Aim: To test the effectiveness of WINROP software tool to screen retinopathy of prematurity (ROP) in Indian preterm infant population including twin neonates.

Methodology: In a retrospective single center study, birth weight, gestational age, comorbidities, and weekly weight measurements (for 5 weeks) were retrieved from 63 preterm infants born between 01/2014 and 04/2015. The obtained data was entered into the WINROP algorithm to obtain ROP outcomes and WINROP alarm.

Results: For a cohort of 63 patients together with twin neonates, the median birth weight was 1250.0 g and gestational age was 30.0 weeks. Of the 63 infants, 22 infants developed Type 1 ROP and 39 infants developed Type 2 ROP. WINROP alarm was triggered in 33 (52.38%) infants. Comorbidities such as malnutrition, respiratory distress syndrome (RDS), blood transfusion, anemia of prematurity and pregnancy-induced hypertension (PIH) was associated with the development of ROP. The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of WINROP to predict type 1 ROP were 63.6%, 53.6%, 42.4% and 73.3%, respectively. In twin neonates, WINROP predicted type 1 ROP with sensitivity, specificity, PPV, and NPV of 100%, 60%, 33.3% and 100%, respectively.

Conclusion: This is the first WINROP validation study in twin neonates from Indian settings. The WINROP model was highly sensitive to detect type 1 ROP in twin neonates, however, due to low specificity and low PPV, the outcome of this study suggests use of WINROP algorithm alongside standard ROP screening in infants including twin neonates with WINROP alarm.

Retinopathy of prematurity (ROP), a gestational age (GA)-related illness, is a leading cause of childhood blindness in premature infants [1]. Pathologically, ROP involves delayed retinal vascularization, vaso-proliferation and intravitreal angiogenesis [2]. Fibrovascular retinal detachment and permanent blindness are the risks associated with ROP [2]. Among the risk factors, low gestational age (GA), low birth weight (BW) and oxygen level are the major risk factors for the development of ROP [1]. Timely screening of premature infants for ROP could improve visual prognosis. The conventional ROP screening involves dilation of pupil and subsequent use of indirect ophthalmoscopy and retinal imaging using a RetCam [3]. However, to minimize the risk and to increase the identification of high-risk infants, researchers have introduced and developed multiple prediction models such as WINROP, CO-ROP, ROP Score and CHOP-ROP [1]. Among these prediction model, studies have indicated sensitivity of WINROP (Weight, Insulin-like growth factor, Neonatal ROP), an online ROP monitoring tool, ranging from 80% to 90% globally among Chinese [4], South African [5] and Australian [6] preterm natal population.

Since, low GA and low BW are the major risk factors in the development of ROP, WINROP prediction model uses weight and gestational age at birth as a dichotomized factor for the screening of ROP [7]. However, studies indicate that heavier and more mature babies can also develop ROP, especially, from middle-income or developing countries such as India indicating role of alternative risk factors in the development of ROP [1]. Thus to accommodate larger number of ROP screening, the National guidelines for the screening for ROP has broad eligibility criteria with gestational age of ≤34 weeks or < 2000 gram or neonates with gestational age above 34 weeks and associated risk factors such as prolonged oxygen support, cardiovascular instability, and sepsis [3].

In the last five years, Indian clinical studies have also adopted the use of WINROP model [8,9] and shown high sensitivity and NPV of the algorithm to predict Type 1 ROP. The model was sensitive to predict Type 1 ROP earliest at 2 weeks than the conventional screening which predicted ROP at 7th week postnatal stage [9]. However, there is scarcity of research, and the existing literature does not completely validate the adoption of WINROP in Indian settings.

Aim and Objectives
The aim of the present study was to test the efficacy of WINROP to predict ROP in Indian preterm infant population. The objective was to (i) To investigate the pattern of ROP in preterm infant population including twin neonates and (ii) To measure the diagnostic performance of WINROP software to predict ROP in twin neonates.

This was a retrospective study conducted on 63 infants born between 01/2014 and 04/2015 and who were at high risk of developing ROP. Infants with the following criteria were included: (i) birth weight less than 1500 grams; (ii) gestational age 30 weeks; (iii) infants with birth weight between 1500 g and 2000 g or (iv) gestational age more than 30 weeks and with an unstable clinical course and at high risk for ROP. The enrolled infant population also included 6 pairs (12 infants) of twin neonates with ROP. Twins were called discordant if their birth weight difference was more than 15%. For twins, the inclusion criteria were that both twin infants were alive. Infants with congestive heart failure, Neonatal nephrotic syndrome, Hydrocephalus excluded due to nonphysiologically weight gain. Data of Final ROP outcome should be available.

WINROP Screening
For WINROP screening following clinical data were retrospectively retrieved including infant’s gestational age (less than 32 weeks at birth), birth weight, associated comorbidity, weekly weight measurements, physiological weight gain and absence of other pathologic retinal vascular disease. The collected data was entered into WINROP software (https://winrop.com/). Clinical examination for ROP was performed weekly/twice in a week and close observation was made to see the progression of ROP to Type 1 ROP. The clinical phenotypes of ROP, namely, Type 1 and Type 2 ROP were classified as per the Early Treatment of Retinopathy of Prematurity (ETROP) cooperative group classification.

Statistical Analysis
The Statistical Package for Social Sciences (SPSS) version 24 was used for descriptive statistics. Association between the variables was estimated by Chi-Square analysis. Independent t-test was performed to evaluate the differences in variables between with and without WINROP group. The sensitivity, specificity, positive predictive values and negative predictive values of the WINROP algorithm were also calculated.

Table 1 illustrates the demographic characteristics and ROP outcomes in 63 preterm infants who were enrolled in this study. The majority were female infants (53.9%) with very low birth weight in the range of 1000-1500 g (69.8%). The median gestational age was 30.0 weeks (30.0-32.0 weeks) and the median birth weight was 1250 g (range: 1052.5-1500.0 g). The median birth weight in the first week was 1310 g (1100.0-1547.50 g) which increased to 1587 g (range: 1353.75-1797.5 g) in the fifth week. Based on birth plurality, out of 63 preterm infants, 51 were single babies and 12 (6 pairs) were twin babies. WINROP alarm was signaled in 52.3% of infants and 39 (61.9%) infants developed Type 2 ROP, 22 (34.9%) developed Type 1 ROP and 2 (3.1%) had no ROP.

Association between birth characteristics and WINROP
Table 2 presents the association of birth characteristics and type of ROP with and without WINROP alarm. There was no association of WINROP alarm with sex (χ² = 0.009, p > 0.05), birthweight (χ² = 3.171, p> 0.05) and type of ROP (χ² = 3.527, p > 0.05). Further, gestational age (t = -1.238, p > 0.05), birthweight (t = -1.788, p > 0.05) and physiological weight gain from first week to fifth week were not associated with WINROP alarm (p > 0.05).

Association between type of ROP and comorbidities
Infants had multiple comorbidities including respiratory distress syndrome (RDS) (76.1%), followed by blood transfusion (44.4%) anemia of prematurity (42.8%), hyaline membrane disease (23.8%) and malnutrition (22.2%). In addition, about 76.1% of infants were born to mother with pregnancy-induced hypertension (PIH). ROP showed highly significantly association with malnutrition (χ² = 20.46, p < 0.001), RDS (χ² = 9.33, p < 0.001) and anemia of prematurity (χ² = 9.58, p < 0.001) and significant association with blood transfusion (χ² = 6.48, p < 0.05) and PIH (χ² = 7.28, p < 0.05). Type 1 ROP was associated with malnutrition and anemia of prematurity and Type 2 ROP was associated with RDS, blood transfusion and PIH.

Effectiveness of WINROP to predict ROP
WINROP alarm was signaled in 33 infants, out of which 14 (42.4%) developed Type 1 ROP and 19 (57.6%) developed non-Type 1 ROP. About 8 infants developed type 1 ROP without any WINROP alarm. In the prediction of type 1 ROP, WINROP tool had a sensitivity of 63.6%, specificity of 53.6%, positive predictive value (PPV) of 42.4% and the negative predictive value (NPV) of 73.3% (Table 3).

WINROP analysis for twin neonates
Table 4 presents the association of birth characteristics and type of ROP with and without WINROP alarm in twin neonates. Sex and type of ROP were not associated with WINROP alarm. The association of WINROP alarm with birth weight was significant (χ² = 4.00, p < 0.05). WINROP alarm was signaled in twins with very low birthweight suggestive of high risk of ROP in infants with very low birthweight (1000-1500 g). Further, the t-test showed significant difference in birth weight between infants with and without WINROP alarm (t = -2.533, p < 0.05) (Table 4). The mean value of birthweight was lower in WINROP alarm group than no alarm group (1092.5 ± 90.7 vs. 1410 ± 293.2 g).

In 6 pairs (12 babies) of twin babies, RDS (83.3%), anemia of prematurity (75%) and hyaline membrane disease (50%) were the most frequent comorbidities. Among the comorbidities in twin babies, malnutrition (χ² = 12.00, p < 0.001) and PIH (χ² = 7.33, p < 0.05) showed significant association with Type 1 ROP, and RDS (χ² = 12.00, p < 0.001) and anemia of prematurity (χ² = 7.33, p < 0.05) showed significant association with Type 2 ROP.

In twin neonates, WINROP alarm was signaled in six infants, out of which two developed Type 1 ROP and six infants with no alarm developed non-Type 1 ROP. The diagnostic performance of WINROP indicated sensitivity of 100%, specificity of 60%, PPV of 33.33% and NPV of 100% to predict type 1 ROP (Table 5).

Clinically, prediction models such as WINROP have been used to detect the risk of ROP in preterm infants [5,10]. The present retrospective study was performed to analyze the predictive ability of WINROP algorithm for the detection of ROP in preterm infant babies in Indian setup. The median BW of 1250 g and GA of 30 weeks was comparable to studies from other Asian population including China [11] and Taiwan (12). WINROP alarm was signaled in 52.3% of infants which was low compared to previous studies from Malaysia (72.8%) [13], India (74.2%) (8) and Saudi (70.9%) [14] but higher than percent of WINROP alarm in study from Australia (42.6%) [6] and another study from India (27.7%) [15]. In the present study, WINROP alarm showed no association with birthweight (p > 0.05), sex (p > 0.05) and type of ROP (p > 0.05). Contrary to the present finding Sute et al (15) showed association of WINROP alarm with gestation age, birthweight and Type 1 ROP. Previous studies have associated multiple comorbidities such as RDS, blood transfusion, dysplasia, large patent ductus arteriosus and septic shock have been associated with ROP (13,15,16), however, in this study, malnutrition (p < 0.001), RDS (p < 0.001), anemia of prematurity (p < 0.01) , blood transfusion (p < 0.05) and PIH (p < 0.05) were associated with ROP. Across different studies (15,16), RDS was the most common comorbidity which was associated with ROP. Studies indicate that disruption of gas exchange leads to multiple complications including increased risk of ROP in preterm infants [17]. Studies state that parenteral nutrition consisting of high energy and protein, and mother’s milk to ensure optimal growth and adiposity in the post-natal weeks may have a protective effect against the development of ROP in preterm infants [18,19]. Likewise, anemia and blood transfusion has been reported as an independent risk factor of ROP in premature infants [20]. On the contrary, in literature the effect of maternal PIH on the occurrence of ROP is not conclusive and requires further analysis [21]. Overall, based on the present findings it can be inferred that multiple neonatal risk factors are associated with the development of ROP and effort should be made to control these factors to reduce the risk of ROP.

In various cohorts, the sensitivity and specificity of WINROP to predict ROP has varied. In the present study the sensitivity to predict type 1 ROP through WINROP was low (63.6%) compared to previous studies from countries such as Saudi (100%) [14], Sweden (100%) [22] and Malaysia (95.2%) [13] which reported high sensitivity but comparable to previous studies from Taiwan (64.7%) [10] and South Africa (72.9%) [5]. Further, the specificity of WINROP was low (53.6%) but comparable to previous studies in literature from Malaysia (33.8%) [13] and Japan (42.7%) [23]. A very few Indian studies have used the WINROP algorithm for the prediction of ROP in preterm infants [8,15]. The use of the WINROP model in Indian cohorts has reported sensitivity of 90.3% [8] and 80% [15], and specificity of 38.4% [8] and 80.6% [15]. Further, the other validation parameter such as PPV of 42.4% and NPV of 73.3% was comparatively lower with the values mentioned in the Indian setup. In our study, 42% [14/33] of infants with WINROP alarm developed type 1 ROP and remaining 57% non-type 1 ROP suggestive of timely testing for the progression of ROP. Based on the given validation parameters of WINROP algorithm, it can be inferred that in infants with WINROP alarm there is a requirement of ROP screening, and this could reduce the number of unnecessary screenings in preterm infants.

The present study is the inclusion of twin neonates and this is the first study from Indian babies using the WINROP model to predict ROP in twin neonates. Multiple studies have reported association of multiple gestations/twins or multiplets with the risk of ROP [1]. Of the two dichotomous variables, namely, birthweight and gestational age which is used in WINROP algorithm [7], twin babies have the same gestational age and perinatal risk factors. Thus, they provide a good model to analyze if birth weight has a role in the progression of ROP [24]. Focused on that, the present study tested the efficacy of the WINROP model to predict ROP in twin neonates. In the present study, in twin babies, WINROP alarm was associated with very low birth weight (p < 0.05) and ROP was significantly associated with comorbidities including malnutrition (p < 0.001), RDS (p < 0.001), anemia (p < 0.05) and PIH (p < 0.05), however, further research is required to ascertain the associated factors in twin neonates with ROP. Previous studies have reported higher risk of ROP in discordant twins with lower birth weights [25], however, Azad et al [24] states that birth weight as a factor to screen ROP in twins should be performed with caution. The authors state that presentation and progression of ROP can vary in twins as heavier siblings were also presented with severe ROP. On the contrary, Sanghvi et al [26] reported that birth order, birthweight and post-gestational neonatal risk factors do not predict the severity of ROP in twins. Furthermore, the WINROP model predicted type 1 ROP in twin babies with a high sensitivity of 100% and NPV value of 100% in this study. As stated by Sanghi et al [8] NPV of 100% presents an ideal situation which can reduce the ROP screening for infants with no alarm. In support of this, Raffa et al [14] argues that since prediction of ROP is important to prevent blindness, sensitivity and NPV are more relevant than other parameters to screen preterm infants with ROP.

The differences in parameters from the previous studies could be multifactorial including the study design, preterm study population, types of ROP, screening criteria for ROP, provisions for perinatal and postnatal care [14,15]. For instance, studies from Indian clinical settings such as Sute et al [15] performed a prospective observational study on 102 singleton preterm infants, whereas our study was retrospective in nature with 63 preterm infants including 12 twin neonates, hence, discrepancies in data within the same geographical settings cannot be avoided. Nevertheless, Ko et al. [10] states that WINROP is an effective tool to predict ROP in infants that meet the criteria of BW and GA of less than 1000 g and less than 28 weeks, respectively. However, for infants from developing country such as Asian preterm infant population where ROP epidemiology and weight gain curve can differ from the developed country, development of individualized algorithm for different geographical zones is recommended [10]. Based on the present findings that WINROP alarm was signaled in non-type 1 ROP also, thus, it is recommended to monitor for the progression of non-Type 1 ROP to Type 1 ROP. In addition, considering the small size of population of twin neonates, further validation studies to precisely estimate the sensitivity and specificity of WINROP algorithm and the generalizability of findings on a larger population of twin neonate is suggested.

The study is limited by retrospective design, small sample size and from a single center. Further studies should include prospective design on a larger infant population including twin neonates and from multicenter. Additionally, to improve the predictive efficacy of the algorithm, WINROP model should be modified to accommodate populations with different characteristics such as larger and older babies, and additionally multiple postnatal risk factors should be incorporated.

Overall, the WINROP model had a moderate sensitivity of 63.6%, low specificity of 53.6%, low PPV of 42.4% and high NPV of 73.3% to predict type 1 ROP. Further, WINROP had a high sensitivity of 100%, a high NPV of 100% but low specificity of 60% and low PPV of 33% to predict type 1 ROP in twin neonates. Based on these performance parameters, it is suggested that WINROP algorithm be used potentially as an accessory tool and standard ROP screening be performed alongside on infants with WINROP alarm.