An important high quality trial has just been published, it has taken me a bit longer than usual to process the new info. Among other reasons a nice review was posted on the “other neonatal blog“, but I wanted to try and put this in context of the other similar published trials. The new trial is Baud O, et al. Effect of early low-dose hydrocortisone on survival without bronchopulmonary dysplasia in extremely preterm infants (PREMILOC): a double-blind, placebo-controlled, multicentre, randomised trial. The Lancet. 2016.
523 babies of less than 28 weeks, and at least 24 weeks, were enrolled and randomized (23 week gestation infants are not generally actively treated in France, so they weren’t included) in the first 24 hours of life. Infants who were severely growth restricted or asphyxiated were not included. They didn’t have to be on a ventilator or even requiring oxygen, so it was a true trial of prophylaxis. Babies received 1 mg/kg/d of hydrocortisone (divided in 2 doses) for 7 days followed by 0.5 mg/kg/d for 3 days, or placebo. The primary outcome was survival without bronchopulmonary dysplasia at 36 weeks. The study had a sequential analysis design, so sample size is not specified strictly in advance, an interim analysis was done every time an additional 100 patients reached the primary outcome. The maximum sample size that they were aiming for was 786 infants, but the study was stopped prior to that, not because they crossed the analysis lines showing efficacy or futility, but because they were running out of money and resources, so in March 2013 they decided they would have to stop in January 2014.
They found an improvement in the primary outcome, that was just “statistically significant”, that is, the estimated probability that such a difference would occur due to chance alone is 0.04, or 1 in 25. Neither of the 2 components of the primary outcome were individually statistically significant, but both were improved in the hydrocortisone group compared to control.
This study has something in common with several other prior trials. I have summarized the previous trials that randomized all very preterm infants, or all very preterm infants on ventilators to low dose hydrocortisone starting before 48 hours. As you can see they all used relatively similar doses of hydrocortisone, and all except the latest study required babies to be intubated. Some of the babies were a bit larger than Baud’s subjects.
|Publication||Patient characteristics||Dose per kg per day used.|
|Watterberg 1999||N=40, MV, 500-999g, <48h||1 mg x 9d, 0.5 x 3d|
|Biswas 2003||N=253, MV, <30 wk, <9h||1 mg x 5d, 0.5 x 2d (plus T3)|
|Watterberg 2004||N=360, MV, 500-999g, 12-48h||1 mg x 12d, 0.5 x 2d|
|Peltoniemi 2005||N=51, MV, 500-1250g, <36h||2 mg x 2d, 1.5 x 2d, 0.75 x 6d|
|Bonsante 2007||N=50, MV, <1250g (24-30wk), <48h||1 mg x 9d, 0.5 x 3d|
|Baud 2016||N=523, 24-<28 wk, <24h||1mg x 7d, 0.5 x 3d|
The Biswas study also treated the experimental group with tri-iodothyronine, but T3 is not effective in improving survival or BPD so it could be left in a review for now.
I haven’t found any other trial data so far, if anyone knows of any, please let me know.
I did my usual thing and put the data in Revman to see what the Forest plots now look like.
The first is the meta-analysis of the effects on Death, which doesn’t look very impressive
The next is the impact of hydrocortisone prophylaxis on BPD, which is similarly not “statistically significant”
Finally the combined outcome of “death or BPD”, which is an outcome that should perhaps be junked, suggests a possible 11% reduction, but with an upper 95% confidence interval that is very close to 1.0
These trials were all of reasonably good quality, with some heterogeneity in the 3rd of these analyses, if you do some sensitivity analyses, taking out the Biswas trial because it also used T3, the results are almost identical.
I think at this point we have to say the benefits of early low-dose universal hydrocortisone prophylaxis are “Not Proven”, either among all very preterm babies or among those who are ventilated. If there is a real effect it is probably not huge, but there is a potential for as much as a 32% reduction in mortality if you look at the confidence intervals for death. Unfortunately you will all guess what comes next, we need another bigger trial. Before designing such a trial however, the long term follow up of these trials should be collated, there have been some indications of adverse effects of adverse long term effects from the small Peltoniemi trial, but the data from the follow-up of Watterberg’s 2004 trial were reassuring. Long term outcomes of the new PREMILOC trial will be essential before either instituting this therapy, or performing another larger trial.
As with all of your posts Keith this one is spot on. Excellent analysis and thank you for the “nod” at the start of the post.
Keith, please educate the mortals and the ethicists as to the reasons a researcher would chose to do a sequencial analysis design over one in which a sample size is specified in advance.
Why isn’t that done more often? Isn’t it a good thing to do?
Isn’t it conveniant that one could just stop an investigation exactly when there are enough patients to reach p = 0.04? (or when runs out of resources?)