Retinopathy of Prematurity: A Global Perspective and Recent Developments

Retinopathy of prematurity (ROP) is a significant world-wide cause of treatable blindness in preterm infants.¹  ROP is a disease of abnormal retinal vascularization that can lead to retinal detachment and severe vision impairment or blindness. Other possible ocular complications of ROP and prematurity include glaucoma, strabismus, myopia, and amblyopia. This review paper aims at summarizing recent developments in ROP knowledge, research, and delivery of care, with a global perspective.

Since 1984, the International Classification of ROP has represented the standardized classification system based on the location, morphologic features, and severity of both active and ciccatricial ROP disease. In 2021, the third version of the International Classification of Retinopathy of Prematurity (ICROP) was published and included additional innovative ophthalmic imaging and extensive research of ROP.²  “The unifying principle of this classification system is: the more posterior the disease and the greater the amount of involved retinal vascular tissue, the more serious the disease.”²  ROP is classified by 4 zones or locations (Fig 1), 5 stages of severity, and the presence of plus disease, a posterior retinal vascular biomarker of severe disease often warranting treatment (Figs 2 and 3, Table 1).³  Aggressive ROP (A-ROP) is characterized by rapid onset of abnormal neovascularization and severe plus disease and may not necessarily progress through the typical stages of ROP (Fig 4). The earlier term aggressive posterior ROP was recently changed in the third edition of the ICROP, acknowledging the fact that A-ROP may extend beyond the posterior retina.²  There have been 8 clarifications or additions to the pre-existing ICROP, which are summarized in Table 1.

The first major randomized clinical trial for treatment of ROP was the Cryotherapy for Retinopathy of Prematurity Cooperative Group (CRYO-ROP) published in 1988.⁴  Current management and treatment of acute ROP is guided by the Early Treatment of Retinopathy of Prematurity Randomized Trial (ETROP), published in 2003 which established type 1 and type 2 disease characteristics (Table 2).^4,6  Treatment with either laser photocoagulation to the avascular retina or intravitreal antivascualar endothelial growth factor (VEGF) is recommended within 48 to 72 hours for type 1 ROP to prevent further abnormal neovascularization and retinal detachment, whereas type 2 ROP requires close monitoring.⁵  In infants with ciccatricial ROP, or stage 4 and 5, surgical management with scleral buckle and/or pars plana vitrectomy may be indicated.^7,8 

In normal retinal vascular development, vasculogenesis begins at the optic nerve head around 12 weeks in utero and continues from the center to the periphery, until at least 22 weeks of gestation.⁹  After this period, new blood vessels are formed via angiogenesis under the primary influence of VEGF. Complete retinal vascular development of the peripheral retina occurs between 40 and 44 weeks of postmenstrual age (PMA). Therefore, in infants born prematurely and at extremely low gestational ages (GA), vascular development is interrupted at a vulnerable time.

The 2-phase hypothesis for the pathogenesis of ROP is important in understanding ROP screening and NICU care done by the pediatrician and ROP staging, treatment rationale, timing and choice of treatment determined by the ophthalmologist. In phase I, immediately following birth, there is delayed physiologic retinal vascularization, vaso-attenuation and obliteration thought to be related to premature neonatal physiologic stressors, extrauterine hyperoxia, low levels of insulin-like growth factor 1 (IGF-1), and delayed expression of VEGF receptor 2.¹⁰  In phase II, approximately 4 to 8 weeks after birth, there is abnormal proliferation of retinal vascular cells and neovascularization of the retina and vitreous, which is stimulated by increasing local VEGF levels from peripheral avascular retina in response to local hypoxia from active metabolic cellular demands.^11–16  Pediatricians have a critical role in systemic care related to target oxygen levels during the initial resuscitation phase and first weeks of life (phase 1) and during the NICU growing periods (phase 2).

Retinal hypoxia from the peripheral avascular retina in ROP is the major biochemical stimulus that leads to production of hypoxia-inducible factors, which then activates the production of angiogenic genes that code for proteins such as erythropoietin, angiopoietin 2 and VEGF.^17,18  VEGF is a dominant vascular survival growth factor in both physiologic retinal vascular development and pathologic neovascularization of ROP.

Numerous ROP risk factors have been identified, including maternal, prenatal, and perinatal factors, demographics, medical interventions, nutrition, and genetics.¹⁹  The most well recognized are low birth weight (BW), low GA,²⁰  and high and fluctuating oxygen levels at birth and during the neonatal period.^16,21–23  The lower the BW and the GA, the higher the risk of severe ROP. Intermittent hypoxia (SpO₂ <80% for 1 min or more) and fluctuations of oxygen level during the neonatal period have also been identified as risk factors for ROP.^21,23,24 

Maintaining appropriate oxygen saturation has been 1 of the mainstays of ROP prevention in the NICU. Resuscitation using 100% oxygen may result in significant exposure to early hyperoxia and contribute to the risk of ROP.²⁵  Five randomized controlled trials with similar protocols were included in the Neonatal Oxygen Prospective Meta-Analysis (NeoPROM).²⁶  This meta-analysis on aggregate patient data showed that neonates allocated to the lower oxygen saturation group (85% to 89%), compared with those allocated to the higher oxygen saturation group (91% to 95%), had decreased incidence of severe ROP (risk ratio [RR] 0.72 [95% confidence interval (CI), 0.61–0.85]) but had increased risk of death by 18 to 24 months of corrected age (RR 1.16 [95% CI, 1.03–1.31]). Another meta-analysis in 2020 arrived at similar conclusions.²⁷  An individual patient data meta-analysis of the NeoPROM trials showed that there was no statistically significant difference in the primary outcome of death or major disability between the lower and the higher oxygen saturation group (RR 1.04 [95% CI 0.98–1.09]). Similar to the aggregate patient data meta-analysis, in the lower oxygen saturation group mortality was greater (RR, 1.17 [95% CI, 1.04 to 1.31]) whereas treatment of ROP was administered to fewer infants (RR, 0.74 [95% CI, 0.63 to 0.86]).²⁸  Thus, there seems to be a tension between severe ROP and mortality, with attempts to reduce 1 at the cost of the other. The United Kingdom’s National Institute for Health and Care Excellence (NICE) guidelines recommend a target oxygen saturation of 91% to 95% in preterm infants born at less than 32 weeks of gestation.²⁹  In a systematic review of recommendations for oxygen saturation targets in preterm infants, Tarnow-Mordi and Kirby found that most systematic reviews, consensus statements, commentaries, and an American Academy of Pediatrics clinical report favored an oxygen saturation target of approximately 90% to 95%.³⁰  They also reported that 5 systematic reviews indicated that lower oxygen saturation targets increased mortality but not the composite outcome of death or disability.³⁰  Recent understanding of the 2 phases of ROP suggests that prevention of hyperoxia during resuscitation and up to 30 to 32 weeks postmenstrual age decreases the risk of ROP, whereas the prevention of hypoxia beyond 32 weeks postmenstrual age decreases the risk. Data from a few retrospective cohort studies on a biphasic or graded increase in oxygen saturation targets to tackle the 2 phases of ROP have reported better outcomes, both for severe ROP and mortality compared with a uniform target.^31–33  To the best of our knowledge, there are no randomized controlled trials comparing graded oxygen targets to a uniform oxygen target of 90% to 95%. Graded targets would require significant education and training of frontline healthcare workers, which would be a challenge in any setting, and more so in developing countries. As retrospective cohort studies are prone to several biases, the current recommendation of a uniform target of 90% to 95% remains the standard of care.

Antenatal corticosteroids have been shown to lower the risk of severe ROP. A prospective cohort study by Travers et al demonstrated that infants born between 23 and 30 weeks GA and exposed to antenatal corticosteroids had lower rate of severe ROP.³⁴  The American College of Obstetrics and Gynecologists recommends that 1 course of antenatal corticosteroid therapy should be administered to women at 24 to 35 weeks GA if they are at high risk for preterm delivery in the next 7 days.³⁵  In addition, the POINTS of Care framework in the NICU has been suggested to reduce the risk of ROP by addressing pain control, oxygen management, infection prevention, nutrition, temperature control, and supportive care measures.³⁶  Hyperglycemia has also been suggested as a potential independent risk factor for the development of severe ROP.³⁷  Optimizing parenteral nutrition has been shown to reduce the risk of any stage ROP.³⁸  Risk of severe ROP may be reduced with supplementation of vitamin E, inositol, and breast feeding, but this has only been shown in observational studies.³⁸ 

It has been estimated that 184 700 (169 600–214 500) premature infants developed any stage of ROP worldwide in 2010, of which around 10.7% developed severe visual impairment.¹  Two-thirds belonged to middle income countries. ROP-related blindness may be associated with the infant mortality rate (IMR) of a country. Countries with intermediate levels of IMR reported the highest incidence of ROP.^1,39  The global epidemiologic trend of ROP in the last 2 decades has been variable depending on high-income, upper-middle, and low-middle income countries (HIC, UMIC and LMIC, respectively). There is scant data from low-income countries where survival rates among the tiniest babies is very low. In a worldwide survey on ROP screening policies, no data on ROP screening was forthcoming from 49 countries.⁴⁰  Among the survey respondents, 14 countries (mostly LMIC) did not have any ROP screening program.⁴⁰  There is an ongoing “third epidemic” of A-ROP in LMICs around the world, representing up to 30% of all type 1 ROP cases.^41–43  Not withstanding interobeserver variations, Table 3 provides an overview of the incidences of ROP across the world.

The current third wave of ROP in sub-Saharan Africa has been recently recognized and reported.^44,45  With improved nursery care, neonatal mortality rates in sub-Saharan Africa have decreased by 40% since 1990.⁴⁴  Since sub-Saharan Africa accounts for 28% of preterm births globally,⁴⁶  as neonatal care expands and neonatal mortality rates decline, more of these surviving preterm infants will be at risk for developing severe ROP. It is likely that as time goes on additional sub-Saharan African countries will experience a higher proportion of blindness from ROP.⁴⁷ 

In a recent survey of sub-Saharan African ophthalmologists and neonatologists, although the median number of infants receiving supplemental oxygen per nursery was 15, the median number of oxygen blenders per unit was 0.⁴⁸  Most units lacked the in-wall oxygen and air lines required for oxygen blenders, resulting in most neonates receiving unregulated oxygenation that in turn can explain the increasing incidence of severe ROP. The most important intervention to decrease the incidence of ROP in these countries is to look at regulating oxygen administration. Many countries experienced an ROP epidemic before instituting oxygen monitoring and regulation.⁴⁹  An automated system to assist regulation of oxygen blender based on real-time digital oximetry reading is not yet commercially available. Better oxygen delivery management could reduce the morbidity caused by ROP for preterm infants across sub-Saharan Africa, in the United States, and worldwide.

ROP screening programs use evidence-based criteria to identify those premature infants at risk for severe ROP and subsequent vision loss or blindness if not treated in a timely manner.⁵⁰  Neonatologists have a critical role in the screening of premature infants and identifying those infants who will need an ophthalmology examination according to BW, GA, or unstable neonatal clinical course. Currently, the ROP screening guidelines for preterm infants vary between countries, especially between developed and developing countries. We have attempted to list all of the countries throughout the world with published ROP screening guidelines in Table 4.

Once screened and referred by the neonatologists, ophthalmologists, or ophthalmic photographers, then the retina is observed to stage the severity of ROP, often called a “screening” exam. Such a retinal exam by the ophthalmologist is “screening” to identify ROP severe enough to require treatment. ROP screening follow-up guidelines per the American Academy of Pediatrics are shown in Table 5. Screening eye exams can generally be terminated when there is: complete retinal vascularization¹ ; zone III vascularization without previous zone I or II ROP; no type I ROP or worse at 45 weeks PMA; or regression of ROP in zone III without abnormal vascular tissue capable of reactivation in zone II or III.⁵¹  Infants that have been treated for ROP need life-long eye examinations, especially in the first 5 years of life.^52–54  Reactivation of ROP is more common after treatment with anti-VEGF drugs, therefore meticulous arrangements for on-going eye examinations after anti-VEGF therapy are critical to identify and manage cicatricial ROP and possible retinal detachment with vision loss.^2,55–59 

Screening examinations are resource intensive and stressful to the infants. ROP eye examination increases Crying Requires xygen Increased Vital Signs Expression Sleep (CRIES) pain score and can lead to bradycardia.^60,61  Furthemore, phenylephrine and cyclogyl used in dilated fundus examination can lead to tachycardia and systemic hypertension, and feeding intolerance, respectively in preterm infants.⁶²  Efforts should be made to allow safe and efficient retinal examination of multiple scheduled infants in the NICU.⁶²  There have been significant research efforts to develop innovative evidence-based ROP screening algorithms with greater specificity.⁶³  Although current screening criteria have high sensitivity, the lower specificity leads to a high screening to detection ratio. In a 29-hospital retrospective cohort study from the United States and Canada, only 12.5% of the screened infants developed severe ROP.⁶⁴  SCREENROP, a Canadian nationwide study that analyzed 32 potential predictors and various prediction models using a cohort of nearly 5000 premature babies across all tertiary nurseries in Canada,⁶⁵  found that only babies with a BW of <1200g or a GA of <30 weeks need to be screened to capture those that require ROP treatment, although the current Canadian screening guidelines include infants of BW <1251g or of GA <31 weeks. Determining screening guidelines with both high specificity and sensitivity has been a long standing effort around the globe to limit the number of infants requiring an ROP examination but ensuring that those infants most at risk for needing treatment are not missed. As mentioned in the previous section, infants most at risk for severe ROP and vision loss will vary from country to country, especially from high to middle and low income countries.

Research efforts aimed at refining screening guidelines have been innovative and contributed to our understanding and knowledge. In 2006, Lofqvist et al published the Weight, Insulin-like growth factor-1 (IGF-1), Neonatal, Retinopathy of Prematurity (WINROP) model.⁶⁶  WINROP’s screening algorithm used postnatal weight measurements and serum IGF-1 levels to monitor risk of ROP development. It was postulated that the application of WINROP would eliminate screening in 20% of the infants.⁶⁶  In WINROP2, Hellstrom et al modified the WINROP study to exclude serum IGF-1 levels and used postnatal weight gain as a proxy to decrease stress on the infants.⁶⁷  Various WINROP validation studies in several countries have shown it to have small variability in sensitivity (84.7% to 100%) and larger variability in specificity (23.9% to 89%).^68–74  Although initially WINROP was a useful tool, helping to identify up to 25% of the babies for earlier termination of screening,⁷⁵  more recently with an increase in the target oxygen saturation of the babies, it lost its ability to predict which babies will develop severe ROP.⁷⁶ 

Other predictive models for ROP that include a measure of post natal weight gain include the Children’s Hospital of Philadelphia (CHOP) ROP model⁷⁷  and the more recent postnatal growth and retinopathy of prematurity (G-ROP) model.⁷⁸  Unlike CHOP ROP, the G-ROP model has shown 100% sensitivity in validation studies in terms of identifying all babies developing type 1 ROP.

The goals of treatment of ROP include both the prevention of vision loss or blindness and the preservation of retinal architecture. Treatment of ROP with ablative surgery of the peripheral avascular retina has been used for many decades. Laser photocoagulation was introduced in the early 1990s and studies showed it to be superior to cryotherapy.⁷⁹  Comparisons of the outcomes after cryotherapy versus laser treatment have revealed a higher percentage of poor structural and functional outcome, higher frequency of refractive errors, particularly myopia, and systemic complications with cryotherapy.^52,80,81  Currently, laser therapy is viewed by many as the standard of care, with a regression rate of the ROP of approximately 90%.^82,83 

Laser treatment is performed under local, general, or conscious sedation anesthesia, and a near confluent pattern of laser burns is delivered via laser indirect ophthalmoscopy to the avascular peripheral retina (Fig 3).⁸⁴  Some authors also suggest treatment on, and anterior to, the ridge in eyes that do not regress despite conventional laser treatment.^85,86  British guidelines for ROP⁸⁷  recommend skilled ophthalmologists who regularly perform the treatment^88,89  to reduce the number of retreatments. For this purpose, in regions where the transportation of the infant is practical, centralization of the treatment is preferable and is performed in countries such as Denmark⁹⁰  and Netherlands.⁹¹ 

With adequate laser photocoagulation, the peripheral retina is destroyed, reducing the risk of further angiogenesis and recurrence of disease. This is a great advantage, reducing the long-term follow-up when the infants are more combative with retinal exams and difficult to ensure outpatient follow up. However, there are a few disadvantages and complications with laser therapy, such as cataracts, anterior segment ischemia, and glaucoma.^92–94  Despite laser treatment, ROP may infrequently progress to retinal detachment (stage 4 or 5).⁹⁵ 

Pharmacological therapy provides the ability to achieve both prevention of vision loss and preservation of retina and may facilitate physiologic retinal vascular development after treatment.⁹⁶  The retinal expression of VEGF is closely related to retinal vascular development, therefore using intravitreal anti-VEGF therapies (Fig 5) may be the preferred treatment option for aggressive and type I ROP. Efficacy, drug selection, dosage of drugs, and potential short- and long-term complications have been extensively investigated over the last 15 years.

Four intravitreal anti-VEGF drugs have been reportedly used for the treatment of ROP: bevacizumab (Avastin; Genentech, South San Francisco, CA); ranibizumab (Lucentis; Gememtecj, South San Francisco, CA); aflibercept (Eylea, Regeneron, Tarrytown, NY); and conbercept (Chengdu Kanghong Biotech Co., Ltd., Sichaun, China). Most of these drugs are used “off label” in North America and are not FDA approved for the indication of ROP treatment, although ranibizumab has been recently approved by the European Medicines Agency (EMA) and Health Canada.

Bevacizumab is the most used drug for the treatment of ROP. It is a humanized monoclonal antibody that blocks all VEGF isoforms.⁹⁷  It has become widely adopted for retinal therapies in adult and pediatric retinal diseases, including ROP. It is the most cost effective of the ROP drugs and has a long systemic half-life of 20 days.⁹⁸  However, it suppresses systemic VEGF for up to 8 to 12 weeks, which is significantly longer than the 3 days for ranibizumab.^99–102 

Intravitreal bevacizumab (IVB) was the first drug to demonstrate short-term safety and efficacy in zone I and posterior zone II ROP. At a dose of 0.625 mg, the Bevacizumab Eliminates the Angiogenic Threat (BEAT)-ROP study reported less reactivation of stage 3 ROP with IVB (4%) than laser (22%) for zone I ROP.¹⁰³  A recent meta-analysis from Wang and Zhang included 17 studies (13 of these studies were not randomized) that compared laser, bevacizumab, and ranibizumab for type I and aggressive ROP.¹⁰⁴  The authors reported similar efficacy and retreatment rates for both therapies, but a higher incidence of complications and myopia associated with laser (odds ratio [OR]: 0.38; 95%CI: 0.19–0.75; P = .005).¹⁰⁴ 

Intravitreal ranibizumab (IVR) has been shown to be as effective as laser in regression of type I ROP in 2 major randomized clinical trials. The first was the “CARE-ROP” study in 2018 which compared IVR to laser at doses of both 0.2 mg and 0.12 mg in regression of Type I ROP, with superior physiologic intraretinal vascularization in the lower dose group. Systemic VEGF levels were not significantly altered from baseline with either doses 2 weeks after therapy.⁹⁶  In 2019, the “Ranibizumab Compared with Laser Therapy for the Treatment of Infants Born Prematurely With Retinopathy of Prematurity” clinical trial reported efficacy of IVR at 0.1 mg and 0.2 mg intravitreal doses compared with laser.⁹⁶  The 0.2mg dose of IVR was superior to laser therapy with fewer unfavorable ocular outcomes in the first 24 weeks and 2 years following therapy.⁹⁶  The long-term safety data from this study will be reported in 5-year extension study.¹⁰⁵ 

Intravitreal aflibercept (IVA) is well established in the treatment of adult retinovascular diseases but is used less commonly for the treatment of ROP. In 2019, a comparison of IVA with IVR in the treatment of ROP reported reactivation in 48.1% in the IVR group and 13.9% in the IVA group, with a 1 year follow up period.¹⁰⁶  There was a statistically significant difference between each group for the time of reactivation (shorter for IVR) and the time to vascularize the peripheral avascular retina (longer for IVA). Although both drugs were effective for the treatment of ROP, IVA was associated with fewer and later reactivation of disease. Vural et al reported regression of disease in 94.4% within 1 week of IVA treatment and late reactivation in 19.4%.¹⁰⁷  The use of IVA in ROP is not as common as IVB or IVR, but it is currently being studied in 2 multicenter randomized clinical trials, NCT04004208 Aflibercept for Retinopathy of Prematurity - Intravitreal Injection Versus Laser Therapy (FIREFLEYE) and NCT04101721, Study to Assess the Efficacy, Safety, and Tolerability of Intravitreal Aflibercept Compared with Laser Photocoagulation in Patients With Retinopathy of Prematurity (BUTTERFLEYE).

The dose of each drug is critical in the delicate balance of therapeutic response, promoting physiologic retinal vascular development and minimizing local and systemic toxicity to the infant.²⁴  Anti-VEGF dose-dependent response to normal retinal vascularization and revascularization has been demonstrated in animal models. In premature infants, dose de-escalating studies using IVB have shown that smaller doses may be as efficacious, but the rate of infants requiring treatment of reactivation of disease is higher.^108–111  Lower doses may result in less systemic suppression of VEGF, which would be important for long term safety in these vulnerable infants still undergoing organogenesis, however, this has not yet been reported.¹¹²  Studies of fluorescein angiograms of the retina postinjection have demonstrated abnormal retinal vascular features not seen after ROP laser, that may be attributed to local drug toxicity.^113,114  Further investigations are needed regarding the most appropriate intravitreal dose.

VEGF is an important angiogenic factor in the development of many organs in the premature infant. Therefore, using anti-VEGF drugs should be carefully studied for possible long-term adverse effects from systemic exposure during this vulnerable developmental period.¹¹⁵  The most preferred pharmacokinetic profile of an intravitreal drug for the treatment of ROP would only require 1 injection; have maximum efficacy with regression of intraretinal and vitreous neovascularization; facilitate physiologic retinal vascular development toward the peripheral retina; be minimally absorbed into the systemic circulation and result in negligible alterations in the premature infant’s baseline serum VEGF levels, thus protecting those organs that require VEGF during the perinatal period. Recent studies have reported both increased risk or no increase in risk for neurodevelopmental delay, secondary to intravitreal anti-VEGF drugs used for the treatment of ROP.^116–122  The question of whether suppressed systemic levels of VEGF in these infants contributes to neurodevelopmental delay is very difficult to answer as extreme prematurity itself is 1 of the greatest risk factors for neurodevelopmental delay.^118,123,124  Most studies in the literature are retrospective, however, very recently, the Rainbow Extension Study published the first prospective evaluation of neurodevelopmental and ophthalmic outcomes after 2 years following treatment. This extension study found no difference in neurodevelopmental outcomes between the 3 treatment groups. We await more randomized clinical trials that evaluate long term neurodevelopmental performance in anti-VEGF therapies.^108,125 

Although ROP specialists are few, improvements in wide-field digital retinal imaging have enabled outreach infants the opportunity to undergo staging and increase accessibility to ROP care. Images captured by nonphysicians can be electronically transferred, read, reported, archived, and accessed for clinical, medico-legal, and research activities.^126–128  Tele-ROP allows doctors or trained nonphysicians who capture images of infants “at risk” of ROP to grade them onsite, or transmit them on a secure digital platform for remote reading performed by an ROP specialist within a stipulated time period.¹²⁶  The true potential of this technology would be reached if images of infants derived from resource-poor settings are returned with a credible report in the shortest possible period.

Pediatrician or neonatologist-led teams can help in scaling up existing or new programs.¹²⁹  This model proposes the use of a low-cost camera-based imaging guided by an ROP specialist remotely. The recent availability of portable, affordable ROP cameras may aid the adoption and propagation of such a model soon.¹³⁰  Finally, advances in automatic software analysis of ROP disease integrated into a tele-ROP platform may help triage the decision and grading process, thereby increasing the number of babies who can be screened and staged in the community.¹³¹ 

Automated image analysis and deep learning systems for ROP have the potential to improve the efficiency, accuracy, and objectivity of ROP diagnosis and quantify disease development and predictive risk.¹³²  Several semiautomated platforms have been evaluated, including ROPtool, Retinal Image multiScale Analysis, and Imaging and Informatics in Retinopathy of Prematurity (i-ROP) Research Consortium, which have demonstrated objective measurement of retinal vascular tortuosity and dilation.^133–136  Of these, ROPtool has shown accurate identification of Plus and Pre-Plus disease from analysis of a single image, and has been validated using wide-field and narrow-field images.¹³³  Very recently, i-ROP evaluated the iROP ASSIST system that combined automatic retinal vessel segmentation, tracing, feature extraction and classification of ROP.¹³⁶ 

Challenges exist within artificial intelligence (AI) deep learning strategies and algorithms in their ability to deliver diagnostic efficacy across varied populations. Campbell et al reported external validation of an AI-based screening severity scale based on image sets from a large ROP telemedicine program in India.¹³⁷  The integration of AI into the screening program detected treatment warranting ROP with 100% sensitivity and 78% specificity.¹³⁷  ROP severity could also be characterized based on the oxygen management capabilities of the neonatal care units, facilitating improved understanding of ROP epidemiology and resources. Although investigations of AI applications in ROP are of great interest and have the potential to improve ROP care, there remains many technical and clinical challenges to implementation, although Vinekar and colleagues have recently published a study in which they integrated the AI diagnosis of ROP stage into a real-world ROP screening model in India.^138,139  As described in a review by Ting et al, the deployment of AI using machine learning and deep learning techniques are hindered by the need to ensure generalizability and explainability, and to overcome associated regulatory and medicolegal issues.⁵⁵ 

Although there have been significant advances regarding our knowledge of ROP globally, best methods for screening and treatment are still debated and evolving. With the advent of technology, ROP care may become more accessible and less resource intensive.

Thank you to both April Ingram, MS and Dr Yasmin Jindani, OD for their assistance in research for this manuscript; and Dr Gerd Holmström, MD, PhD for her input on the treatment of the ROP section and review of the article.

ROP

retinopathy of prematurity

GA

gestational age

BW

birth weight

PMA

postmenstrual age