Evaluation of modified retinopathy of prematurity screening guidelines using birth weight as the sole inclusion criterion.

PURPOSE: The purpose was to determine if birth weight (BW) alone can be the sole criterion for screening infants at risk for retinopathy of prematurity (ROP).
MATERIALS AND METHODS: In this retrospective, observational case series, 208 infants were screened for ROP using the American Association for Pediatric Ophthalmology and Strabismus (AAPOS) Guidelines (1997). Variables examined included gestational age (GA), birth weight (BW), and a composite variable BWGA Index [(grams × weeks)/1000], which takes into consideration both the birth weight and gestational age of the infant. Infants were divided into two groups: Group 1, BW ≤1250 g, and Group 2, BW >1250 g. Multivariate analysis was performed to detect factors predictive of ROP. Receiver operator characteristic (ROC) curves were generated to determine the efficacy of screening using the BW, GA, and BWGA Index. Statistical analyses were performed with logistic regression with a P-value of 0.05 or less indicating statistical significance.
RESULTS: Varying stages of ROP were present in 116 of 416 eyes. Of the 105 eyes in Group 2, only 1 eye developed stage 1 ROP. Only Group 1 eyes developed stage 3 or higher ROP. The ROC curve for BW alone gave an area under the curve (AUC) of 0.797 (standard error [SE] = 0.0329, P < 0.0001); for GA, AUC was 0.801 (SE = 0.0340, P < 0.0001) and for the BWGA Index, the AUC was 0.808 (SE = 0.0324, P < 0.0001). Using 1250-g BW as a criterion for ROP screening would have decreased the number of screenings by 24%, and did not exclude any ROP higher than stage 1.
CONCLUSION: Data from our neonatal intensive care unit suggest that birth weight ≤ 1250 g alone is an adequate parameter to identify premature infants at risk for ROP.

INTRODUCTION

Retinopathy of prematurity (ROP), in its advanced stages, is a visually devastating disease that warrants early detection and timely treatment. Smaller neonates with birth weight (BW) ≤1250 g are at the highest risk of developing advanced ROP and thus have a less favorable course than infants born with BW >1250 g.1–3

The previous guidelines for ROP screening from the Joint Statement of the American Academy of Pediatrics (AAP), the American Association for Pediatric Ophthalmology and Strabismus (AAPOS), and the American Academy of Ophthalmology (AAO) mandate screening for infants with BW ≤1500g or gestational age (GA) ≤28 weeks, with infants up to 2000 g to be screened at the discretion of the attending pediatrician or neonatologist.12 This was later revised in 2006 to include infants of GA ≤ 30 weeks.3 The need to include neonates with BW greater than 1250 g in the screening process is controversial. There has been evidence that using a revised screening criterion of BW less than 1250 g or GA less than 30 weeks results in up to 40% decrease in the number of infants screened.4–8

However, the risk of missing sight-threatening ROP may outweigh any benefits.

In this study, we retrospectively evaluated whether a revised guideline of BW < 1250 g alone has any impact on reducing the number of screenings without missing sight-threatening ROP. The prevalence of ROP and its sequelae were analyzed in a sample of premature infants with GA < 28 weeks or BW < 1500 g. We used the 1997 and 2001 guidelines for this study, as the subjects included were screened between 1999 and 2002.12 The screening variables were then examined to determine whether BW alone that we used in this study was sufficient for efficient ROP detection and treatment. Specifically, BW ≤1250 g was tested as the sole parameter used for ROP screening in this retrospective study.

MATERIALS AND METHODS

We conducted a retrospective chart review of infants screened for ROP by the Department of Ophthalmology at the neonatal intensive care unit (NICU) of a university hospital. An approval for this study was obtained from the local Institutional Review Board. All infants born at or before 32-week gestation or weighing ≤ 1500 g were included in the screening process. Two hundred eight infants born between July 1999 and January 2002 were screened. An experienced retinal surgeon performed dilated, indirect, funduscopic exams, using a 30 D lens with an eyelid speculum and scleral depression, under topical anesthesia. The initial examination was performed at a 6-week chronological age, or at a 31- to 33-week GA, whichever was earlier.12 All eyes examined were staged according to the International Classification for ROP by the Committee for the Classification of Retinopathy of Prematurity.9

Weekly examinations were conducted in eyes with ROP. Infants without ROP were examined every 2 weeks until the retina was completely vascularized. All infants with threshold ROP underwent laser treatment under general anesthesia in the operating room and were followed weekly until resolution. Treatment was based on the guidelines available at the time.10

The data collected during examination included the maximum stage and zone of each eye, whether the eye was treated, the presence of plus disease, BW, GA, race, and gender.11 A composite variable BWGA Index which takes into consideration both the BW and GA of the infant was calculated using the formula (BW × GA)/1000. Infants were grouped according to BW with Group 1 consisting of infants weighing ≤1250 g and Group 2 consisting of infants with BW > 1250 g.

We calculated the sensitivity, specificity, and positive predictive value of existing strategies and for the proposed criterion for routine screening for ROP in our population. Data were analyzed with simple and multiple logistic regression analyses to select factors predictive of ROP with Statistical Analysis System, version 8.4, software (SAS Inc., Cary, NC, USA). The predicted probability of ROP was estimated using models based on the GA, BW, and BWGA Index. A P-value of 0.05 or less was considered statistically significant.

Receiver operator characteristic (ROC) curves were generated to determine the efficacy of the proposed screening criteria of the BWGA Index, BW, GA as well as the variable probability of ROP (prob_ROP) which is the estimated probability of ROP given values for BW and GA using the fitted two-variable logistic regression model. An ROC curve is a graph plotting the combination of sensitivity (true-positive rate) and the complement to specificity (that is, 1 – specificity, false-positive rate) across a series of cutoff values covering the whole range of values of a given disease marker. An ROC curve analyzes the discrimination performance of quantitative tests (or criteria, as in our study) throughout the range of possible values and helps identify the optimal cutoff value. The overall discrimination performance of a given test is measured by calculating the area under the ROC curve (AUC). In our study, the AUC was calculated for the above factors using Analyze-It, an Excel add-on (Microsoft Corp., Redmond, WA, USA). Tests of pair-wise comparisons of the AUCs were made at the 5% level of significance.

RESULTS

The cohort demographics are summarized in Table 1. A total of 416 eyes of 208 infants born with a BW of ≤1500 g or at a 28-week GA or less, were screened for ROP. Of the 208 infants, 116 (56%) were males and 92 (44%) were females. One hundred forty-seven (71%) infants were born in the university hospital while 61 (29%) were born in other hospitals but received care at the same Neonatal Intensive Care Unit (NICU).

Table 1

Demographics of the infants included in the study

Retinopathy of prematurity evaluation

The presence or absence of ROP with respect to BW and GA is shown in Figure 1. On fundus evaluation, 116 (28%) eyes had some form of ROP, while no ROP was detected in more than half (72%) of the eyes screened. In the study cohort, 158 (76%) infants were born with a BW ≤ 1250 g (Group 1) and 50 (24%) infants had a BW > 1250 g (Group 2). Of the 100 eyes in Group 2, only 1 eye developed ROP. The BW of the infant was 1375 g, with a GA of 29 weeks. The infant had stage 1 ROP in one eye (OD) that resolved and did not progress. All other eyes with ROP were in Group 1. Table 2 presents the level of ROP in the study of cohort. All eyes with stage 3 and higher ROP were in Group 1 [Table 2]. For the 27 eyes that had stage 3 or higher ROP, 25 were under 1000 g BW. Twenty-six eyes developed threshold disease and underwent laser treatment. Despite treatment, two eyes progressed to advanced ROP and required vitreoretinal surgery.

Figure 1

Birth weight and gestational age plotted with respect to the presence or absence of retinopathy of prematurity. Presence of retinopathy of prematurity = “*,” no retinopathy of prematurity = “+;” horizontal line drawn at BW = 1250 g

Table 2

Retinopathy of prematurity evaluation according to the International classification for retinopathy of prematurity by the committee for the classification of retinopathy of prematurity

Factors predictive of retinopathy of prematurity

The AAO joint guidelines in our population gave a sensitivity of 91.4%, specificity of 40.7%, and a positive predictive power of 37.3% for identifying ROP. Using BW ≤ 1250 as the sole screening parameter, the sensitivity was 98.3% (95% confidence interval [CI]: 90.8–100.0%), specificity 32.7% (95% CI: 25.2–40.8%), positive predictive value (PPV) 35.1%, negative predictive value (NPP) 98.1% and an efficiency of 51.0% [Table 3]. Statistically significant univariate logistic regression models for predicting ROP (with a single variable) were found for BW (P < 0.0001, OR = 0.613 per 100 g), GA (P < 0.0001, OR = 0.578), and BWGA Index (P < 0.0001, OR = 0.86). The multiple logistic regression model with both GA (P = 0.0135) and BW (P = 0.0389) found both variables to be statistically significant. Estimated probabilities of ROP under various values of the screening variables are shown in Figure 2a–c.

Table 3

Sensitivity, specificity, and positive predictive value at existing and proposed criteria

Figure 2

(a) Logistic regression coefficient estimates for retinopathy of prematurity (ROP) using only birth weight (BW). (b) Logistic regression coefficient estimates for ROP using only gestational age (GA). (c) Logistic regression coefficient estimates for retinopathy of prematurity (ROP) using only the BWGA Index [(birth weight in grams × gestational age in weeks)/1000]

ROC curves were used to evaluate the screening power of BW and GA individually, BWGA Index, and the logistic model with the BW and GA as predictors [Figure 3a–c]. The ROC curve for BW as a sole screening criterion gave an AUC of 0.797 (standard error [SE] = 0.0329, P < 0.0001, 95% CI: 0.732–0.861); for GA, the AUC was 0.801 (SE = 0.0340, P < 0.0001, 95% CI: 0.735–0.868) and for the BWGA Index, the AUC was 0.808 (SE = 0.0324, P < 0.0001, 95% CI: 0.744–0.871).

Figure 3

(a) Receiver operator characteristic curve for birth weight. (b) Receiver operator characteristic curve for gestational age. (c) Receiver operator characteristic curve for BWGA Index [(birth weight in grams × gestational age in weeks)/1000]

We found that using 1250 g or less as the screening criterion was probably more efficient than using 1500 g or less (51.0% vs. 28.4%), reducing the number of unnecessary infant screenings in our population by 24%.

DISCUSSION

Existing national screening guidelines for ROP use both BW and GA as parameters, with 1500 g as the cutoff for BW with minor differences in the GA from 33 weeks to 28 weeks.12 The recommendations for screening published by the Royal College of Ophthalmologists (RCO) use the cutoffs of <31-week GA and BW < 1500 g for screening infants and the 1998 Canadian Pediatric Society (CPS) use the parameters of 30-week GA or <1500 g BW.12–15 In our study, the screening criterion of BW alone or GA alone is just as efficient in detecting any ROP as using the combined criteria of BW and GA. AUC estimates for each of BW, GA, BWGA Index, and prob_rop (GA, BW) were similar and showed that the variables under consideration have a good overall screening power. The highest observed AUC was with the logistic model of BW and GA (0.814). None of the pair-wise comparisons of the ROC curves produced statistically significant differences. BW versus GA alone was not statistically different (P = 0.8454).

Studies have questioned the use of 1500 g as the upper limit for the screening protocol.4–8 Lee et al.7 and Ho et al.8 have reported that using BW < 1200 g resulted in a decrease of up to 40% screenings in their study population. All infants in these two studies diagnosed with threshold or stage 4 disease weighed less than 1250 g, ranging from 490 to 1135 g, regardless of GA.78 In addition, no threshold disease was observed in infants from either study population with a BW ≥ 1200 g. Several studies in the United Kingdom have tested the effectiveness of RCO screening guidelines and have questioned the need to include infants with a BW > 1250 g.1415 In these UK-based studies, no infant with a BW of 1251–1500 g developed threshold disease.1415 Alternately, in the single largest series of infants with a BW > 1250 g screened for ROP, Hutchinson et al. have shown that screening parameters should include all infants with a BW < 1500 g or GA < 32 weeks to not miss any infants who require treatment.16 Of 1118 infants (BW > 1250 g) screened, Hutchinson et al.16 reported 7 cases of ROP requiring treatment in infants with a BW ranging from 1600 to 1874 g. However, Hutchinson et al's.16 cohort included many infants with associated systemic conditions, all with a ≤32-week GA.

Outliers in study cohorts have been reported who developed threshold disease despite a BW > 1250 g.8 For example, Ho et al. noted two cases of ROP, both with a 30-week GA and a BW > 1250 g.8 However, both cases had only self-limiting ROP.8 It has also been reported that these outliers, in spite having a higher BW, could have other concurrent risk factors for the development of ROP such as severe neurological injury, maternal diabetes, and perinatal blood loss.416 In our study, only one baby in our population who had disease would have been missed by implementing a criterion of BW ≤ 1250. This infant had one eye with stage 1 disease in zone 2 that did not progress. All patients who developed stage 3 or higher ROP were in Group 1, with the heaviest infant of BW 1076 g. In order to detect ROP in larger infants, a one-time ROP screening has been suggested to catch outliers and the update of the Joint Statement included screening babies with a BW >1500 g determined to be at a high risk by their attending pediatrician or neonatologist.17

Most of the current screening protocols include BW or GA to identify the correct time to begin the screening process.1819 Using a single criterion would provide consistency across national guidelines for a reproducible screening for ROP. In our study, logistic regression and the ROC curves indicated that BW alone is as statistically significant as GA to predict the probability of ROP. BW had a slightly greater sensitivity than GA. It is more accurate to use the neonate's weight at birth than having to rely on the mother's recollection of the last menstrual cycle, which is unknown or unreliable in up to 40% of pregnancies. Ultrasound measurements of the crown–rump length offer improved estimations of the GA; however, a controversy exists regarding the timing of such studies, with a decreasing accuracy as the pregnancy progresses.20 In our study population, using the cutoff of BW <1250 g was more efficient than using a 1500 g BW by 24%. The sensitivity of this criterion was 98.3%, which is 6.9% higher than the AAO guidelines. Using a BW ≤ 1250 g as the sole screening criterion avoids many unnecessary ROP exams, and was successful in the timely diagnosis and treatment of infants with a high risk ROP in our population.

Although our cohort is from a single institution with a small sample size in comparison to the 4099 infants from 70 NICUs in the CRYO-ROP study,21 we had a similar incidence of threshold ROP (5.8% and 6%, respectively) in spite of screening infants with a BW up to 1500 g. The majority (56.7%) of our study population were black infants who appear to develop severe ROP at a significantly lower rate than white and Hispanic infants.21 Our recommendations may not be applicable to other medical centers with a large Hispanic population or populations where larger, more mature babies are developing severe disease, called aggressive ROP.22

Our data suggest that screening infants with a BW cutoff of ≤ 1250 g seems unlikely to miss any ROP higher than stage 1 while being vigilant in high-risk cases and detecting threshold ROP. Specific factors such as neurologic insult, sepsis, candidemia, and blood loss that put larger infants at risk for ROP should be identified. Large, national studies are required to further refine ROP screening guidelines.