There remains some doubt about the impacts on cerebral development, and thus on developmental progress of anti-VEGF treated babies. It is possible that there are cerebral effects of VEGF inhibition (or interception) and it isn’t clear what the exposure of the brain really is. In this new trial (Stahl et al, that I just posted about), most babies had undetectable serum aflibercept concentrations throughout the study, which is reassuring. Data from other observational cohort studies has shown some association between anti-VEGF “-mab” administration, and poorer outcomes. In this study from the NICHD network, which Roger Soll refers to in his editorial accompanying the FIREFLEYE trial publication, their long term primary outcome, “death or severe NDI” was not much different between bevacizumab and surgery treated babies (adjusted Odds Ratio was 1.42; with 95% CI 0.94 to 2.14). However, mortality was somewhat higher (most of the deaths, and the difference in deaths, was prior to hospital discharge; 9% of the treated babies vs 3.5%), and Bayley III cognitive composite scores were shifted lower by a mean of 3 points, so somewhat more babies were under 85 (aOR 1.78 [95% CI 1.09 to 2.91]). The motor scores were slightly lower in the bevacizumab group, and the language scores were almost identical. The babies in the bevacizumab group also had longer assisted ventilation, oxygen therapy, and hospitalisation, so perhaps they had more severe BPD.

Published follow up of the CNN and CNFUN Canadian cohort also shows lower scores among the bevacizumab treated babies compared to laser surgery, but the pattern is different, cognitive scores were identical between groups, but language and motor scores were worse. There were very few deaths between treatment and discharge. The CNN bevacizumab babies also seem to have been a bit sicker in terms of their lung disease than the laser treated infants. The difference in severity of lung disease is understandable, as bevacizumab was emerging as a treatment during these periods, we tended to use it for babies who we were most worried about deteriorating during surgery, so the simplicity of bedside intravitreal injections made us prefer it among the babies with the most unstable pulmonary status.

The real impacts at long term will require follow up of randomized trials, at present the observational data are conflicting and confusing, in addition to the two large multicentre cohorts discussed above, there are several others, and a systematic review in 2020 that found a total of 8 studies, actually showed no overall difference in outcomes. Another systematic review, (Kaushal et al) also published in 2020 included data from 13 studies. There are however, errors in their meta-analysis which I have just noted as I was reviewing it now. They calculated the mean and SD of the data from the CNN, estimating them from the median and IQR which were published. But the SDs they calculated are tiny, and are, I think, incorrect, using the method they claim to have used (I checked with an online calculator which is supposed to be based on the method the authors used, but gives SDs which are larger and more reasonable). As a result the CNN data are given huge weight compared to the other studies, which means that their calculations for the continuous outcomes are in error, I think.

The Cochrane review of this intervention doesn’t include any long term follow-up, but there are now some data available from comparative trials, or at least from the RAINBOW trial. (Marlow N, et al. 2-year outcomes of ranibizumab versus laser therapy for the treatment of very low birthweight infants with retinopathy of prematurity (RAINBOW extension study): prospective follow-up of an open label, randomised controlled trial. The Lancet Child & Adolescent Health. 2021;5(10):698-707)

A caution about the introduction to the trial, as you can see from the conflicts of interest statement, there was heavy involvement of Novartis in the trial. Novartis markets both bevacizumab (Avastin) and ranibizumab (Lucentis), which are derived from the same antibody. Novartis, however, want everyone to use ranibizumab for eye injection because they charge many times as much for it, they prepare it in the small doses needed for eye injections, and, according to them, it is less likely to cause systemic effects. The molecules are very different in size, with ranibizumab being a much smaller molecule which is probably cleared from the circulation much faster, which might possibly make it safer. Comparative trials in adults, however, haven’t shown much difference in either efficacy or complications. Bevacizumab is in fact “non-inferior” to ranibizumab in adult wet macular degeneration efficacy. (Moreno TA, Kim SJ. Ranibizumab (Lucentis) versus Bevacizumab (Avastin) for the Treatment of Age-Related Macular Degeneration: An Economic Disparity of Eye Health. Semin Ophthalmol. 2016;31(4):378-84).

All that being said, the first author is Neil Marlow, and I am sure he wouldn’t have written this unless he was sure about the data. These long term outcomes show equivalent development in almost all domains, but ranibizumab led to much less high myopia, and less ocular structural abnormalities than laser. There is very little follow up of the bevacizumab, trials, the BEAT-ROP trial only reported follow up from 16 infants from one centre, and therefore no power to show anything. Kennedy KA, et al. Medical and developmental outcomes of bevacizumab versus laser for retinopathy of prematurity. J AAPOS. 2018;22(1):61-5 e1. Even though cognitive scores were 20 points higher among those who received bevacizumab rather than laser there was such a wide range, and small numbers, that this might well have been a chance difference.

The long term, therefore, must be classified as uncertain, with some concerns from observational studies, and some reassurance from the little data available from RCTs. Which is a very unsatisfactory state of affairs when trying to counsel parents.