Hidden sample size change may have altered result

The IRIS trial in this months British Journal of Obstetrics & Gynaecology (BJOG) compared two speeds for administering MgS04 to mothers for fetal neuroprotection. The idea was that the slower one might reduce side effects. The trial report is here. It was registered here.  The planned sample size was 140, and the primary outcome any maternal adverse effect. The authors recruited 51 women and found no difference. But they hid their failure to achieve the intended sample by writing, at best incorrectly: “We aimed to recruit 51 women in total.”

Fortunately the editor asked Dwight Rouse, one of the most experienced obstetric triallists in the world, to write a commentary. He checked the trial registration site and picked up the sample size change: “the authors failed to acknowledge it, but should have, the intended sample size when they originally registered the trial was 140 women”.

Good for him. Not good for the editors to fail to make the authors correct the manuscript. Online readers will likely miss Rouse’s commentary.

But it gets worse.  The trial registration site now reads “Prospectively registered, target sample size 51, status complete”.  Only on the trial registration history here will you find that on 3 March 2013, the sample size was altered from 140 to 51 because “Sample size was based on a reduction in the experience of any adverse effects of the magnesium sulphate infusion (the primary outcome).”   This makes no sense. If the rate of the primary endpoint was lower than anticipated, it would be a reason for increasing the sample size!

Does it matter?  The results were as follows:

Any adverse event. 15/25 (60%) 60-minute loading; 21/26 (81%) 20-minute loading; risk ratio (RR) 0.74; 95% confidence interval (95% CI) 0.51-1.08. Adverse event leading to ceasing the loading treatment (1/25 in the 60-minute loading group; 2/26 in the 20-minute loading group; RR 0.52; 95% CI 0.05-5.38.  

The authors conclusion:

A slower rate of administering the loading dose of magnesium sulphate did not reduce the occurrence of maternal adverse effects overall. 

But if the trial had gone on to its planned sample size of 140 and if the percentage of adverse events in each group among the latter 89 had been the same as the first 51 – supposition, but a reasonable supposition – the results might have been 42/70 (60%) v 56/70 (81%). RR 0.74, 95% CI 0.59 – 0.92. P = 0.0078   i.e. slow infusion does reduce side effects.

I know I’m a boring obsessive, but if the question matters, and this one surely does because MgSO4 treatment reduces cerebral palsy, then getting the right answer matters.

This was government funded research.  If pharma did this sort of thing, imagine the fuss AllTrials.net would make.

Jim Thornton