In the March print edition of the Journal of Pediatrics, the report of the Insulin-like growth factor 1/IGF binding protein 3 trial, as a preventive for retinopathy. Ley D, et al. rhIGF-1/rhIGFBP-3 in Preterm Infants: A Phase 2 Randomized Controlled Trial. J Ped. 2019;206:56-65 e8.
Many years of fascinating work out of Sweden, by Dr Hellstrom in particular, have suggested a rôle for IGF in the development of retinopathy of prematurity. The evidence trail goes like this: extreme preterm babies with intra-uterine growth restriction are at increased risk of retinopathy, impaired early postnatal growth is also associated with more retinopathy. Babies with IUGR have lower IGF concentrations, as do babies with poor postnatal growth, and IGF levels correlate negatively with RoP risk. New vessel formation in the retina is important in the vasoproliferative phase (which is just saying the same thing twice!), and such new vessel formation is dependent on VEGF. Insulin-like Growth Factor 1 plays a rôle in retinal vascular development, it seems to be required for VEGF to have an effect. IGF1 concentrations decrease after preterm birth, and then slowly increase. The low initial IGF1 concentration is thought to participate in poorer retinal development initially, but then, later on when the IGF1 levels increase, it permits VEGF to act, promoting neo-vascular proliferation. Earlier better nutrition leads to an increase in early IGF concentrations and a decrease in later concentrations, and is associated with less RoP.
All of which information has led to the important question: if we supplement IGF early, can we decrease retinopathy?
That it turn led to the performance of studies to show that recombinant human IGF-1 can be given intravenously, that it increases measured IGF levels, and that we have some idea about dose requirements.
And then to an RCT to show whether or not IGF infusions decrease retinopathy. During the course of these investigations, a mixture of IGF and an IGF binding protein was developed. IGF-1 usually is transported in the circulation bound to an IGFBP, the most important of which appears to be number 3, IGFBP-3, and it is a mixture of recombinant IGF-1 and recombinant IGFBP-3 which was tested in this trial.
If you were to design such a study, you would include only babies at substantial risk of serious RoP, you would ensure that the sample size was large enough to demonstrate an impact, and you would determine whether or not there was an impact on mortality and other potential adverse consequences.
Unfortunately with studies designed, funded, and run, by pharma the justifications are different, they want to design studies with the highest likelihood of showing a positive effect, at the lowest cost, in order to ensure that profits are maximized. They also don’t want to have eligibility criteria too restrictive, otherwise the eventual licence is likely to be restrictive.
The best study of IGF prophylaxis against RoP would, therefore, be designed to include only babies of 24 weeks or less (in whom vision-threatening retinopathy is a serious issue) who were already on optimal nutritional protocols. They would evaluate serious RoP (and not particularly stage 1 disease) and have active surveillance for all potential and possible adverse impacts in this critical group of patients. Reductions in stage 1 disease might be important in terms of pathophysiology, but they are not very important clinically. Inclusion of babies born at 27 weeks gestation, for example would have little value, as they very rarely develop vision-threatening RoP, and so would massively increase the required sample size.
How about the current study? It was a randomized unmasked study which included :
Infants with gestational age at birth of 230/7-276/7 weeks … Exclusion criteria included monozygotic twins, detectable gross malformation, known/suspected chromosomal abnormality, genetic disorder/syndrome, a persistent blood glucose level of <2.5 mmol/L or >10 mmol/L on the day of birth, anticipated need for administration of rh erythropoietin during treatment, a history of maternal diabetes requiring insulin, and clinically significant neurologic disease (germinal matrix hemorrhage allowed)
Babies were randomized and started on protocol within 24 hours of birth; intervention group babies received IGF-1/IGFBP-3 infusion continuously up to the start of their 30th week PMA. Normal dose was 250 microg/kg/d of a 50 microg/mL solution, which they therefore received for between 2 to 7 weeks.
The primary outcome variable of this trial was maximum severity of RoP, expressed according to ICROP grading, and analyzed as an ordinal result (0,1,2,3,3+,4 and 5; in the text it states “ROP outcome stages in the current study were classified as 0, 1, 2, 3, and >3” but in the table the results are given for ≥ 3, the numbers are quite small in each group, however).
Results: there were more controls without RoP, 48%, compared to babies who received IGF, 29.8%. The absolute risk increase of 18% in all stages of RoP is most compatible with an absolute RoP difference between a 1.1% decrease and a 35% increase in all stages with the use of IGF. There were also more IGF babies who developed ≥ stage 3 RoP, 25.5% compared to 18% among the controls (there were no retinal detachments (stage 4 or 5) in either group). This is exactly the opposite of what was expected, and calls into question the premise of the study.
There were also more deaths with IGF than among the controls, 20% vs 12%. It isn’t clear how many of those who died after their eye exam actually had RoP, or what the incidence of RoP was among survivors; but we do know that the majority of the deaths occurred before their first eye exam (11/12 IGF infant deaths, and 5/7 controls) so the other 3 babies won’t affect the results much.
The results, therefore, show an absolute increase in severe RoP, with the intervention, of 7.5%, which is most compatible with an absolute difference of between an 8% decrease and a 24% increase in severe RoP. In terms of relative risk that is an RR of 1.31, (compatibility intervals of 0.6 to 2.9) In other words the results show an increase in all stages of RoP and an increase in severe RoP, both of which are compatible with small decreases or major increases.
The authors of the study state that IGF use decreased the occurrence of severe BPD, but if we analyse the trial using standard methods, i.e. no BPD versus BPD (4/49 placebo cf 4/47 IGF with no BPD, using the newer definition), there is no difference between groups. If we just concentrate on those who were in oxygen at 36 weeks, i.e. previous definition of BPD, and what is called here moderate/severe BPD using the newer definition, there were 27 of 49 evaluated surviving placebo babies vs 19 of 47 evaluated surviving IGF babies with moderate/severe BPD); p=0.22 using Chi-square with Yates’ correction, the 15% absolute decrease in BPD among survivors is compatible with a difference between an increase of 5% and a decrease of 33% with IGF. To express that as a relative risk, the RR for BPD with IGF compared to control was 0.84, most compatible with an RR between 0.54 and 1.3, among surviving babies.
The analysis which is reported, and which is touted as being significant, ranks the severity of BPD; that analysis was not mentioned in the study registration (which is admittedly a bare-bones document), and gives a p-value of 0.04. (p-hacking, anyone?). But that analysis is only for BPD among surviving infants who were still in the study at 36 weeks for their evaluation.
I actually think that treating BPD as a spectrum, rather than as a dichotomous outcome, is a better idea than our traditional BPD/no BPD, but I don’t think this division into moderate and severe depending on whether you are over 30% oxygen is necessarily the way to go, It replaces one way of dichotomising the outcome with another. Does being in 35% oxygen at 36 weeks have a greater impact on respiratory outcomes of importance than requiring 27%? For an evaluation of outcomes at 36 weeks, I guess it is, however, justifiable, but should be supplemented with data of more clinical importance, such as respiratory disease during the first year of life. The other reason for being very concerned about the likely reproducibility of these results is the very high rate of severe BPD in the controls. I think likely as a random fluke, there were 4.5 times more severe BPD cases among controls than moderate BPD. Indeed a frequency of 45% severe BPD among infants less than 28 weeks gestation is enormously high, higher, for example, than any of the European regions that recently reported the variation in severe morbidity incidences across multiple European regions. (Edstedt Bonamy AK, et al. Wide variation in severe neonatal morbidity among very preterm infants in European regions, (Archives of disease in childhood Fetal and neonatal edition. 2019;104(1):F36-F45).
In contrast to the apparent decrease in BPD severity, mortality was higher with IGF treatment (12% placebo vs 20% IGF) by chi-square with Yates’ correction p=0.33, this gives a relative risk of death with IGF treatment of 1.68, most compatible with a relative risk of death with IGF of between 0.71 and 4.0.
If the outcome were to be the more frequently used “death or BPD”, that occurred in 34/56 controls and 31/59 intervention babies: p-value from Yates’ corrected chi-square = 0.75. Or for “death or severe BPD” 29/56 vs 22/59, p=0.4237. The authors state that they did a supplementary analysis putting the deaths in the severe BPD group, and note that a “trend for a decrease in severe BPD among treated infants remained (37.3% in rhIGF-1/rhIGFBP-3 group vs 51.8% in the standard of care group)” but no statistic is provided. I re-ran the analysis as a Mann-Whitney U test, putting deaths as a separate, worse group than severe BPD (i,e, coding them as 0 for no BPD, 1 for mild, 2 for moderate, 3 for severe, and 4 for dead), the p-value was 0.49.
I find these results, especially the increase in retinopathy, somewhat surprising, but I guess that is why we have to do prospective randomized trials. The increase in mortality is also concerning, is at least as striking, and, I would suggest, much more important, than the possible shift in severity of BPD.
Why might this be so? There are numerous explanations possible, firstly the compatibility intervals are wide and it is possible that another study might find a decrease in RoP; but the best estimate of a potential positive impact is a small reduction. It is also possible that the association of early lower IGF-1 concentrations among preterm infants who later develop RoP is just an association and not causative, and that the increase in RoP shown by this study is evidence of the permissive impact of IGF-1 on VEGF activity in the retina, later on during the proliferative phase
When I searched the registration documents for the trial to try to find what analyses had been planned for the primary outcome variable, I was surprised to find the large number of revisions (25), many of which were just procedural (addition of centers and so on), but analyzing the documents, almost everything about this trial has changed between initial registration and final publication. The sponsor changed, the intervention changed (it started as a trial of IGF-1, and then IGFBP-3 was added later), the duration of the intervention (originally up to 31 weeks 6 days, changed to 29 weeks and 6 days), the sample size and the exclusion criteria have all changed.
I think this makes a mockery of trial registration, they basically studied a different medication in a different group of babies, with a different sample size, for a different duration than the original registration.
As far as I can see most of those changes were made with the revision in May 2014, which is prior to the actual start of the phase of the study that is reported in this manuscript, which started enrolling in September 2014. So why not just register this as a different trial? That would leave the original documents untouched, and would make it easier for everyone to find what is going on. Changing trial registration documents this extensively means that it will remain difficult to find what studies have been planned, or performed, and never published. Which is one of the important functions of registration.
In summary, this trial seems to show that a prolonged infusion of IGF-1/IGFBP-3 among very preterm infants leads to an increase in retinopathy, an increase in mortality, and a possible decrease in BPD severity, but only compared to a control group with an unusual very high incidence of severe BPD.
Neonatal Research 