Patients and Methods
Study Design
We performed a population-based, cross-sectional study using the KID to estimate the incidence of ROP and to determine predictors of ROP among at-risk newborns. All research adhered to the tenets of the Declaration of Helsinki. Institutional review board (IRB) approval was not required by the Stanford IRB, as this was a retrospective study that did not involve unique patient identifiers.
Data Source
The KID is maintained by the Agency for Healthcare Research and Quality as part of the Healthcare Cost and Utilization Project (HCUP).
7
It is the largest publicly available pediatric inpatient care database in the United States, incorporating discharge data from 2,500 hospitals across 22 states in 1997 to more than 4,100 hospitals across 44 states in 2012. The KID uses systematic random sampling to select 10% of uncomplicated in-hospital births and 80% of complicated in-hospital births and other pediatric cases from each frame hospital. The sample frame includes pediatric (20 years or less) discharges from community and nonrehabilitation hospitals in the participating HCUP Partner States. Using the KID database compiled for the three most recent years (2006, 2009, and 2012), we performed a retrospective analysis of all newborns diagnosed with ROP. Data compiled for the years 2000 and 2003 were also included to analyze trends in ROP incidence and in-hospital mortality over time. We excluded the data compiled for the year 1997, when the KID first started, as several changes were made to the database in 2000 that would cause a discontinuity between the 1997 and 2000 data, including revising the definition of total discharges. Missing data are reported for each variable. Mortality refers specifically to death during the hospitalization.
Case Identification
We performed a retrospective review of the KID for all reported cases of ROP as defined by the International Classification of Diseases, 9th Revision (ICD-9) medical diagnosis codes (Table
A). We limited our population of interest to newborns with length of stay (LOS) longer than 28 days to limit the effect of mortality on the dataset, as most newborn deaths in premature infants occur within the first week. Furthermore, this allowed us to capture the population at greatest risk for ROP as a stay of this length indicates an unstable clinical course, one of the criteria for ROP screening.
8–9
Newborns who underwent laser photocoagulation, scleral buckle, and/or vitrectomy were counted as cases. Demographic data was recorded in the database. Birth weights, estimated gestational age (EGA), comorbidities, and surgical interventions were assessed by determining the weighted frequencies of the recorded ICD-9 codes.
Outcomes
Outcomes of newborns with and without ROP were examined, including LOS, inpatient mortality, and associated complications identified with ICD-9 codes: birth trauma, intrauterine hypoxia, chronic respiratory disease, respiratory distress syndrome (RDS), perinatal infection, fetal hemorrhage, intraventricular hemorrhage, necrotizing enterocolitis (NEC), periventricular leukomalacia, continuous invasive mechanical ventilation, noninvasive mechanical ventilation, lack of respiratory support, and blood transfusions (Supplemental Table). In patients with ROP, the proportions receiving laser photocoagulation, vitrectomy, and scleral buckle were calculated and stratified by birth weight.
Statistical Analysis
The data were analyzed with SAS Enterprise Guide 7.1 (SAS Institute, Cary, NC). Variables were tested for normality with the Kolmogorov-Smirnov test to assess for outliers and to determine the appropriate statistical test. Crude univariate (unadjusted) testing was used to compare the cohort with and without ROP and to calculate the odds ratio (OR) and 95% confidence interval (CI) as measures of the association between ROP and each potential risk factor. All continuous variables were analyzed by the Student t test while dichotomous variables were analyzed by the Rao–Scott Chi-square test.
Predictors of ROP were then assessed using multivariate logistic regression modeling performed based on crude univariate analysis and review of the available literature. The model included clinically significant predictors with a
P
value of less than .10 and less than 20% missing data in the univariate analyses. Backward selection was performed on the predictors and the final multivariate model retained predictors with a
P
value of less than .05. The final model, therefore, contained independent risk factors for ROP. We used the area under receiver operating characteristic curve (AUROC) to estimate the prognostic ability of the model to discriminate between patients with and without ROP. We adjusted for overfitting using 10-fold cross validation.
All data were weighted according to HCUP recommendations prior to analysis to calculate national estimates.
Results
Patient Demographics
Table
1
provides the demographic characteristics of the newborns in this study with and without ROP with an LOS greater than 28 days, compiled for the years 2006, 2009, and 2012. On univariate analysis, newborns with ROP were more likely to be female and to have been identified by their parents as black than non-ROP controls. Conversely, infants identified by their parents as Native American had lower odds of ROP development. Although the relationship between black race and ROP became nonsignificant after controlling for EGA, the relationship between Native American race and ROP remained significant even after controlling for sex, EGA, and birth weight. There was no difference in the type of insurance (private vs. other) between newborns with ROP and non-ROP controls. Newborns with EGA less than 36 weeks had an increased odds of ROP. The odds were highest in newborns with an EGA of 24 weeks.
Effect of Birth Weight on ROP and LOS
Newborns classified by the World Health Organization as extremely low birth-weight (ELBW; less than 1,000 grams [g]) and very low birth weight (VLBW; 1,000 g to 1,499 g) with a LOS greater than 28 days had significantly higher odds of developing ROP as compared to normal birth weight (Table
1
). A peak frequency of ROP of 30.22% occurred in newborns between 750 g and 999 g. Table
2
shows the mean LOS in newborns with and without ROP stratified by birth weight. For the same birth weight, infants with ROP had a significantly higher LOS than newborns without ROP (
P
< .0001) for all weight categories. The greatest discrepancy in LOS occurred between newborns weighing less than 500 g with and without ROP.
Incidence of Retinopathy of Prematurity in Dataset
Of the 18,822,055 discharges recorded during the three representative years analyzed (2006, 2009, and 2012), 27,481 newborns were identified with ROP. Within our population of interest of 153,706 newborns with LOS of more than 28 days, the total incidence of ROP was 17.9%. Table
3
presents the incidence of ROP and in-hospital mortality during the following years of the KID database: 2000, 2003, 2006, 2009, 2012. The overall incidence of ROP has increased from 14.70% to 19.88%, whereas in-hospital mortality among newborns with ROP and LOS longer than 28 days has ranged from 0.92% in 2000 to 0.45% in 2012 (
Figure
).
Complications and Comorbidities Associated with Newborns with ROP
The complications and comorbidities in newborns with and without ROP and with an LOS of longer than 28 days are shown in Table
4
. On univariate analysis, compared to newborns without ROP, newborns with ROP had an increased odds of chronic respiratory disease, RDS, perinatal infection, fetal hemorrhage, intraventricular hemorrhage, continuous invasive mechanical ventilation, noninvasive mechanical ventilation, and blood transfusions. Compared to newborns without ROP, newborns with ROP had a lower odds of birth trauma, intrauterine hypoxia, NEC, and lack of respiratory support. There was no significant difference in the odds of periventricular leukomalacia between newborns with and without ROP. The most common comorbidities in both ROP and non-ROP newborns were RDS, continuous invasive mechanical ventilation, and perinatal infection.
Multivariate Logistic Regression
We assessed the following variables as possible risk factors in regression modeling based on crude univariate analyses and review of the current literature: sex, race, EGA, birth weight, continuous invasive mechanical ventilation, noninvasive mechanical ventilation, lack of respiratory support, RDS, and intraventricular hemorrhage (Table
5
). Backward selection was performed on the predictors and the final multivariate model retained predictors with a
P
value of less than .05, including sex, EGA, birth weight, and lack of respiratory support. The AUROC for this model was 0.703, but 10-fold cross validation revealed a true AUROC of 0.695 (95% CI, 0.691–0.698). The overoptimism was 0.008, indicating a very low degree of overfitting. Newborns with female sex, EGA less than 36 weeks, and birth weight less than 2,000 g had significantly increased odds of developing ROP among infants with a LOS greater than 28 days. Significance of any one variable did not change after adjusting for in-hospital mortality.
Surgical Interventions for ROP
Overall, 8.31% (2,284 of 27,481) of newborns required treatment of ROP prior to discharge from the hospital. Of these newborns, 8.21% (2,257 of 27,481) underwent laser photocoagulation, 0.18% (49 of 27,481) underwent pars plana vitrectomy, and 0.03% (nine of 27,481) underwent scleral buckle. Incidence of laser photocoagulation in infants with ROP was highest in infants with weight less than 500 g (45.19%; 156 of 345), followed by infants weighing between 500 g and 749 g (29.03%; 1,144 of 3,941).
Discussion
Our study represents the most updated report of ROP using a nationwide cohort of newborns. In contrast to previous retrospective analyses using the National Inpatient Sample (NIS) database, which includes patients of all ages, our current study uses the KID sample, the largest publicly-available inpatient care database in the United States focusing on the pediatric population. The design of the KID, with an emphasis of sampling complicated in-hospital births from each frame hospital, makes it ideal for research on rare pediatric conditions requiring hospitalization, including ROP.
10
Our cohort represents the largest group of premature newborns in this time period with ROP that has been examined (more than 26,000 infants with ROP out of 18.8 million discharges over three study years). Our final predictive model had an AUROC of 0.695, demonstrating that the model can successfully distinguish between the diseased and healthy subpopulations in our study cohort.
Within our population of interest for newborns with a LOS greater than 28 days, the ROP incidence increased by one-third from 15.6% between 1997 and 2005 to 17.9% in our study spanning the years 2006 to 2012.
5
One important factor leading to this jump may be the widespread implementation of the American Academy of Pediatrics and American Academy of Ophthalmology ROP screening guidelines in 2006. The guidelines recommended that all infants with weights less than or equal to 1,500 g, who have an EGA equal to or less than 32 weeks (now less than 30 weeks with revised 2013 guidelines), and who have an unstable clinical course or are at high risk for developing ROP as determined by an attending pediatrician or neonatologist should be serially evaluated until the criteria for screening termination are met.
8–9
Furthermore, advancements in life-preserving technologies have led to increased survival of premature and LBW babies.
11
As EGA and LBW are both well-known to be inversely related to ROP risk,
12–15
these developments in neonatal care have increased the population of babies at risk for ROP.
16–18
In addition to contributing to the increasing incidence of ROP with more systematic screening guidelines and advancements in neonatology, these factors may also play a role in decreasing the overall in-hospital mortality in newborns with ROP (Table
3
).
In our analysis, premature infants with ROP were significantly more likely to be female. This held true even after adjustment for EGA and birth weight. This is consistent with the 1997 to 2005 NIS data,
5
but differs from the CRYO-ROP results in which no difference in sex was found in patients with or without ROP.
19
Further evaluation is warranted to conclusively determine the relationship between sex and ROP.
As in prior studies, EGA, VLBW, and ELBW were strongly associated with ROP (Table
1
).
5–6,12–15
In this study, 30% of patients diagnosed with ROP were 1,250 g or greater. This is almost a 50% increase in ROP incidence in this weight category compared to the 1997 to 2005 NIS study
5
and illustrates the importance of evaluating newborns who meet the age criteria for ROP screening regardless of their birthweight.
8–9
Of note, we identified a small percentage of normal-weight infants with ROP, which has increased by more than 60% from the 0.19% reported in the 1997 to 2005 NIS study.
5
Infants in this group were likely to satisfy EGA criteria for ROP screening, as they were on average of lower EGA than VLBW and ELBW infants.
8–9
Univariate analysis showed that the newborns in our sample with ROP and a LOS of longer than 28 days had significantly higher incidence of intraventricular hemorrhage (IVH), perinatal infection, blood transfusion, and mechanical ventilation than newborns without ROP. However, all relationships were nonsignificant after controlling for birth weight and EGA. Our results did indicate that babies not exposed to oxygen therapy had lower odds of developing ROP, which is consistent with previous studies.
20–24
Several complications and comorbidities had a significantly lower incidence in our population of interest as compared to all newborns with LOS longer than 28 days, including birth trauma, intrauterine hypoxia, and necrotizing enterocolitis. The mechanisms of action to explain these trends have not been studied, and prior studies examining comorbidities have shown conflicting results, making it difficult to interpret the significance of these associations.
5,25–28
One possibility is the suggestion by Lad et al. that aggressive treatment of comorbidities may explain the negative association with ROP.
6
The overall percentage of newborns with ROP and a LOS greater than 28 days who required treatment prior to discharge (8.31%) is similar to the percentage reported in the 1997 to 2005 NIS study (> 8.18%).
5
In contrast, an Australian study found that approximately 17% of patients diagnosed with ROP from 1992 to 2009 required laser therapy.
29
The discrepancy may be a result of greater use of nonsurgical treatment options for ROP, namely the anti-VEGF inhibitors bevacizumab (Avastin; Genentech, South San Francisco, CA) and ranibizumab (Lucentis; Genentech, South San Francisco, CA).
30–37
The KID does not have records of any newborns receiving intravitreal injections of bevacizumab or ranibizumab. However, we believe this is due to the lack of an accurate ICD-9 code for off-label use of these drugs, or that providers were using CPT codes instead, which are not captured by the KID.
Our study draws from a large, nationwide pediatric database to report the incidence of ROP in the United States. The KID has several advantages over the NIS, including its emphasis on sampling complicated in-hospital births from each frame hospital, making it ideal for study of rare diseases. However, the KID is similarly limited by its capture of inpatient data alone, failing to record morbidity and mortality that occurs outside of the hospital setting. Furthermore, analysis of the database is limited by accurate and consistent classification of ICD-9 codes by providers. Likewise, the severity of several comorbidities and the extent and duration of exposures is often not captured by ICD-9 codes. For instance, the volume of blood transfusions or duration of oxygen therapy, which are both known risk factors for ROP, were not addressed in our study as the data is not recorded through ICD-9 codes. Moreover, the KID does not capture longitudinal data on individual patients, so long-term outcomes and efficacy of treatments are also not available. Lastly, as with any retrospective study, missing data within the KID prevent a complete assessment of the entire population of interest, and may affect calculated odds ratios.
ROP is a devastating disease that can lead to childhood blindness if not detected and treated early. In the present study, we surveyed more than 18 million live births in the United States between 2006 and 2012, with an overall ROP incidence of 17.9% among newborns with a length of stay greater than 28 days. Our results confirm existing knowledge that LBW and EGA are important risk factors for the development of ROP, and suggest that sex and lack of respiratory support are independently associated with ROP. Finally, we show that the incidence of ROP has been increasing from 14.70% in 2000 to 19.88% in 2012, a trend inversely related to the decline in newborn mortality in the United States in the same time period.