Authors’ Response

We thank Dr Rodriguez for her insightful comments on our article in which we report neurodevelopmental outcomes of preterm infants with retinopathy treated with bevacizumab or laser surgery. We agree entirely that the results of our nonrandomized study are confounded by baseline differences between groups and the likelihood that treatment options varied by center and over time, a point that is emphasized throughout the article.

For statistical adjustment, before embarking on the analysis, we a priori chose covariates known to be associated with neurodevelopmental outcomes in preterm infants, such as gestational age, sex, severe intracranial hemorrhage (ICH) or WMI, bronchopulmonary dysplasia (BPD), surgical necrotizing enterocolitis, and maternal public insurance, while controlling for center as a random effect. We did not include birth weight because of its collinearity with gestational age. We did not include duration of ventilator support or supplemental oxygen because of collinearity with BPD and because length of respiratory support could be attributable to differences in ROP therapy. Severe ICH or WMI was chosen as the brain injury marker on the sonogram because of its demonstrated association with neurodevelopmental outcomes. We reviewed our data and found that there were 23 infants who had missing data for ventricular enlargement after 28 days of age and through 36 weeks’ postmenstrual age (PMA), compared with severe ICH or WMI through 36 weeks’ PMA. Therefore, per your suggestion, we have developed random-effect models (see Table 1) for neurodevelopmental outcomes and mortality, adjusting for birth weight (instead of gestational age), sex, severe ICH or WMI through 36 weeks’ PMA, BPD, surgical necrotizing enterocolitis, and maternal public insurance as a sensitivity analysis. The significant results for mortality that we observed earlier did not change.

The subgroups of infants who died after treatment with bevacizumab (n = 26) or laser surgery (n = 14) involved small numbers. We did not conduct comparisons of baseline characteristics in the 2 groups, precluding any conclusions.

Unfortunately, we do not have data on zone of ROP at treatment and could not include it as a potential confounder of outcomes or of choice of therapy.

The sensitivity analysis to compare infants at centers that treated exclusively with laser surgery or bevacizumab revealed a significant difference in death (but not in NDI), in line with our findings overall. However, selection bias related to differences in case mix, treatment preferences, and outcomes across these centers, as is typical of any multicenter cohort, may have influenced outcomes and cannot be teased out. Center was included in the logistic regression model.

The findings of the randomized-controlled BEAT-ROP trial of comparable mortality between groups has been discussed in the article. A recent small study revealed similar visual and adjusted neurodevelopmental outcomes at a mean age of 1.5 years in groups of children with ROP without requirement of treatment and those (with a lower gestational age) treated with intravitreal bevacizumab.¹  In a before-after comparison of laser surgery and bevacizumab, no differences were found in the overall primary outcomes of death, cerebral palsy, bilateral visual impairment, and hearing loss; Bayley Scales of Infant Development scores were available in 13 and 9 infants in the bevacizumab- and laser-treated groups, respectively.²  The single-site subgroup of BEAT-ROP did not have any deaths.³  These small studies may not have been powered to detect differences between groups.

We believe that we have carefully acknowledged caveats to our findings of a statistical association of bevacizumab treatment of severe ROP with death and adverse cognitive outcomes after adjustment for multiple covariates that were chosen a priori. We have conducted additional sensitivity analyses in an attempt to address variation in choice of therapy. We have specifically discussed the comparability of the groups, and we have discussed that the cohort study design was not ideal for comparing mortality differences between groups or for uncovering associations between single comorbidities or therapies with childhood outcomes. We have underscored the need for rigorous appraisal of the risks and benefits of bevacizumab in a large randomized trial with neurodevelopmental follow-up. Until then, the choice of therapy for severe ROP should be guided by clinicians on the basis of ongoing review of available data and a careful individualized assessment of risks and benefits, both visual and neurodevelopmental.