A monotonic normal dynamic linear model was fit to the dose-response data. The probability of success for dose (d), pd, was estimated using a logistic function of the log odds for each dose, θd, where pd = exp(θd) / [1 + exp(θd)]. The model shares information across doses by setting a normal prior on the log odds, θd, where each θd has a mode of θd −1 with variance component τ2d.This prior structure borrows information from the previous dose to inform the log odds of success for the subsequent dose. The first dose is modeled with a normal distributed prior with mean (SD) of 0 (2). In addition, the model imposes monotonicity of θd, such that θd increases as a function of dose. Therefore, the model assumes that as the dose increases, the probability of success also increases. CrI indicates credible interval.
Nonauthor Collaborators. The Pediatric Eye Disease Investigator Group members.
Customize your JAMA Network experience by selecting one or more topics from the list below.
Kraker RT, Wallace DK, Beck RW, et al. Choice of Dose Level for a Randomized Clinical Trial of Low-Dose Bevacizumab vs Laser for Type 1 Retinopathy of Prematurity. JAMA Ophthalmol. 2021;139(10):1143–1144. doi:10.1001/jamaophthalmol.2021.3192
Premature infants with severe retinopathy of prematurity (ROP) are often treated using anti–vascular endothelial growth factor drugs such as bevacizumab. Intravitreal bevacizumab reaches the systemic circulation and is associated with reduced serum vascular endothelial growth factor levels in premature infants.1 Vascular endothelial growth factor is necessary for normal development of organs, so blocking its systemic action could be harmful to premature infants.2,3 We conducted a dose de-escalation phase 1 study, which found that very low doses of bevacizumab are effective in treating severe ROP.4,5 To determine which dose to carry forward to a large-scale randomized clinical trial, we applied a bayesian approach to analyze data across 8 dose levels. Herein we report those results (NCT04634604).
The study included infants with type 1 ROP and no prior treatment for ROP. Approval was obtained from each of the 11 institutional review boards at participating sites.4 Written consent was obtained from parent(s) of participants. Families were compensated $50 per completed 1 year-adjusted age visit. Eight dose levels of bevacizumab were evaluated (0.250 mg, 0.125 mg, 0.063 mg, 0.031 mg, 0.016 mg, 0.008 mg, 0.004 mg, and 0.002 mg), each in 10 μL after dilution, if necessary, by individual research pharmacies. Additional details of the study have already been reported.4,5 Within each dose level, the proportion of study eyes meeting success criteria and the 1-sided 95% lower confidence limit for the success proportion was tabulated. Using data across all dose levels, a bayesian statistical model was used to calculate the probability that each of the 8 doses was (1) at least 95% effective in treating type 1 ROP and (2) the lowest tested dose that was at least 95% effective.
A total of 120 infants were analyzed. The mean (SD) gestational age at time of birth was 24.8 (1.6) weeks, and 71 infants (59%) were male. The parents of 19 infants (16%) self-reported their infant as African American, 17 (14%) as Hispanic, and 65 (54%) as White. The probability that a dose of 0.250 mg was at least 95% effective was 100%; 0.125 mg, 100%; 0.063 mg, 94%; 0.031 mg, 84%; 0.016 mg, 64%; 0.008 mg, 33%; 0.004 mg, 5%; and 0.002 mg, less than 1% (Table and Figure, A). The probability that a dose of 0.250 mg was the lowest tested dose that was at least 95% effective was less than 1%; 0.125 mg, less than 1%; 0.063 mg, 6%; 0.031 mg, 10%; 0.016 mg, 20%; 0.008 mg, 31%; 0.004 mg, 28%; and 0.002 mg, 5% (Table and Figure, B).
A limitation of this dose de-escalation study is that the sample size was small. With small sample sizes at each dose, the bayesian statistical model uses information across all dose levels to estimate the true dose response curve. When considering which dose to study in a future randomized clinical trial, a committee of ophthalmologists placed greater emphasis on the probability that a dose was at least 95% effective than on the probability that the dose was the lowest tested dose that was at least 95% effective. A dose of 0.008 mg has the highest probability (31%) of being the lowest dose at least 95% effective, and the probability that it is at least 95% effective is 33%. Thus, there are concerns for relatively lower efficacy and higher risk of disease progression at this dose. The probability that a 0.063-mg dose is at least 95% effective was 94%; however, the probability this dose is the lowest tested dose at least 95% effective is only 6%. Thus, there are fewer concerns for risk of disease progression secondary to ineffective treatment, but it is likely that a lower dose is equally effective. Given that a dose of 0.063 mg is still 4- to 10-times lower than the doses most commonly used in current practice (0.250 mg and 0.625 mg), 0.063 mg may be effective and also reduce the risk of adverse systemic effects. As a result, the dose of 0.063 mg was selected for a randomized clinical trial to evaluate the effectiveness of bevacizumab vs laser as treatment for acute ROP in infants (NCT04634604).
Corresponding Author: Raymond T. Kraker, MSPH, Jaeb Center for Health Research, 15310 Amberly Dr, Ste 350, Tampa, FL 33647 (email@example.com).
Accepted for Publication: June 22, 2021.
Published Online: August 19, 2021. doi:10.1001/jamaophthalmol.2021.3192
Author Contributions: Mr Kraker had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Concept and design: Kraker, Wallace, Beck, Saunders, Lorenzi, Melia.
Acquisition, analysis, or interpretation of data: Kraker, Wallace, Saunders, Lorenzi, Li.
Drafting of the manuscript: Kraker, Saunders, Lorenzi.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Kraker, Saunders, Lorenzi, Melia, Li.
Obtained funding: Kraker.
Administrative, technical, or material support: Kraker.
Supervision: Kraker, Beck, Melia.
Conflict of Interest Disclosures: Dr Kraker reported grants from the National Institutes of Health/National Eye Institute as coordinating center for the Pediatric Eye Disease Investigator Group during the conduct of the study. Drs Wallace and Beck reported grants from the National Eye Institute during the conduct of the study. Dr Saunders reported grants from Jaeb Center of Health Research Foundation during the conduct of the study. Dr Lorenzi reported personal fees from Jaeb Center of Health Research Foundation during the conduct of the study. Dr Melia reported grants from the National Eye Institute to their institution during the conduct of the study and nonfinancial support and grants from Regeneron for clinical trials run by the DRCR Retina Network investigating aflibercept outside the submitted work. No other disclosures were reported.
Funding/Support: This study was supported by the National Eye Institute of National Institutes of Health (grants EY011751, EY023198, and EY018810).
Role of the Funder/Sponsor: The funder had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Group Information: The Pediatric Eye Disease Investigator Group members are listed in the Supplement.