Aims To determine the incidence of retinopathy of prematurity (ROP) based upon a national database and to identify baseline characteristics, demographic information, comorbidities, and surgical interventions.
Methods The National Inpatient Sample, a representative sample of all US hospital discharges from 1997 to 2002, was queried for all newborn infants with and without ROP. Primary outcome variables included demographics, comorbidities, hospital length of stay (LOS), and hospital charges. Multivariate logistic regression was used to predict risk factors for ROP.
Results 4.67 million live births were recorded during the study period. The total incidence of ROP was 0.12% overall and 7.35% for premature infants with LOS greater than 14 days. Newborns with ROP were more likely to be born at a teaching hospital and to have higher LOS and hospitalisation charges. The odds ratios for the development of ROP were greatest in infants weighing less than 1250 grams. The multivariate regression model revealed that only respiratory distress and intraventricular haemorrhage were predictive of the development of ROP and Hispanic infants were 33% more likely to develop ROP.
Conclusion This study represents the largest cohort of newborns analysed for ROP. The multivariate analysis emphasised the role of birth weight in extended-stay infants, as well as Hispanic race, respiratory distress syndrome, and intraventricular haemorrhage.
Statistics from Altmetric.com
Retinopathy of prematurity (ROP) is the primary cause of visual impairment in premature infants.1 Factors demonstrated to be associated with an increased risk of the development of ROP are short gestational period and low birth weight,2 sepsis, intraventricular haemorrhage,3 blood transfusions,4 and mechanical ventilation.5 Although light was reported to increase the incidence of ROP,6 the reduction in ambient light exposure did not result in an increased incidence of ROP.7
Our goal was to evaluate the incidence of ROP based upon a national database. In addition, we sought to identify baseline characteristics, demographic information, comorbidities, and surgical interventions on patients with ROP. Finally, odds ratio estimates from multivariate logistic regression models were sought.
The National Inpatient Sample (NIS) is maintained by the Agency for Healthcare Research and Quality as part of the Healthcare Cost and Utilization Project (HCUP).8 The NIS is a 20% representative sample of all US hospital discharges, stratified by geographical region, hospital size, urban versus rural location, and teaching versus non-teaching status. The teaching status of the hospital was determined by hospital affiliation with either a medical school or an Accreditation Council for Graduate Medical Education residency programme. Hospital size was divided into small, medium, and large with a range of greater than 100 to 450 beds, depending on urban/rural and teaching status. The NIS is the largest, all-payer US administrative database that incorporates discharge data from approximately 1000 hospitals and five to seven million discharges annually. The NIS 20% sample is based on a stratified probability sample of all US hospitals, to provide a national estimate of inpatient health services. Only inpatient data found in a discharge abstract are available in the NIS. The NIS does not have unique patient identifiers, and as a result, patients cannot be followed longitudinally. Further information regarding the NIS is available from the Agency for Healthcare Research and Quality, which administers the database as part of the Healthcare Cost and Utilization Project.
In this retrospective cohort study, cases were identified by the International Classification of Diseases, 9th Revision (ICD-9) procedure codes for the following entry variables: V30 newborn ICD-9 code to capture all live births, ICD-9 362.21 ROP code, and LOS greater than 14 days.
The period studied was from 1 January 1997 to 31 December 2002; this period was chosen so that the results would be sufficiently recent to avoid secular trend bias. Most newborn deaths in premature infants occur within the first week, and the exclusion of children aged less than 14 days limits the effect of mortality upon the dataset. ICD-9 codes for surgical interventions, comorbidity, demographic information and birth weights were determined from the database, and were then used to construct adjusted odds ratios from a multivariate logistic regression model.
Newborn outcomes with and without ROP were examined, including length of stay (LOS), inpatient mortality, total hospital charges, and associated complications (blood transfusion, birth trauma, intrauterine hypoxia, respiratory distress, perinatal infection, fetal haemorrhage, haemolytic disease, necrotising enterocolitis, and intraventricular haemorrhage). Specific ICD-9 codes are available on request. In patients with ROP, the proportion of laser photocoagulation and scleral buckle was calculated stratified by birth weight.
Dichotomous and continuous variables were examined by Student’s t-test and χ2 analyses. Linear and logistic regression analyses were applied to LOS and mortality variables to correct for potential confounders. In addition, logistic regression analyses indicated potential predictors of ROP. A p value of less than 0.05 was set as significant. The statistical program SAS (version 8.1; SAS Institute, Cary, North Carolina, USA) was used for database analysis.
Patient and database characteristics
Using the NIS, 4 675 577 live births were recorded during the six-year period (1997–2002), which represents approximately 20% of the US hospital discharges during that time period. A total of 22 939 000 newborns were thus represented, of which 22 906 005 newborns were without ROP and 28 011 were with ROP (table 1).
Newborns with ROP were more likely to be born at a teaching hospital (78.1%) and to have an increased LOS (67.7 days for infants with ROP versus 36.4 days for infants without ROP), and increased hospitalisation charges (US$192 000 for infants with ROP versus US$94 096 for infants without ROP). Mortality was lower in the ROP group (0.45%) than the non-ROP group (2.14%). All p values were less than 0.0001 (table 1).
Incidence of ROP in overall dataset
Table 1 summarises the demographic characteristics of infants in the study population stratified by the presence or absence of ROP. The total incidence of ROP was noted to be 28 011, representing 0.12% or approximately one out of 820 newborns. The percentage incidence of ROP per year is depicted in fig 1. There was no statistical significance between the incidence of ROP from year to year during the time period analysed (one-way analysis of variance, p>0.05). For premature infants with LOS greater than 14 days, this incidence was 7.35%.
The mean LOS in newborns with ROP and newborns without ROP categorised by birth weight is presented in table 2. The total number of newborns with ROP and LOS greater than 14 days (27 198) is stratified by birth weight and presented in fig 2. The frequency of ROP was greatest (24 198 out of 231 527 representing 10.45%) in infants with LOS greater than 14 days weighing less than 1500 grams compared with 0.22% (3000 out of 1 384 641) in infants weighing more than 1500 grams. A peak incidence of 17.7% (8531 out of 48 116) was noted between 750 and 999 grams. A value of 12.4% incidence (20 784 out of 167 439) was determined in newborns weighing less than 1250 grams.
Comorbidities in newborns with ROP
The comorbidities present in infants with ROP and LOS greater than 14 days are shown in fig 3. The most common complications in both ROP and non-ROP newborns were respiratory distress syndrome (72.8% of infants with ROP versus 41.8% of infants without ROP) and perinatal infection (43.2% of infants with ROP versus 34.4% of infants without ROP). All of the comorbidities shown in fig 3 had a significantly higher incidence in ROP newborns compared with non-ROP newborns (fig 3; p<0.0001).
Surgical intervention for ROP
In the cohort of premature infants with ROP with LOS greater than 14 days analysed in this study, 7.7% (2112 out of 27 435 reported infants) underwent laser photocoagulation; 0.25% (35 out of 14 174) infants had scleral buckle or pars plana vitrectomy coded on the discharge diagnosis and entered into the database system.
Table 3 shows the proportion of these ROP newborns who underwent these surgical procedures by birth weight. Laser photocoagulation rates were highest in newborns weighing less than 1000 grams, with a peak frequency in newborns weighing less than 500 grams (119 out of 417 infants or 28.7%). No incidence is reported for newborns weighing 2000–2500 grams. Rates of scleral buckle or pars plana vitrectomy in premature newborns with ROP with LOS greater than 14 days were 0.4% for newborns weighing 500–749 grams and 0.2% for newborns with birth weights of 750–999 grams. No study infant with ROP and LOS greater than 14 days who had a birth weight of less than 500 grams or greater than 1000 grams was reported to undergo scleral buckle or vitrectomy.
Multivariate logistic regression analysis
The data were entered into a multivariate logistic regression analysis to determine adjusted odds ratios for the development of ROP in newborns with LOS greater than 14 days. Low birth weight was strongly associated with the development of ROP (fig 4). The odds ratios were greatest in infants weighing less than 1250 grams, with lower limits of 95% confidence levels in excess of 45 in the weight classifications between 500 and 1250 grams. For example, newborns with LOS greater than 14 days with birth weights of 750–999 grams were 63.82 times more likely to have ROP than newborns weighing more than 2500 grams at birth (p<0.05). Newborns with birth weights of 1000–1249 grams were 77.26 times more likely to have ROP than those weighing more than 2500 grams (p<0.05).
Only respiratory distress (adjusted odds ratio 1.47, 95% confidence interval (CI) 1.42 to 1.51) and intraventricular haemorrhage (1.1, CI 1.05 to 1.14) were found to be predictive of the development of ROP in the multivariate regression model (fig 5). Hypoxia (adjusted odds ratio 0.78, CI 0.70 to 0.86), perinatal infection (0.85, CI 0.82 to 0.87), and necrotising enterocolitis (0.81, CI 0.75 to 0.85) were not associated with the development of ROP in this model. When race was entered into the model, Hispanic infants were 33% more likely than infants of other races to develop ROP. For all other races, there was no statistically significant relationship between race and the development of ROP. In particular, 95% confidence levels crossed 1 for black infants (0.95 to 1.02) and Asian infants (1.00 to 1.20).
This retrospective study represents the largest cohort of newborns analysed for ROP. Retrospective analyses have previously examined the incidence and epidemiological risk factors for ROP in single centres worldwide9–19 and in multiple US centres.20 21 Retrospective studies using national inpatient databases have been performed in Australia22 and the United States.23
We examined factors such as LOS, birth weight, sex, race, mortality, hospitalisation charges, comorbidities, and treatment for newborns with ROP over the period 1 January 1997 to 31 December 2002. Although providing many broad interpretations based on a national sample, this approach has several limitations. The differentiation of patients in the database requires an accurate definition of ICD-9 codes. A second limitation was caused by variations in the completeness of the data. The NIS database represents a stratified sample across the United States. Samples at one extreme of weight may not be as representative as other subsamples, therefore the prevalence of ROP in the group of newborns weighing less than 500 grams may be underestimated. Third, because laser photocoagulation, vitrectomy and scleral buckle are frequently ambulatory procedures and the HCUPnet NIS database only reports data on inpatient discharges, the total number of these procedures performed cannot be ascertained. It is possible that the data may either underestimate the true incidence of scleral buckle or vitrectomy and panretinal photocoagulation because some procedures may have occurred after hospital discharge, or may overestimate the true incidence if the procedures were performed for diagnoses other than ROP. Another limitation of the study is that highly relevant information, such as the timing of the ROP examinations during hospitalisation and the identity of the physician performing the examinations, is not available from the NIS database. A final limitation is that the HCUPnet NIS database does not contain information on long-term outcomes or efficacy.
In the six-year period, data for over 4.67 million infants were analysed. These data were obtained from a broad socioeconomic and ethnic patient population (table 1). A total of 28 011 newborns with ROP were identified, representing an incidence of 0.12%, 40% lower than the 0.2% incidence reported in a cohort of 1.1 million infants in New York state during 1996–2000.23 The apparent decrease in the incidence of ROP may be partly the result of recent advances in neonatology.15 17 24 Alternatively, it is possible that the difference between our results and the New York study is caused by the method of ICD-9 coding for ROP (362.21) and may have resulted from a failure to enter the codes and random transcription mistakes.
Both the CRYOROP and ETROP trials reported a higher incidence of ROP.25–27 The different methodologies employed in these prospective, multicentre trials make it very difficult for comparisons to be made with the results obtained from databases. For example, in the CRYOROP and ETROP trials, each patient was examined for the presence of ROP, its zone, extent, stage, and presence of Plus disease by ophthalmologists experienced in the longitudinal and sequential changes of ROP.25–27 These studies were designed in a fashion to capture high-risk infants who might benefit from treatment, and it stands to reason that this emphasis on high-risk patients might lead to an increase in the reported incidence. In addition, patients were longitudinally followed. In both the present study and the New York cohort, data were analysed from databases inputted from unknown sources. All patients, not just high-risk patients, were evaluated, probably diluting the reported incidence of disease. Furthermore, we lacked the ability to follow one patient longitudinally, and therefore do not know if a particular patient eventually developed ROP, possibly in an outpatient setting. For these reasons, it is easier to compare this trial with similar analyses obtained from other databases than it is with clinical trials.
In premature infants with LOS greater than 14 days, the incidence of ROP was 61.25 times greater (7.35%) than in newborns with LOS of less than 14 days, suggesting that comorbidities and health status may play an important role in the development of ROP. Conversely, newborns with ROP had higher LOS (67.7 days for infants with ROP versus 36.4 days for infants without ROP) and hospitalisation charges (US$192 000 for infants with ROP versus US$94 096 for infants without ROP). These results are in agreement with numerous studies that have demonstrated that ROP is associated with complex medical problems and prolonged oxygen requirements.10 23 28–30 All comorbidities studied showed a higher incidence in newborns with ROP than in newborns without ROP (fig 3). These results paralleled the findings of a New York cohort,23 in which necrotising enterocolitis and intraventricular haemorrhage were significantly higher in the ROP group than in the non-ROP group. A high illness severity score was shown to be an important predictor of the severity of ROP warranting surgery.19 In our study, the most common comorbidities found in newborns with ROP were respiratory distress syndrome and perinatal infection (fig 3). In a multivariate regression model, respiratory distress and intraventricular haemorrhage were found to be predictive of the development of ROP (fig 5). In the model, both necrotising enterising colitis and perinatal infection appeared to be protective for the development of ROP (fig 4). All research to date indicates that comorbidity is associated with a greater risk of the development of ROP as well as a more advanced disease state. It is possible that improved treatment of these conditions has led to this paradoxical finding. More likely, this change may reverse itself when a longer timeframe is examined.
Newborns with ROP and LOS greater than 14 days were more likely to be born at a teaching hospital (78.1%) than newborns without ROP and LOS greater than 14 days. Although mortality was lower in infants with ROP (0.45%) than in the population without ROP (2.14%), this may be skewed by the 14-day LOS minimum for the ROP group.
In our study as well as in numerous previous studies,9–17 20 23 31birth weight was determined to be an important factor associated with ROP. The CRYO-ROP study reported an ROP incidence of 65.8% in newborns with birth weights of less than 1251 grams.20 Numerous studies have found ROP incidence rates ranging from 10% to 46% among low birthweight newborns worldwide.9–19 Few studies have, however, systematically analysed the frequency of ROP in infants with larger birth weights. In addition, because many of the previous studies took place in sites with a special interest in ROP, the data may not be representative of the general population. In the New York cohort, the incidence of ROP was 20.3%, 27.3% and 33.2% for infants with LOS over 28 days with birth weights of less than 1500 grams, 1200 grams, and 1000 grams, respectively.23 Similarly, in our study of a larger national cohort, the incidence of ROP was greatest in newborns weighing less than 1500 grams and a peak incidence of ROP was found for newborns weighing between 750 and 999 grams (fig 2). A multivariate logistic regression analysis showed that low birth weight was strongly associated with the development of ROP (fig 4). In one report, the mean birth weights of newborns with severe ROP from poorly developed countries was 903–1527 grams, compared with 737–763 grams in highly developed countries,31 suggesting that the population of newborns who develop ROP differs between highly developed and less developed countries.31.
Previous studies have demonstrated that race is another important independent risk factor for the development of ROP.19 29 Our multivariate regression analysis model revealed that Hispanic infants were 33% more likely than infants of other races to develop ROP. Charles et al32 observed that Hispanic infants in an inner-city population had a significantly higher incidence of severe ROP than black infants. The authors reported that severe ROP developed significantly more frequently in Hispanic than in black infants, but they also noted significantly lower birth weights in the Hispanic infants for which their univariate analysis could not control.32 In a recent retrospective study, Eliason and collaborators33 found no significant difference between Hispanic and white non-Hispanic infants in the risk of developing ROP. In their analysis, the slightly greater incidence of ROP in the Hispanic group was explained by the lower birth weight and gestational age in this population. When they controlled for birth weight and gestational age, there was no significant effect of ethnicity on the incidence of ROP.33
For all other races, there was no statistically significant relationship between race and ROP development. This echoes the findings of the CRYO-ROP study, in which the incidence of ROP was highest among non-black, non-white infants,20 but is discordant with the New York cohort in which there was no significant relationship between Hispanic ethnicity and the development of ROP.23 We demonstrated that the majority of premature newborns with ROP were male; no gender difference in the incidence of threshold ROP was found in the CRYO-ROP study20 or the New York cohort.23
In the population of premature newborns with ROP with LOS greater than 14 days analysed in this study, 7.7% underwent laser photocoagulation and 0.25% underwent scleral buckle or pars plana vitrectomy. These rates are slightly lower than those in the data from the New York cohort, in which 9.5% of ROP infants underwent laser photocoagulation and 0.5% underwent scleral buckle or vitrectomy,23 possibly as a result of the differences in the population enrolled and the outpatient status at the time of the treatment. In our study, laser photocoagulation rates were highest in newborns weighing less than 1000 grams and no incidence was reported for newborns weighing 2000–2500 grams, in line with previous reports.23
Despite recent advances in diagnosis and treatment, ROP continues to represent a leading cause of childhood blindness worldwide, with enormous costs to society. This study analysed the largest reported cohort of newborns with ROP, involving over 4.5 million infants during a six-year span, with a reported incidence of ROP of 0.12%. The multivariate analysis emphasised the role of birth weight in extended stay infants with ROP, as well as Hispanic race, respiratory distress syndrome, and intraventricular haemorrhage. Future directions would have the goal of removing the weighting factor and refining the internal HCUPnet NIS database to allow for inconsistency in checking.
This study demonstrates that large retrospective cohort studies employing national inpatient databases provide an extremely important tool for clinical ophthalmic research and clinical care.
Competing interests: The authors have no proprietary interest in any of the materials discussed in this article.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.