Routine erythropoietin administration from early life in very preterm babies has been postulated to be neuro-protective. Unfortunately, despite some early promising results, 2 large well-performed RCTs showed no benefit, including the largest with 750 babies under 28 weeks, which showed very small differences between groups and nothing that appeared to be a real effect of the erythropoietin.

Another trial of about the same size, but with less immature babies (800 of less than 32 weeks gestation) used a dose about half of those other 2 trials, it was published in Annals of Neurology for some reason. Of note, it was registered on clinicaltrials.gov after it was completed, and the registered primary outcome variable is MDI<70. Even though they registered the trial after they performed it, the published primary outcome variable is not the same, in fact, the published manuscript methods section does not mention a primary outcome variable, the results show in table 2 as “Primary Outcomes” 3 outcomes, Death, Disability, and Death + Disability. The sample size reported in the publication is 743, the registration document (submitted after the trial was finished) states the sample size is 490.  For several years the Annals of Neurology has required that RCTs are registered, that the registered primary outcome is the same as that reported by the article, and that the publication complies with CONSORT guidelines. Fail, fail, fail.

You might ask why I am spending so much time criticising a journal and an article when it was published in 2016? That article purported to show a substantial reduction in major intracranial haemorrhage with erythropoietin administration (EPO) from 16% to 7%, an enormous reduction in PVL from 20% to 11%, a reduction in sepsis from 26% to 20% and a lowering of “Death + Disability” from 27% to 13%. The Cochrane review of outcomes of early erythropoietin in preterm infants (which you can reach full text via the Vermont Oxford website https://public.vtoxford.org/cochrane-at-von/) is heavily dependent on that trial “Song 2016”. For serious IVH, for example, that trial contributes 73% of the weight in the analysis, for PVL it is 90%. If the data are unreliable, which I contend they are, then the systematic review is unreliable also. The Cochrane Systematic Review (SR) does not include the results of the most recent large trial Juul et al, as it was published after the SR was completed, and after the first author of the SR Arne Ohlsson sadly died, last year. That trial from the PENUT consortium showed NONE of those effects, no difference in IVH, PVL or sepsis or in developmental delay.

The Song 2016 trial is also the only trial in the SR that shows a reduction in NEC. As a result of only that trial the overall impact of early EPO appears to show a reduction in NEC. The same authors from Zhengzhou appear to have carried on with their same poor research practices, and immediately performed a subsequent trial (Wang Y, et al. Erythropoietin prevents necrotizing enterocolitis in very preterm infants: a randomized controlled trial. J Transl Med. 2020;18(1):308). Song 2016 enrolled patients until June 2013, (or perhaps January 2013 according to the registration document) the new trial using the same dosage regime (500 units/kg every 48 hours for 2 weeks) started recruiting in January 2014. Enrolment continued until June 2017 and the trial was registered 2 years later in April 2019. 1285 babies of less than 33 weeks gestational age were randomized. The primary outcome in the registration document is “Incidence of NEC” not further specified. In the publication which, strangely, has appeared in BMC Journal of Translational Medicine, the primary outcome is the same, but whether all grades of NEC or stage 2 and higher was the primary outcome is never specified.

If we concentrate on “confirmed NEC” as they refer to it, which is Bell grade 2 or higher, it is very frequent in this trial, 17% of the controls under 28 weeks, but the numbers of babies involved under 28 weeks is just 110 (53 controls and 57 erythropoietin). The incidence of confirmed NEC in more mature babies is also very high, at 30 to 36 weeks it is over 3%, compared to, for example, recent data from the Canadian Neonatal Network which shows an incidence of 1% at that gestational age.

Strangely, in both the new trial, and in Song 2016, the controls received a saline placebo, but the trials are reported as not being masked. I am not sure why you would go to the trouble of preparing and administering a placebo saline injection if the caregivers knew it was a placebo! Interestingly, the CONSORT flow chart shows that there were 1327 potentially eligible infants screened and only 20 refused consent, which is the highest consent rate I can remember seeing for a long time.

Mortality was 50/644 controls, and 34/641 EPO babies, which looks like a difference which is likely to be due to chance, but amazingly enough there is no statistical test of mortality in the manuscript, and compatibility intervals for mortality differences are also never reported.

What also surprises me about this article is the failure to report any other clinical outcomes. NEC and mortality are reported, but the outcomes that this group has reported previously as being reduced by EPO (IVH, severe IVH, PVL, sepsis, retinopathy) are nowhere to be seen.

The Journal of Translational Medicine is supposed to adhere to CONSORT standards, but they are woefully lacking in this publication. One might wonder why a multicentre clinical trial in preterm infants with major potential clinical importance was submitted to this journal; perhaps it was not the first choice, or perhaps the editorial standards suited the authors, the article was submitted to that journal May 20th this year, and accepted July 26th. Many open access journals publish the names of reviewers, and make available all versions of the article from initially submitted to final revisions, but BMC Journal of Translational Medicine does not.

The data in this publication are not consistent with the results of Juul, who showed a very minor difference in NEC incidence among infants who were all less than 28 weeks gestation; 6% with EPO and 8% in controls for stage 2 or 3 NEC. I discussed that high-quality trial previously https://neonatalresearch.org/2020/01/17/erythropoietin-for-brain-protection-in-the-very-preterm-not-worth-penuts/

This new trial creates a big problem for all of us interested in evidence-based practice. I think that the data are unreliable, but the sample size is so large that the results will overwhelm any attempt at systematic review of EPO impacts on NEC. Fortunately, there are few babies under 28 weeks, but even an analysis of EPO for prophylaxis of NEC among babies under 28 weeks will be heavily impacted because of the large number of events. I guess we can be grateful that all the other usual clinical three-letter outcomes were NOT reported! At least SR of early EPO impacts on IVH, PVL, BPD, RoP and sepsis won’t be impacted by the results of this trial.

Unfortunately, the same group appears to be continuing to perform their studies. There is another registration document on clinicaltrials.gov for a trial by this group (registered April 2016, enrolment was expected to start in May 2016 and be complete in December 2019) which is apparently underway. This, of course, overlaps with the trial just reported which finished enrolment in June 2017. In this most recent publication, they refer to the next trial as being among infants of under 28 weeks gestation, and examining retinopathy. In the registration document retinopathy of prematurity (not further specified) is listed as one of several short term secondary outcomes. The primary outcomes are listed as being two separate outcomes Mortality and the Incidence of neurological disability at 2 years old.

I for one won’t be giving routine EPO to my preterm babies to prevent NEC based on this data. For now, all that I think we can be sure about is that early routine EPO stimulates red cell production, which, depending on transfusion thresholds, is very likely to reduce blood transfusions. Any other potential benefits are unproven.

Prospective registration of RCTs has numerous benefits, including ensuring the completeness of the scientific record, ensuring that the analyzed primary outcome variables are those initially planned, and avoiding the addition of post hoc outcomes, unless they are clearly reported as being post hoc outcomes. Allowing publication of RCTs which are registered after they have been completed negates all those benefits, even when the authors still can’t get the sample size or the primary outcome correct!