By Nora Laberee
Scheie Vision Annual Report 2019
Retinopathy of prematurity (ROP) is a leading cause of vision loss in children. If not caught and treated in time, it can lead to lifelong blindness. ROP affects over 30,000 new premature infants in the United States each year. While many of these cases will improve without treatment, the most severe cases are important to diagnose and treat quickly to prevent retinal detachment and blindness.
The American Academy of Pediatrics publishes ROP screening guidelines, which currently recommend performing retinal examinations for infants with a birth weight below 1,501 grams or a gestational age of 30 weeks or less (for perspective, normal human gestation is 40 weeks). Bigger and older babies who have an unstable clinical course, per the judgment of the neonatologist, should also be screened. Based on these guidelines, roughly 70,000 infants each year undergo repeated eye examinations, even though less than half of them will even develop ROP. These examinations can be uncomfortable and physically stressful for very small babies. In addition, coordinating and performing these examinations requires significant time and resources from neonatologists, ophthalmologists, nurses, and coordinators. A decade ago, Gil Binenbaum, MD, MSCE, set out to find a better way to screen for ROP.
Dr. Binenbaum is the Richard Shafritz Endowed Chair of Pediatric Ophthalmology Research at the Children's Hospital of Philadelphia (CHOP), and Associate Professor of Ophthalmology at the Scheie Eye Institute. By closely considering how ROP develops, Dr. Binenbaum and his collaborators have been able to better predict which babies will develop severe ROP.
Why ROP Develops
The causes of ROP are complex. Premature birth introduces a baby to a hyperoxic, or high oxygen, environment, because oxygen levels for the fetus are actually low inside the womb. Hyperoxia results in decreased production of a protein called vascular endothelial growth factor (VEGF), which the retina produces to induce growth of blood vessels when oxygen levels are low. In the new hyperoxic environment of the premature infant, VEGF production in the retina stops, so retinal vessel development also halts. Immature blood vessels also can be damaged by hyperoxia in a process called vaso-obliteration. The results of both processes are poor retinal vessel development.
A second effect of premature birth is a drop in the growth hormone insulin-like growth factor 1 (IGF-1), which premature infants do not produce well. VEGF activity requires adequate IGF-1 levels. As the retina develops in the weeks following birth, it shifts from relative hyperoxia to hypoxia, or low oxygen, because the poorly developed retinal vessels do not deliver enough oxygen. In response, the retina secretes VEGF to promote blood vessel growth, but because the baby is not making much IGF-1, the VEGF does not function effectively, and the retinal vessels still do not grow well. By the time the baby starts to produce higher amounts of IGF-1, VEGF levels in the eye are too high, and the retinal blood vessels grow uncontrollably and in the wrong direction. The retina can be pulled off into a retinal detachment, and vision is lost.
Low levels of IGF-1 are a key factor in this process, and therefore slow weight gain, a surrogate measure for low IGF-1, is predictive of the later development of severe ROP. This is the key piece of information that Dr. Binenbaum and his team took advantage of to better predict which infants need retinal examinations.
Revising ROP Screening Criteria
While Dr. Binenbaum has a number of research interests (for example, he was the principal investigator of the collaborative project between Penn and the Philadelphia Museum of Art that showed how art classes can improve the observational skills of medical students), he has been intrigued by ROP since his days as a fellow. Soon after, he began conducting early studies on infant weight gain, IGF-1, and ROP with the support of the Penn Vision Scientist NIH K12 program and his mentors, Dr. Graham Quinn and Dr. Maureen Maguire. In 2012, he received an R01 grant from the National Eye Institute to form the Postnatal Growth and Retinopathy of Prematurity (G-ROP) Study Group.
We worked closely with Gui-shuang Ying, director of the G-ROP data center at Penn, and an executive committee of national experts, to design two large studies,” he explains. “First, a retrospective model development study, and second, a prospective model validation study to make sure that the model was reliable before using it clinically.”
Twenty-nine hospitals participated in the first study (G-ROP-1), and 7,483 premature infants at risk for ROP were retrospectively studied. Published in JAMA Ophthalmology in July 2018, the study produced new evidence-based screening criteria for ROP. These criteria not only greatly improved specificity, or the ability to rule out the babies who will not develop severe ROP, but also improved sensitivity, which refers to the ability to detect babies who will develop severe ROP and must have examinations. By using a hybrid modeling approach, Dr. Binenbaum and his team were able to lower the birth weight and gestational age thresholds. In order to capture the few larger babies who developed severe ROP, they had to add criteria for slow postnatal weight gain. If an infant met any one of the six resulting criteria, he or she required retinal examinations.
These new “G-ROP Criteria” correctly predicted 459 out of 459 infants with severe ROP. At the same time, if the criteria had been used to decide who receives examinations at the study hospitals instead of the current guidelines, the new criteria would have saved 2,269, or 30%, of infants from unnecessary exams.
Validating the Model
With these encouraging results from the retrospective study under their belts, the study group began a prospective validation study, which tested the criteria in a new cohort of infants. While many groups had previously attempted to improve ROP screening guidelines by incorporating measures of slow weight gain, they universally ran into the same problem: the number of babies in the model development study had been too low, and the models did not perform well enough when tested in new groups of infants. "When developing a model, if the model is overly complex relative to the number of cases of severe ROP in the study, the model can be overfitted,” Dr. Binenbaum explained. “This means the model may be predicting random variation, instead of true associations.”
The G-ROP study group tried to avoid overfitting by studying thousands of infants when creating the G-ROP criteria. Even so, there was a still a chance that overfitting had occurred, or that some other unknown factors might make the criteria not perform well when applied to new groups of infants. Therefore, it was important to validate the criteria in a second study before using them clinically.
The study group prospectively studied 3,980 new infants at 41 hospitals in this second study (G-ROP-2). “We applied the same G-ROP criteria developed in the first study, and we found that the criteria held up to validation,” Dr. Binenbaum said. “The model identified 100% of babies with severe ROP and reduced the number of babies needing exams by 36%.” These results were just published in JAMA Ophthalmology in November 2019. This was the first growth-based model for ROP to maintain 100% sensitivity when validated.
Between the G-ROP-1 and G-ROP-2 studies, a total of 11,463 premature infants were studied. The G-ROP criteria predicted 100% of the 677 infants who developed severe ROP, while the currently used birth weight and gestational age guidelines missed a small number of these severe ROP cases. At the same time, the G-ROP criteria greatly reduced the number of infants requiring examinations compared to the current criteria. “We estimate that if you scale these results to ROP screening across the United States, about 25 additional cases of severe ROP would be caught, and about 14,000 fewer infants would receive examinations each year, if we used the G-ROP criteria instead of the current birthweight and gestational age criteria,” Dr. Binenbaum explained.
“With successful validation, we now believe that these new criteria can be used clinically to reduce the number of babies who require exams,” Dr. Binenbaum said. “In fact, some may argue that they should be used, because they have a higher sensitivity than the current criteria, in addition to higher specificity.”
Looking forward, Dr. Binenbaum recognizes that modifying ROP screening criteria is not a simple process. He thinks that the G-ROP criteria may first need to be incorporated into the published guidelines, in order for clinicians to feel comfortable using them. In addition, he points to the possibility of using the G-ROP criteria in conjunction with the original subjective “third” criteria, which suggests examining babies with an unstable clinical course who do not otherwise meet the criteria. This would allow neonatologists to continue to use their clinical judgment in requesting ROP examinations for infants they feel are at high risk.
“We also recommend that if and when physicians start using the new criteria, that we continue to monitor their performance,” Dr. Binenbaum said. “So if there are outliers who are identified, we share those data with each other so we can update the criteria if needed, and so we understand what’s being missed, if something is being missed.”
With such a large and comprehensive dataset, Dr. Binenbaum and his team plan to continue examining the rich data and gaining as much knowledge as possible from them. The team hopes to provide new insights that can help to improve outcomes for neonatologists and ophthalmologists treating ROP, and most importantly, for babies who develop ROP.