Abstract
Purpose:
To develop prediction models for severe retinopathy of prematurity (ROP) based on risk factors in preterm Thai infants to reduce unnecessary eye examinations in low-risk infants.
Methods:
This retrospective cohort study included preterm infants screened for ROP in a tertiary hospital in Bangkok, Thailand, between September 2009 and December 2020. A predictive score model and a risk factor-based algorithm were developed based on the risk factors identified by a multivariate logistic regression analysis. Validity scores, and corresponding 95% confidence intervals (CIs), were reported.
Results:
The mean gestational age and birth weight (standard deviation) of 845 enrolled infants were 30.3 (2.6) weeks and 1264.9 (398.1) g, respectively. The prevalence of ROP was 26.2%. Independent risk factors across models included gestational age, birth weight, no antenatal steroid use, postnatal steroid use, duration of oxygen supplementation, and weight gain during the first 4 weeks of life. The predictive score had a sensitivity (95% CI) of 92.2% (83.0, 96.6), negative predictive value (NPV) of 99.2% (98.1, 99.6), and negative likelihood ratio (NLR) of 0.1. The risk factor-based algorithm revealed a sensitivity of 100% (94, 100), NPV of 100% (99, 100), and NLR of 0. Similar validity was observed when “any oxygen supplementation” replaced “duration of oxygen supplementation.” Predictive score, unmodified, and modified algorithms reduced eye examinations by 71%, 43%, and 16%, respectively.
Conclusions:
Our risk factor-based algorithm offered an efficient approach to reducing unnecessary eye examinations while maintaining the safety of infants at risk of severe ROP. Prospective validation of the model is required.
Keywords: Postnatal weight gain, prediction, prevalence, retinopathy of prematurity, risk factors, Thai
Retinopathy of prematurity (ROP) is a disease of abnormal development of immature retinal vessels in preterm infants. The incidence of ROP is inversely correlated to gestational age (GA) and varies from 15% to 67% according to differences in screening criteria, population characteristics, and neonatal care practices.[1] In cases of severe ROP, visual impairment can occur and may be associated with adverse neurodevelopmental outcomes.[2] Timely screening for ROP is imperative to facilitate early intervention and mitigate visual loss.[3]
Birth weight (BW), GA, and oxygen exposure were previously found to increase the risk of ROP.[1,4,5,6] While some factors remain inconclusive, the association between low postnatal insulin-like growth factor-1 (IGF-1) and ROP has led to a growing interest in the use of serum IGF-1.[7] Infant weight gain has been proposed as a surrogate measure for serum IGF-1 in clinical settings due to its unavailability.[8]
Standard ROP screening guidelines in well-developed countries typically recommend screening for infants with a GA of <30–31 weeks and a BW of <1501 g.[3,9] Interestingly, infants who develop ROP in low-to-moderate Human Development Index (HDI) countries tend to have higher BW and GA than those in high-HDI countries.[10] In our center, we observe severe ROP in infants with a GA up to 33 weeks. To include these infants, we extended our GA screening criteria to ≤33 weeks. Unfortunately, ROP screening program has substantial costs, and serial eye examinations with indirect ophthalmoscopy put infants at risk of both stress-related physiologic changes as well as significant pain.[11,12] Identifying ROP risk factors beyond GA and BW could help minimize the number of unnecessary eye examinations and in focusing resources to high-risk infants.
Our objective was to develop prediction models, including a predictive score model and a risk factor-based algorithm, to anticipate severe ROP in preterm Thai infants. Derived from our risk factors, we hypothesized that these prediction models would exhibit high sensitivity and negative predictive values (NPVs) as well as low negative likelihood ratios (NLRs), permitting us to effectively determine preterm infants at risk of severe ROP within our population.
Methods
Study design, population, and setting
This was a retrospective cohort study conducted at the Department of Pediatrics and the Department of Ophthalmology, Faculty of Medicine Siriraj Hospital, Mahidol University, Bangkok, Thailand, between September 2009 and December 2020. Siriraj Hospital is a tertiary referral center with approximately 7000–9500 deliveries reported per year. At Siriraj Hospital, an experienced pediatric ophthalmologist performed indirect ophthalmoscopy for ROP screening. The screening criteria included infants with a BW ≤1500 g, GA ≤33 weeks, or preterm infants with an unstable clinical course according to neonatologists. We included all infants who met the Siriraj ROP screening criteria. Infants with unavailable ophthalmologic and weight records throughout the first 28 days of life (DOL), who were deceased or lost to follow-up before DOL 42, or were referred from other hospitals after DOL 14 were excluded.
This study’s protocol was approved by the Siriraj Institutional Review Board, Faculty of Medicine Siriraj Hospital, Mahidol University, Bangkok, Thailand (COA Si 331/2021).
Study procedures and data collection
Infants’ first eye examinations took place at 4 weeks chronological age. Follow-up eye examinations were scheduled until the retinal vessels were fully developed per previous ophthalmologic findings.
Eligible infants were evaluated using the tenth revision of the International Classification of Diseases (ICD-10) for ROP (H35.1) and the ninth revision, Clinical Modification (ICD-9-CM) for extended ophthalmologic work-ups for retinal disease (9503) and eye examinations or procedures not otherwise specified (9509). Demographic and clinical data (including ophthalmologic data) of mother–infant dyads were extracted from medical records. Data was collected and managed in REDCap® electronic data capture hosted by Mahidol University.
Operational definitions
The data collector team defined the operational definitions used in this study. Following Papile’s classification, intraventricular hemorrhage (IVH) included grades I–IV.[13] Infants ≤32 weeks of gestation or with a BW <1500 g were routinely screened, and the most severe grade observed in serial brain ultrasounds was recorded. Severe IVH included both grades III and IV. Diagnosis of hydrocephalus, including both congenital and posthemorrhagic hydrocephalus, was obtained from radiologic reports. The Early Treatment for ROP randomized trial for ROP typing was used to classify severe ROP (types 1 and 2 combined).[14] The cumulative days of oxygen therapy was determined by the number of days supplemented oxygen (fraction of inspired oxygen [FiO2] >0.21) was administered across different modes of delivery. The difference between weight on DOL 28 and BW determines weight gain during the first 4 weeks of life. We chose this variable because the first eye examinations took place at 4 weeks chronological age. Infants with unstable conditions were not weighed and were assigned an average weight. The body weight on DOL 28 was determined by the average for weights available from DOL 25 to 31.
Imputation of missing data
Thirty infants (3.6%) had no data for weight gain during the first 4 weeks of life. This missing data was imputed using three methods: assigned as “weight gain >325 g” (best case scenario), “≤325 g” (worst case scenario), or daily weight gain patterns of infants with equivalent GA. For example, an infant with a GA of 24 weeks would have a weight gain pattern from DOL 1 (BW) to 31 which is based on all infants whose GA is from 24 weeks 0 days to 24 weeks 6 days. This was used to impute body weight from DOL 25 to 31. Immediate postoperative body weight after sacral teratoma removal (performed on DOL 1) was used in place of actual body weight on DOL 1 for one infant. Four out of 30 infants recorded severe ROP.
Statistical analyses
Sample size calculations for prediction model development followed the four-step approach described by Riley et al.[15] A sample size of 849 was calculated based on a severe ROP prevalence of 8% and a margin of error of 0.05.
Data analyses was performed using Predictive Analytics Software version 18 (SPSS Inc., Chicago, IL, USA). Types 1 and 2 ROP were grouped together under severe ROP. Nonsevere ROP included no ROP and other ROP. Comparisons between continuous variables were performed using t-tests and were presented as mean and standard deviation (SD) for normally distributed data or using Mann–Whitney tests and were presented as median and interquartile range (IQR) for non-normally distributed data. Using cutoff points with the highest sensitivity and specificity obtained from a receiver operating characteristic (ROC) curve, continuous data (GA, BW, and weight gain) were transformed into categorical data. Comparisons between categorical data were performed using Pearson’s Chi-square or Fisher’s exact test. Results were presented as odds ratio (OR) with 95% confidence interval (CI). Multivariate logistic regression analysis was performed, and the results were presented as adjusted ORs with 95% CI. P value <0.05 was considered statistically significant.
Development of prediction models for severe ROP
Predictive score model: From the logistic regression analysis, independent risk factors were selected as parameters for the predictive score model. Scoring of each parameter was based on the Wald statistic from the multivariate logistic regression, rounded to the nearest digit. The cutoff value for the predictive model’s total score was determined by the ROC curve. Based on Youden’s J statistic, the optimal cutoff value to discriminate infants at risk of severe ROP was the total score with the highest J. Diagnostic performance metrics including sensitivity, specificity, predictive values, and likelihood ratios were reported.
Risk factor-based algorithm model: Independent risk factors and hydrocephalus were selected as criteria in the algorithm. Meeting at least one criterion qualified the infant being at risk of severe ROP.
Results
Of 962 eligible infants, 117 were excluded [Fig. 1]. ROP screening eye examinations were conducted on 845 infants. Of these infants, 824 (97.5%) met the criteria of GA ≤33 weeks or BW ≤1500 g, while the remaining 21 infants were screened due to neonatologists’ concerns. Table 1 lists the demographic and perinatal data of mother–infant dyads.
Figure 1.
Flow diagram of patient enrollment
Table 1.
Demographic and perinatal data of all infants
| All infants (n=845) | Type 1 ROP (n=38) | Type 2 ROP (n=26) | Other ROP (n=154) | No ROP (n=627) | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Neonatal data | ||||||||||
| Gestational age, week | ||||||||||
| Mean (SD) | 30.3 (2.6) | 27.1 (2.2)a,b | 26.8 (1.5)c,d | 28.1 (2.1)a,c,e | 31.2 (2.1)b,d,e | |||||
| Median (min, max) | 30.6 (24, 36.9) | 26.4 (24, 32.9) | 26.6 (24.4, 31.1) | 27.9 (24, 33.7) | 31.3 (24, 36.9) | |||||
| Birth weight, gram | ||||||||||
| Mean (SD) | 1264.9 (398.1) | 844.9 (189.0)a,b | 780.0 (127.7)c,d | 960.9 (267.6)a,c,e | 1385.1 (368.4)b,d,e | |||||
| Median (min, max) | 1240 (370, 2940) | 805 (540, 1260) | 775.0 (580, 1030) | 930 (370, 1820) | 1350 (560, 2940) | |||||
| Female, n (%) | 402 (47.6) | 18 (47.4) | 12 (46.2) | 72 (46.8) | 300 (47.8) | |||||
| Cesarean section, n (%) | 544 (64.4) | 17 (44.7)b | 15 (57.7) | 90 (58.4)e | 422 (67.3)b,e | |||||
| Multiple births, n (%) | 210 (24.9) | 8 (21.1) | 4 (15.4) | 33 (21.4) | 165 (26.3) | |||||
| Small for gestational age, n (%) | 160 (18.9) | 5 (13.2) | 4 (15.4) | 29 (18.8) | 122 (19.5) | |||||
| APGAR at 1 min (n=835) | ||||||||||
| Median (IQR) | 6 (4, 8) | 4 (1.75, 6) | 2 (1, 6) | 5 (2, 7) | 7 (4, 8) | |||||
| Less than 4, n (%) | 228 (27.3) | 16 (45.7)b | 18 (69.2)c,d | 61 (39.9)c,e | 133 (21.5)b,d,e | |||||
| APGAR at 5 min (n=832) | ||||||||||
| Median (IQR) | 8 (7, 9) | 7 (5, 8) | 7 (2, 8) | 7 (5.5, 9) | 9 (7, 9.75) | |||||
| Less than 4, n (%) | 64 (7.7) | 5 (14.3)b | 8 (30.8)c,d | 20 (13.2)c,e | 31 (5)b,d,e | |||||
| Maternal data | ||||||||||
| Thai nationality, n (%) | 798 (94.4) | 36 (94.7) | 25 (96.2) | 140 (90.9) | 597 (95.2) | |||||
| Maternal diabetes, n (%) (n=479) | 92 (10.9) | 5 (13.2) | 0 | 9 (5.8) | 78 (12.5) | |||||
| Maternal hypertension, n (%) | 199 (23.6) | 4 (10.5) | 6 (23.1) | 37 (24.0) | 152 (24.2) | |||||
| Maternal chorioamnionitis, n (%) | 67 (7.9) | 5 (13.2) | 3 (11.5) | 16 (10.4) | 43 (6.9) | |||||
| Antenatal steroid, n (%) | 745 (88.2) | 28 (73.7)b | 19 (73.1)d | 131 (85.1) | 567 (90.4)b,d |
APGAR, appearance, pulse, grimace, activity, and respiration; IQR, interquartile range; Max, maximum; Min, minimum; N, number; ROP, retinopathy of prematurity; SD, standard deviation. Statistically significant differences (P<0.5) are marked with letters (a, b, c, d, e). aComparison between type 1 and other ROP. bComparison between type 1 and no ROP. cComparison between type 2 and other ROP. dComparison between type 2 and no ROP, eComparison between other and no ROP
Over the 10-year study period, the prevalence of ROP and severe ROP was 26.8% and 7.6%, respectively. Thirty-four (4%) infants required treatment for their ROP, of which 33 (97%) were diagnosed with type 1 and one (3%) with type 2. Infants who developed types 1 and 2 ROP had lower mean GA and BW compared to those who did not. Maximum reported GA and BW of infants with types 1 and 2 ROP were 33 weeks and 1260 g, respectively [Table 1].
To identify the risk factors for severe ROP, imputed data on “weight gain during the first 4 weeks of life” were derived using daily weight gain patterns. Factors associated with severe ROP by univariate analysis are listed in Table 2. Variables independently associated with an increased risk of severe ROP after performing binary logistic regression are listed in Table 3. Independent risk factors were then selected for the predictive score model. Each factor was assigned a weight based on the Wald statistics shown in Table 3. The ROC curve analysis revealed a cutoff point of 15, with the highest area under the curve of 0.92 (95% CI: 0.89, 0.95). The corresponding diagnostic performance metrics, and their 95% CIs, are shown in Table 4. Five infants with severe ROP were not identified by the predictive score model – two of them required ROP treatment. Overall, the predictive score model reduced the number of eye examinations by 71%.
Table 2.
Factors associated with severe ROP by univariate analysis
| n | Nonsevere ROP (n=781) | Severe ROP (n=64) | OR (95% CI) or P | |||||
|---|---|---|---|---|---|---|---|---|
| GA (weeks), mean (SD) | 845 | 30.6 (2.4) | 27.0 (1.9) | <0.001* | ||||
| GA ≤28 weeksa, n (%) | 845 | 136 (17.4) | 54 (84.4) | 25.6 (12.7, 51.6)* | ||||
| BW (g), mean (SD) | 845 | 1301.5 (389.2) | 818.5 (168.8) | <0.001* | ||||
| BW <1050 ga, n (%) | 845 | 206 (26.4) | 58 (90.6) | 27.0 (11.5, 63.5)* | ||||
| Female, n (%) | 845 | 372 (47.6) | 30 (46.9) | 1 (0.6, 1.6) | ||||
| Cesarean sectiona, n (%) | 845 | 512 (65.6) | 32 (50) | 0.5 (0.3, 0.9)* | ||||
| APGAR at 1 min <4, n (%) | 835 | 194 (25.1) | 34 (55.7) | 3.8 (2.2, 6.4)* | ||||
| APGAR at 5 min <4, n (%) | 832 | 51 (6.6) | 13 (21.3) | 3.8 (1.9, 7.5)* | ||||
| Maternal diabetes, n (%) | 479 | 87 (19.4) | 5 (16.7) | 0.8 (0.3, 2.2) | ||||
| Maternal hypertension, n (%) | 845 | 189 (24.4) | 10 (15.6) | 0.6 (0.3, 1.2) | ||||
| Maternal chorioamnionitis, n (%) | 845 | 59 (7.6) | 8 (12.5) | 1.7 (0.8, 3.8) | ||||
| Not receiving antenatal corticosteroidsa, n (%) | 845 | 83 (10.6) | 17 (26.6) | 3.0 (1.7, 5.5)* | ||||
| RDS required surfactanta, n (%) | 845 | 123 (15.7) | 33 (51.6) | 5.7 (3.4, 9.6)* | ||||
| PDA required treatmenta, n (%) | 845 | 267 (34.2) | 48 (75.0) | 5.8 (3.2, 10.4)* | ||||
| NEC stage ≥II, n (%) | 845 | 103 (13.2) | 13 (20.3) | 1.7 (0.9, 3.2) | ||||
| Culture-proven sepsisa, n (%) | 845 | 88 (11.3) | 17 (26.6) | 2.8 (1.6, 5.2)* | ||||
| Hydrocephalusa, n (%) | 845 | 18 (2.3) | 5 (7.8) | 3.6 (1.3, 10.0)* | ||||
| Severe grade IVHa, n (%) | 845 | 31 (4.0) | 13 (20.3) | 6.2 (3.0, 12.5)* | ||||
| Postnatal steroida, n (%) | 845 | 51 (6.5) | 22 (34.4) | 7.5 (4.2, 13.5)* | ||||
| Cumulative number of PRC transfusions during the first 4 weeks, median (IQR) | 845 | 1 (0, 2) | 5 (3, 6) | <0.001* | ||||
| Cumulative number of PRC transfusions during the first 4 weeks ≥3a, n (%) | 845 | 194 (24.8) | 54 (84.4) | 16.3 (8.2, 32.7)* | ||||
| Cumulative days of oxygen supplementation during the first 4 weeks, median (IQR) | 845 | 5 (1, 16) | 26 (20, 28) | <0.001* | ||||
| Cumulative days of oxygen supplementation during the first 4 weeks (>17 days)a, n (%) | 845 | 187 (23.9) | 53 (82.8) | 15.3 (7.8, 29.9)* | ||||
| Weight gain during the first 4 weeks of life (g), mean (SD) | 845 | 417.7 (178.7) | 217 (115.4) | <0.001* | ||||
| Weight gain during the first 4 weeks of life ≤325 ga, n (%) | 845 | 238 (30.5) | 56 (87.5) | 16.0 (7.5, 34.0)* |
APGAR=Appearance, pulse, grimace, activity, and respiration, BW=Birth weight, CI=Confidence interval, GA=Gestational age, IQR=Interquartile range, IVH=Intraventricular hemorrhage, NEC=Necrotizing enterocolitis, OR=Odds ratio, PDA=Patent ductus arteriosus, PRC=Packed red cell, RDS=Respiratory distress syndrome, ROP=Retinopathy of prematurity, SD=Standard deviation. aVariables that were included in the multivariate logistic regression analysis. *Significant at P<0.05
Table 3.
Multivariate logistic regression analysis for risk factors of severe retinopathy of prematurity
| n | Wald | aOR (95% CI) | P | |||||
|---|---|---|---|---|---|---|---|---|
| GA ≤28 weeks | 845 | 8.0 | 3.5 (1.5, 8.3) | 0.005 | ||||
| BW <1050 g | 845 | 9.1 | 4.7 (1.7, 13.0) | 0.003 | ||||
| No antenatal steroid | 845 | 5.7 | 2.5 (1.2, 5.5) | 0.017 | ||||
| Postnatal steroid | 845 | 4.9 | 2.2 (1.1, 4.5) | 0.026 | ||||
| Cumulative days of oxygen supplementation (>17 days) during the first 4 weeks of life | 845 | 5.6 | 2.6 (1.2, 5.7) | 0.018 | ||||
| Weight gain during the first 4 weeks of life ≤325 g | 845 | 7.6 | 3.3 (1.4, 7.8) | 0.006 |
aOR=Adjusted odds ratio, BW=Birth weight, CI=Confidence interval, GA=Gestational age
Table 4.
Validity of the predictive score model and the risk factor-based algorithm to predict severe retinopathy of prematurity
| Predictive score | Risk factor-based algorithma |
|||||||
|---|---|---|---|---|---|---|---|---|
| Daily weight gain pattern | Best-case scenario | Worst-case scenario | ||||||
| Sensitivity, % (95% CI) | 92.2 (83.0, 96.6) | 100 (94.3, 100) | 100 (94.3, 100) | 100 (94.3, 100) | ||||
| Specificity, % (95% CI) | 76.3 (73.2, 79.2) | 46.6 (43.1, 50.1) | 46.7 (43.3, 50.2) | 44.9 (41.5, 48.5) | ||||
| Negative predictive value, % (95% CI) | 99.2 (98.1, 99.6) | 100 (99.0, 100) | 100 (99.0, 100) | 100 (99.0, 100) | ||||
| Positive predictive value, % (95% CI) | 24.2 (19.2, 29.9) | 13.3 (10.6, 16.6) | 13.3 (10.6, 16.7) | 13.0 (10.3, 16.2) | ||||
| Negative likelihood ratio, % (95% CI) | 0.1 | 0 | 0 | 0 | ||||
| Positive likelihood ratio, % (95% CI) | 3.9 | 1.9 | 1.9 | 1.8 | ||||
CI=Confidence interval. aAccounted for gestation ≤28 weeks, birth weight <1050 g, no antenatal steroid, received postnatal steroid, cumulative oxygen supplementation (>17 days), and weight gain ≤325 g during the first 4 weeks of life, and hydrocephalus
Risk factors used in the predictive score model were selected to develop a risk factor-based algorithm. Hydrocephalus was included as a criterion due to its potential to cause nonphysiologic weight gain, expanding the algorithm’s framework to seven criteria. Infants meeting at least one criterion were identified as being at risk of severe ROP. In addition to imputing missing data using daily weight gain patterns, two alternative scenarios (best case and worst case) were tested. Table 4 summarizes these diagnostic performance metrics and their corresponding 95% CIs. The risk factor-based algorithm reduced the number of eye examinations by 43%.
Refinement of the risk factor-based algorithm was accomplished by replacing the variable “cumulative days of oxygen supplementation during the first 4 weeks of life (>17 days)” with “≥1 day of oxygen supplementation during the first 4 weeks of life.” The diagnostic performance metrics and their corresponding 95% CIs are presented in Table 5. The modified algorithm resulted in a 16% reduction in performed eye examinations.
Table 5.
Validity of the modified risk factor-based algorithm to predict severe retinopathy of prematurity
| Modified risk factor-based algorithma |
||||||
|---|---|---|---|---|---|---|
| Daily weight gain pattern | Best-case scenario | Worst-case scenario | ||||
| Sensitivity, % (95% CI) | 100 (94.3, 100) | 100 (94.3, 100) | 100 (94.3, 100) | |||
| Specificity, % (95% CI) | 17.3 (14.8, 20.1) | 17.4 (14.9, 20.2) | 16.3 (13.8, 19) | |||
| Negative predictive value, % (95% CI) | 100 (97.2, 100) | 100 (97.3, 100) | 100 (97.1, 100) | |||
| Positive predictive value, % (95% CI) | 9 (7.1, 11.4) | 9 (7.1, 11.4) | 8.9 (7, 11.2) | |||
| Negative likelihood ratio, % (95% CI) | 0 | 0 | 0 | |||
| Positive likelihood ratio, % (95% CI) | 1.2 | 1.2 | 1.2 | |||
CI=Confidence interval. aAccounted for gestation ≤28 weeks, birth weight <1050 g, no antenatal steroid use, received postnatal steroid, at least 1 day of oxygen supplementation, and weight gain ≤325 g during the first 4 weeks of life, and hydrocephalus
Discussion
We developed a predictive score model and a risk factor-based algorithm to anticipate severe ROP in preterm Thai infants. The prevalence of ROP and severe ROP at Siriraj Hospital was 26.8% and 7.6%, respectively. Approximately 4% of infants who underwent a screening eye examination required treatment for ROP. After adjusting for potential confounding factors, independent risk factors associated with severe ROP included GA ≤28 weeks, BW <1050 g, not receiving antenatal steroids, postnatal steroid administration, >17 days of cumulative oxygen supplementation, and weight gain ≤325 g during the first 4 weeks of life. Two prediction models for severe ROP were developed from these factors. The predictive score model had promising sensitivity as well as NPV and NLR values, reducing the number of eye examinations by 71%. However, five cases of severe ROP were not accounted for, including two of which required treatment. The risk factor-based algorithm exhibited improved sensitivity as well as NPV and NLR values, reducing unnecessary eye examinations by 43%. The modified algorithm, especially useful in resource-limited settings, exhibited similar sensitivity as well as NPV and NLR values, reducing eye examinations by 16%.
Lower GA and BW were consistently determined as risk factors for ROP in several studies,[1,4,5,6] including ours. Both criteria have been used as universal criteria for ROP screening.[6] However, GA and BW of infants who developed severe ROP in our center were higher than those of infants from developed countries.[10] Our expanded criteria resulted in a significant clinical workload and the potential for complications due to eye examination. Of the 21 infants screened based on neonatologists’ concerns, none developed severe ROP. It is crucial to explore the risk factors beyond GA and BW to spare low-risk infants from unnecessary eye examinations.
A recent meta-analysis found that antenatal steroid administration was a preventive factor for ROP of any stage, particularly severe ROP.[5] These findings aligned with our study. Angiogenesis and inflammation are reduced through the repression of tumor necrosis factor alpha following antenatal steroid administration.[5,16] Furthermore, administration also mitigates injuries caused by reactive oxygen species, through increased antioxidant enzyme activity and decreased oxidative stress markers in premature infants.[17] In contrast, pathogenesis of ROP can be promoted through postnatal steroid administration that alters IGF-1 and vascular endothelial growth factor expression.[18] It may be possible that receiving postnatal steroids represented prolonged ventilator exposure, as demonstrated by our study where more oxygen supplementation led to a greater risk of severe ROP. O’Donovan and Fernandes[19] explained that due to inadequate antioxidant defense systems in premature infants, high arterial oxygen concentrations can lead to ROP. This aligned with a recent systematic review which suggested that the duration of oxygen supplementation was a significant risk factor for ROP.[4]
Despite the use of different time frames to determine weight gain, low weight gain was found to be associated with severe ROP and served as a surrogate marker for lower levels of serum IGF-1.[20,21,22] Hellstrom et al.[7] demonstrated that serum IGF-1, required for proper retinal blood vessel development, is associated with ROP in preterm infants. A recent meta-analysis regarding weight gain-based algorithms for ROP screening developed for different populations revealed adequate sensitivity and NLR to rule out severe ROP and minimize the need for unnecessary eye examination.[23] However, a recent meta-analysis revealed that the algorithm yielded lower sensitivity, specificity, and NLR in some low-to-middle-income countries compared to high-income countries.[23] This may be partly due to different perinatal care systems across countries. It is prudent that different countries apply unique screening criteria to ensure coverage for all infants at risk of severe ROP and safely reduce the number of unnecessary eye examinations in low-risk infants.
Maternal hypertension and chorioamnionitis did not exhibit an association with severe ROP in our study. This aligned with a meta-analysis that found inconclusive results regarding the correlation between maternal hypertension and ROP of any stage and severe ROP (pooled OR [95% CI]: 1.12 [0.90, 1.40] and 0.80 [0.47,1.35], respectively).[24] Conversely, chorioamnionitis was demonstrated as a risk factor for ROP. This association may be partially because chorioamnionitis can lead to preterm birth.[25] A recent study identified histologic chorioamnionitis and funisitis as new risk factors for ROP of any stage and severe ROP (OR [95% CI]: 1.8 [1.4, 2.2] and 1.4 [1.1, 1.6], respectively).[26] A recent retrospective cohort study reported a positive association between maternal diabetes and ROP, while others found conflicting results.[4,27] Unfortunately, substantial portions of maternal diabetes data were missing, preventing us from drawing definitive, associative conclusions.
Ideal ROP screening programs can definitively determine whether an infant requires eye examinations, while maintaining absolute certainty that infants not at risk of severe ROP are excluded. To safely exclude these infants, prediction models must possess a sensitivity and NPV of 100% and an NLR of zero. This study developed a predictive score model for severe ROP with promising sensitivity (92.2%), NPV (99.2%), and NLR (0.1). As misclassification can lead to severe visual outcomes, for example, retinal detachment or blindness, missing five cases of severe ROP – two requiring treatment – is not acceptable.
Hydrocephalus was added to the risk factor-based algorithm due to its potential to cause nonphysiologic weight gain.[20] The seven-criteria algorithm proved more robust, with a sensitivity and NPV of 100% and an NLR of 0%. In addition, its validity and robustness were demonstrated consistently across the three data imputation methods. Compared to the predictive score model, the seven-criteria risk factor-based algorithm reduced eye examinations by 43%, rather than 71%. While the predictive score model projects more examination omissions, infants’ safety takes precedent. Thus, the risk factor-based algorithm provides a safer means to exclude low-risk infants from unnecessary eye examinations.
The “duration of oxygen exposure” criterion in the seven-criteria risk factor-based algorithm applies to a neonatal intensive care unit setting with well-equipped oxygen blender to control FiO2 and pulse oximeter to monitor oxygen saturation (SpO2) during oxygen supplementation. In hospitals with limited resources, the use of oxygen blenders and pulse oximeters may prove unfeasible. Brief exposure to high FiO2, with inadvertently high SpO2, may put infants at risk for severe ROP. By replacing “duration of oxygen exposure” with “any exposure to oxygen” in our modified algorithm, similar sensitivity, NPV, and NLR were obtained. While the reduction in examinations decreased only by 16%, the safety of infants was prioritized. This ensured that no infants with severe ROP were overlooked. Further developing this algorithm to be web or mobile phone based may promote accessibility and usage.
The integration of artificial intelligence (AI) into telemedicine programs presents a novel solution to overcome the challenges of ROP screening in regions with limited access to pediatric ophthalmologists.[28,29] Coyner et al.[29] demonstrated a reduction in eye examinations for low-risk infants using AI-integrated telemedicine to screen for ROP. This innovative method, in conjunction with a fundus photograph, allows ophthalmologists to fulfill infant screening criteria and monitor for ROP.
Despite the large number of infants studied over a 10-year period, further prospective validation is required due to the study’s retrospective, single-centered design. Data on body weight measurements were missing. This was addressed through data imputation, with the robustness of results confirmed through worst- and best-case scenario analyses. Conclusive results regarding the association between maternal diabetes and severe ROP were not obtained due to unavailable maternal diabetes data. Furthermore, changes in neonatal care (from invasive to noninvasive respiratory support), surfactant and steroid usage, and patent ductus arteriosus management should be noted as potential limitations.
Conclusion
This study highlights the importance of tailored ROP risk factor assessment. Derived from specific risk factors of preterm Thai infants, our risk factor-based algorithms significantly reduced unnecessary eye examinations while maintaining the safety of infants at risk. However, this predictive model has a major limitation regarding history of maternal diabetes in assessing the risk factor to predict severe ROP in preterm Thai infants. Further prospective studies that externally validate the models in a larger population of preterm Thai infants are required.
Abbreviations
aOR, adjusted odds ratio; AUC, area under the curve; BW, birth weight; CI, confidence interval; COA, certificate of approval; DOL, day of life; GA, gestational age; HDI, Human Development Index; IGF-1, insulin growth factor-1; IVH, intraventricular hemorrhage; NLR, negative likelihood ratio; NPV, negative predictive value; OR, odds ratio; ROC, receiver operating characteristic; ROP, retinopathy of prematurity.
Financial support and sponsorship
Nil.
Conflicts of interest
There are no conflicts of interest.
Acknowledgements
We thank Nutchavadee Vorasan of Siriraj Genomics, Faculty of Medicine Siriraj Hospital, Mahidol University, Thailand, for helping us with the use of the REDCap database. We also thank Katherine Copeland and Satsavat Limpanich for their English editing.
References
- 1.Raghuveer TS, Zackula R. Strategies to prevent severe retinopathy of prematurity: A 2020 update and meta-analysis. Neoreviews. 2020;21:e249–63. doi: 10.1542/neo.21-4-e249. [DOI] [PubMed] [Google Scholar]
- 2.Morken TS, Dammann O, Skranes J, Austeng D. Retinopathy of prematurity, visual and neurodevelopmental outcome, and imaging of the central nervous system. Semin Perinatol. 2019;43:381–9. doi: 10.1053/j.semperi.2019.05.012. [DOI] [PubMed] [Google Scholar]
- 3.Fierson WM; American Academy of Pediatrics Section on Ophthalmology; American Academy of Ophthalmology; American Association for Pediatric Ophthalmology and Strabismus; American Association of Certified Orthoptists Screening examination of premature infants for retinopathy of prematurity. Pediatrics. 2018;142:e20183061. doi: 10.1542/peds.2018-3061. [DOI] [PubMed] [Google Scholar]
- 4.Kim SJ, Port AD, Swan R, Campbell JP, Chan RVP, Chiang MF. Retinopathy of prematurity: A review of risk factors and their clinical significance. Surv Ophthalmol. 2018;63:618–37. doi: 10.1016/j.survophthal.2018.04.002. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Yim CL, Tam M, Chan HL, Tang SM, Au SCL, Yip WWK, et al. Association of antenatal steroid and risk of retinopathy of prematurity: A systematic review and meta-analysis. Br J Ophthalmol. 2018;102:1336–41. doi: 10.1136/bjophthalmol-2017-311576. [DOI] [PubMed] [Google Scholar]
- 6.Bahmani T, Karimi A, Rezaei N, Daliri S. Retinopathy prematurity: A systematic review and meta-analysis study based on neonatal and maternal risk factors. J Matern Fetal Neonatal Med. 2022;35:8032–50. doi: 10.1080/14767058.2021.1940938. [DOI] [PubMed] [Google Scholar]
- 7.Hellstrom A, Engstrom E, Hard AL, Albertsson-Wikland K, Carlsson B, Niklasson A, et al. Postnatal serum insulin-like growth factor I deficiency is associated with retinopathy of prematurity and other complications of premature birth. Pediatrics. 2003;112:1016–20. doi: 10.1542/peds.112.5.1016. [DOI] [PubMed] [Google Scholar]
- 8.Lo HC, Tsao LY, Hsu WY, Chi CY, Tsai FA. Changes in serum insulin-like growth factors, not leptin, are associated with postnatal weight gain in preterm neonates. JPEN J Parenter Enteral Nutr. 2005;29:87–92. doi: 10.1177/014860710502900287. [DOI] [PubMed] [Google Scholar]
- 9.Royal Colleges of Paediatrics and Child Health Screening of retinopathy of prematurity (ROP)-clinical guideline. 2022 Available from: https://www.rcpch.ac.uk/resources/screening-retinopathy-prematurity-rop-clinical-guideline. [updated 2022] [Google Scholar]
- 10.Gilbert C, Fielder A, Gordillo L, Quinn G, Semiglia R, Visintin P, et al. Characteristics of infants with severe retinopathy of prematurity in countries with low, moderate, and high levels of development: Implications for screening programs. Pediatrics. 2005;115:e518–25. doi: 10.1542/peds.2004-1180. [DOI] [PubMed] [Google Scholar]
- 11.Mukherjee AN, Watts P, Al-Madfai H, Manoj B, Roberts D. Impact of retinopathy of prematurity screening examination on cardiorespiratory indices: A comparison of indirect ophthalmoscopy and retcam imaging. Ophthalmology. 2006;113:1547–52. doi: 10.1016/j.ophtha.2006.03.056. [DOI] [PubMed] [Google Scholar]
- 12.Sun X, Lemyre B, Barrowman N, O’Connor M. Pain management during eye examinations for retinopathy of prematurity in preterm infants: A systematic review. Acta Paediatr. 2010;99:329–34. doi: 10.1111/j.1651-2227.2009.01612.x. [DOI] [PubMed] [Google Scholar]
- 13.Papile LA, Burstein J, Burstein R, Koffler H. Incidence and evolution of subependymal and intraventricular hemorrhage: A study of infants with birth weights less than 1,500 gm. J Pediatr. 1978;92:529–34. doi: 10.1016/s0022-3476(78)80282-0. [DOI] [PubMed] [Google Scholar]
- 14.Early Treatment For Retinopathy Of Prematurity Cooperative Group Revised indications for the treatment of retinopathy of prematurity: Results of the early treatment for retinopathy of prematurity randomized trial. Arch Ophthalmol. 2003;121:1684–94. doi: 10.1001/archopht.121.12.1684. [DOI] [PubMed] [Google Scholar]
- 15.Riley RD, Ensor J, Snell KIE, Harrell FE, Jr, Martin GP, Reitsma JB, et al. Calculating the sample size required for developing a clinical prediction model. BMJ. 2020;368:m441. doi: 10.1136/bmj.m441. [DOI] [PubMed] [Google Scholar]
- 16.Yossuck P, Yan Y, Tadesse M, Higgins RD. Dexamethasone alters TNF-alpha expression in retinopathy. Mol Genet Metab. 2001;72:164–7. doi: 10.1006/mgme.2000.3124. [DOI] [PubMed] [Google Scholar]
- 17.Vento M, Aguar M, Escobar J, Arduini A, Escrig R, Brugada M, et al. Antenatal steroids and antioxidant enzyme activity in preterm infants: Influence of gender and timing. Antioxid Redox Signal. 2009;11:2945–55. doi: 10.1089/ars.2009.2671. [DOI] [PubMed] [Google Scholar]
- 18.Smolkin T, Steinberg M, Sujov P, Mezer E, Tamir A, Makhoul IR. Late postnatal systemic steroids predispose to retinopathy of prematurity in very-low-birth-weight infants: A comparative study. Acta Paediatr. 2008;97:322–6. doi: 10.1111/j.1651-2227.2008.00629.x. [DOI] [PubMed] [Google Scholar]
- 19.O’Donovan DJ, Fernandes CJ. Free radicals and diseases in premature infants. Antioxid Redox Signal. 2004;6:169–76. doi: 10.1089/152308604771978471. [DOI] [PubMed] [Google Scholar]
- 20.Binenbaum G, Bell EF, Donohue P, Quinn G, Shaffer J, Tomlinson LA, et al. Development of modified screening criteria for retinopathy of prematurity: Primary results from the postnatal growth and retinopathy of prematurity study. JAMA Ophthalmol. 2018;136:1034–40. doi: 10.1001/jamaophthalmol.2018.2753. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Cao JH, Wagner BD, McCourt EA, Cerda A, Sillau S, Palestine A, et al. The Colorado-retinopathy of prematurity model (CO-ROP): Postnatal weight gain screening algorithm. J AAPOS. 2016;20:19–24. doi: 10.1016/j.jaapos.2015.10.017. [DOI] [PubMed] [Google Scholar]
- 22.Wongnophirun A, Khuwuthyakorn V, Tantiprabha W, Wiwatwongwana A. Association between severe retinopathy of prematurity and postnatal weight gain in very low-birthweight infants at Chiang Mai University Hospital, Thailand. Paediatr Int Child Health. 2020;40:85–91. doi: 10.1080/20469047.2019.1631588. [DOI] [PubMed] [Google Scholar]
- 23.Athikarisamy S, Desai S, Patole S, Rao S, Simmer K, Lam GC. The use of postnatal weight gain algorithms to predict severe or type 1 retinopathy of prematurity: A systematic review and meta-analysis. JAMA Netw Open. 2021;4:e2135879. doi: 10.1001/jamanetworkopen.2021.35879. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Zhu T, Zhang L, Zhao F, Qu Y, Mu D. Association of maternal hypertensive disorders with retinopathy of prematurity: A systematic review and meta-analysis. PLoS One. 2017;12:e0175374. doi: 10.1371/journal.pone.0175374. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Villamor-Martinez E, Cavallaro G, Raffaeli G, Mohammed Rahim OMM, Gulden S, Ghazi AMT, et al. Chorioamnionitis as a risk factor for retinopathy of prematurity: An updated systematic review and meta-analysis. PLoS One. 2018;13:e0205838. doi: 10.1371/journal.pone.0205838. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.El Emrani S, Jansen EJS, Goeman JJ, Lopriore E, Termote JUM, Schalij-Delfos NE, et al. Histological chorioamnionitis and funisitis as new risk factors for retinopathy of prematurity: A meta-analysis. Am J Perinatol. 2023 doi: 10.1055/a-2215-0662. doi: 10.1055/a-2215-0662. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Opara CN, Akintorin M, Byrd A, Cirignani N, Akintorin S, Soyemi K. Maternal diabetes mellitus as an independent risk factor for clinically significant retinopathy of prematurity severity in neonates less than 1500 g. PLoS One. 2020;15:e0236639. doi: 10.1371/journal.pone.0236639. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Campbell JP, Singh P, Redd TK, Brown JM, Shah PK, Subramanian P, et al. Applications of artificial intelligence for retinopathy of prematurity screening. Pediatrics. 2021;147:e2020016618. doi: 10.1542/peds.2020-016618. doi: 10.1542/peds. 2020-016618. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Coyner AS, Oh MA, Shah PK, Singh P, Ostmo S, Valikodath NG, et al. External validation of a retinopathy of prematurity screening model using artificial intelligence in 3 low- and middle-income populations. JAMA Ophthalmol. 2022;140:791–8. doi: 10.1001/jamaophthalmol.2022.2135. [DOI] [PMC free article] [PubMed] [Google Scholar]