Trial endpoint troubles

The ROLO trial (BMJ Sept 2012 – click here) tested the effect of a low glycaemic diet in pregnancy, for women who had previously delivered a very big baby. Controls got no special dietary intervention.

The idea was good. Birthweight above 4kg often causes trouble, and a low glycaemic diet minimises the blood-sugar driven insulin rise rise after food, which crosses the placenta and makes babies bigger. But it didn’t work. Oh well. Let’s look at the trial registration

The trial was registered here in August 2009, and participants were recruited from Jan 2007-Jan 2011.  Although the planned sample size was 700, they don’t explain why 800 were actually randomised, of whom 759 ended up being analysed, but that’s a minor point.

The registered primary outcome was “mean birth weight centiles and ponderal indices […] measured at 14, 28, and 34 weeks, at birth and 3 months post-partum”.  Ignoring the prenatal time points, that still leaves two outcomes measured twice, i.e. four primary outcomes.  That’s three too many – a trial should have one primary outcome.

But the reported primary outcome was different again – raw birth weight. Birth weight centile, and ponderal index were secondary. None were reported at three months.

Turning to secondary outcomes the initial registration listed: maternal weight gain in pregnancy, urinary metabolomics, cord insulin, leptin and IGF-1, and placental weight, villous and vascular development, all measured at 14, 28, and 34 weeks, at birth and 3 months post-partum.  On 1 Sept 2009  maternal weight gain at the various time points became the only planned secondary outcomes.

However, in the paper, besides maternal weight gain at the four time points, another eleven secondary outcomes are listed in table 2, and six more in table 3.  None appear in the trial registration document.

Does this matter?

It matters because if triallists don’t choose one primary outcome and stick to it, their tests of statistical significance will be wrong. If they don’t pre-specify and report all secondary outcomes, readers can never be sure that outcomes have not been selected to support a particular point of view.

When this paper eventually gets included in the relevant Cochrane review, it will be marked as “susceptible to bias” because of potential selective reporting.  Which is a pity because it seems to be otherwise a well conducted trial. The BMJ should know better.

Trial funding – Health Research Board of Ireland, and National Maternity Hospital Medical Fund, i.e. Irish taxpayer.

Final point: if anyone knows what ROLO stands for, let ripe-tomato.org know

Jim Thornton