between T1 and two hours.  But look closely.  Trial registration occurred in Feb 2011, three months after the trial paper was submitted to Critical Care, and three YEARS after the trial was completed!  That’s not trial registration in any meaningful sense. We cannot now be sure that the primary endpoint, statistical methods and sample size were really pre-specified.

The way the researchers got consent from women to participate is also interesting.  They claim that women gave written consent in accordance with the Declaration of Helsinki, but this can’t be true.  There would have been no time to fully inform most women. They would have been bleeding, frightened and exhausted.  So they must have used some sort of “consent ritual”.   Presumably waved a form in front of her and said “sign here”.  Out of 154 eligible women apparently only 2 refused consent!  Hmmmm!

It would have been more honest to use waived or deferred consent.  In my small experience of doing this for the WOMAN trial about 1/4 of women, or relatives, say no and we respect their wishes.   Click here.

The title of the article was “High-dose tranexamic acid reduces blood loss in postpartum haemorrhage”. I don’t think so.  I’m not saying it doesn’t work.  It might or might not.  But this trial tells us almost nothing.

At least the authors end by saying there is a need for a large study like the WOMAN trial.  Indeed there is.

Jim Thornton