M
Pasiabods(B(0wspm2
TawtR(sns
FOHSCPmSRRC4CS1d
J
Race, Gender, and Clinical Risk Index forBabies (CRIB) Score as Predictors of Severe
Retinopathy of Prematurityichael B. Yang, MD,a Edward F. Donovan, MD,b and Jordan R. Wagge, BSa
urpose: To determine whether race, gender, and the Clinical Risk Index for Babies (CRIB) illness severity scorere predictors of threshold or severe prethreshold retinopathy of prematurity warranting surgery (ROP warrantingurgery) and whether racial and gender differences in ROP are correlated with racial and gender differences inllness severity. Methods: This was a retrospective analysis of premature infants 401-1250 g at birth that weredmitted to the University Hospital of Cincinnati (January 1998 to May 2003). Birth weight, gestational age, multipleirth, birth in the study hospital or elsewhere, race, gender, CRIB score, and eye findings were abstracted. Theutcome variable was ROP warranting surgery. Results: Of 299 patients (596 eyes) with adequate eye and CRIBata, 35 patients (11.7%) [66 eyes; 11.1%] developed ROP warranting surgery. Multiple logistic regression analysishowed that higher CRIB score (P � 0.0001; odds ratio [OR] 1.21), male gender (P � 0.005; OR: 2.68), nonblack raceP � 0.0005; OR: 4.32), lower gestational age, and multiple birth are predictive factors for ROP warranting surgery.ecause birth weight and gestational age comprise 2 of the 6 components of the CRIB score, a CRIB subscore
CRIBSUB) consisting of the remaining 4 components was tested and remained a significant predictor (P �.00001). Birth weight was a significant predictor when CRIBSUB was in the model but not when the CRIB scoreas used. The CRIB score was a predictor of neonatal mortality, but race and gender did not predict the CRIB
core or neonatal mortality. Conclusions: Nonblack race, male gender, and higher CRIB illness severity scores areredictors of ROP warranting surgery. In our population, there were no racial or gender differences in neonatalortality or CRIB scores to explain the racial and gender differences in severity of ROP. (J AAPOS 2006;10:
53-261)
Rs
irVAittvddC
ccspisaC
he Multicenter Trial of Cryotherapy for Retinop-athy of Prematurity (CRYO-ROP) demonstrated aracial difference in the incidence of threshold ROP
mong premature infants �1251 g at birth.1,2 White raceas an independent risk factor for the development of
hreshold ROP,3 and the overall incidence of thresholdOP among white infants was twice that of black infants
7.4% vs. 3.2%).1,2 Because factors associated with illnesseverity, such as episodes of hypoxemia, septicemia, andumber of blood transfusions, have been associated withevere ROP,4,5 the observed racial difference in threshold
rom the aDepartment of Ophthalmology/Abrahamson Pediatric Eye Institute, Cincinnati,hio, bDepartment of Neonatology/Child Policy Research Center, Cincinnati Children’sospital Medical Center, University of Cincinnati College of Medicine, Cincinnati, Ohio.upported in part by a grant provided by the Knights Templar Eye Foundation, Inc.,hicago, Illinois (to M.B.Y.).resented at the 31st Annual Meeting of the American Association of Pediatric Ophthal-ology and Strabismus, Orlando, Florida, March 9 to 13, 2005.ubmitted March 9, 2005.evision accepted December 12, 2005.eprint requests: Michael B. Yang, MD, Abrahamson Pediatric Eye Institute, Cincinnatihildren’s Hospital Medical Center, 3333 Burnet Ave MLC 4008, Cincinnati, OH5229-3039 (e-mail: [emailprotected]).opyright © 2006 by the American Association for Pediatric Ophthalmology andtrabismus.091-8531/2006/$35.00 � 0
ioi:10.1016/j.jaapos.2006.01.004
ournal of AAPOS
OP may be the result of a racial disparity in illnesseverity.
Previous studies have demonstrated a racial differencen illness severity as shown by a lower neonatal mortalityate (NMR) among very-low birth-weight (�1500 g,LBW) black infants as compared with white infants.6-8
mong premature black infants, there also is a lowerncidence of neonatal morbidity, such as respiratory dis-ress syndrome and the need for neonatal assisted ventila-ion.9,10 Similarly, the observed female advantage in sur-ival among premature infants also may suggest a genderifference in illness severity,8,11,12 although no genderifference in the incidence of threshold ROP was found inRYO-ROP.1
To minimize the potential confounding effect of a ra-ial disparity in health on ROP outcome, Tadesse andolleagues examined premature infants who had Candidaepsis and similar duration of oxygen exposure.13 In com-arison with white infants, black infants still had a lowerncidence of stage 3 or threshold ROP. These resultsuggest that the racial difference in ROP is not caused byracial disparity in health. However, the 2 characteristics,andida sepsis and duration of oxygen exposure, may not
ndicate a similar severity of illness because the underlying
June 2006 253
mbaom
mhibACspkMtsR
idwWdRds
S
TBa(d4itstfeirHtwiwwbwpcut
Rtisw
C
Tdtcbusacbwasgnvhn
T
B
G
C
M
M
M
MA
Journal of AAPOSVolume 10 Number 3 June 2006254 Yang, Donovan, and Wagge
orbidity of premature infants may not be reflected solelyy similarities in the development of these 2 disease char-cteristics. Consequently, a global and integrated methodf measuring illness severity may address the problemore adequately.14,15
The Clinical Risk Index for Babies (CRIB) score esti-ates illness severity using data collected in the first 12
ours after birth.16 The CRIB score is highly predictive ofn-hospital mortality for premature infants and has alsoeen useful in predicting certain neonatal morbidities.16-18
few studies have examined the predictive value of theRIB score on the incidence of ROP.18-20 However, these
tudies typically have been limited by problems such as aoor definition of ROP outcome, a failure to control fornown predictors of severe ROP, or small sample size.oreover, we are aware of no study that has considered
he potential interaction between race, gender, and illnesseverity scores in a multivariate model of risk for severeOP.The purpose of this study is to determine whether
llness severity, as estimated by the CRIB score, is a pre-ictive factor for threshold or severe prethreshold ROParranting surgery (hereafter ROP warranting surgery).e further hypothesize that previously established racial
ifferences in ROP and possible gender differences inOP may be explainable in part by racial and genderifferences in illness severity as measured by the CRIBcore.
UBJECTS AND METHODS
his study was approved by the Institutional Reviewoards of the University Hospital of Cincinnati (UHC)nd Cincinnati Children’s Hospital Medical CenterCCHMC) and conformed to their policies and proce-ures. This is a retrospective study of premature infants01-1250 g at birth that were admitted to the neonatalntensive care unit (NICU) of UHC from January 1998hrough May 2003. Although UHC is a level 3 NICU,everely ill patients requiring additional care were some-imes transferred to CCHMC, and infants dischargedrom UHC or CCHMC typically received follow-up eyexaminations as outpatients at CCHMC. The infants weredentified using the Neonatal Research Network VLBWegistry maintained by the National Institute for Childealth and Human Development (NICHD). Gender and
he basic CRYO-ROP prognostic variables3 of birtheight, gestational age at birth, multiple birth status, birth
n the study hospital or elsewhere (inborn status), and raceere abstracted from the registry. The race of the infantas assigned as the mother’s race and was categorized aslack or nonblack because the number of nonblack, non-hite infants in the overall population was very small (9atients). The abstracted information was verified duringhart review that was performed to record the 6 variablessed to calculate the CRIB score. Because the ROP data in
he registry was incomplete, results of eye examinations for e
OP also were abstracted by chart review. Exclusion cri-eria were ambiguous genitalia, incomplete data prevent-ng the calculation of a CRIB score, and an inadequateequence of eye examinations to determine whether ROParranting surgery developed.
RIB Score
he CRIB score is calculated from 6 variables measureduring the first 12 hours after birth: birth weight, gesta-ional age, congenital malformation, maximum base ex-ess, and maximum and minimum appropriate FiO2 (Ta-le 1).16 Even although the CRIB score was developed forse in the first 12 hours after birth, it has been useduccessfully in studies of infants in the first 12 hours afterdmission to NICU.21 In our study, inborn infants typi-ally were admitted to the NICU within 15 minutes afterirth. The few outborn infants usually were admittedithin 24 hours after birth. Maximum and minimum FiO2
re considered “appropriate” if the O2 saturation mea-ured simultaneously by pulse-oximetry or arterial bloodases is between 88% and 95%. As long as an infant wasot receiving respiratory support (defined as mechanicalentilation, positive airway pressure ventilation, helmetood, blow-by, or nasal cannula techniques; William Tar-ow-Mordi, personal communication), the maximum base
able 1. CRIB scoreFactor Score
irth weight (g)�1350 0851–1350 1701–850 4�700 7
estational age at birth (wk)�24 0�24 1
ongenital malformationsNone 0Not acutely life-threatening 1Acutely life-threatening 3aximum base excess in first 12 h (mmol/L)��7.0 0�7.0 to �9.9 1�10.0 to �14.9 2��15.0 3inimum appropriate FiO2 in first 12 h�0.40 00.41–0.60 20.61–0.90 30.91–1.00 4aximum appropriate FiO2 in first 12 h�0.40 00.41–0.80 10.81–0.90 30.91–1.00 5aximum CRIB score 23
dapted from Lancet 1993;342:193-8.16
xcess and the maximum and minimum FiO2 were con-
satcswaowvdCo
E
TaatimmnppouomzurutotT22iiegfiR
cpicqtOcch
csmncrztcie(eo“rbrnmqndne
S
SItRaeturtCeawlcptaMth
R
A1wa
Journal of AAPOSVolume 10 Number 3 June 2006 Yang, Donovan, and Wagge 255
idered to be normal even if no blood sample was testednd even if the simultaneous O2 saturation was greaterhan the appropriate range.16 The CRIB score is notalculated in patients with inevitably lethal malformations,uch as anencephaly, trisomy 18, or renal agenesis; other-ise, congenital malformations are graded as none, not
cutely life-threatening, or acutely life-threatening basedn diagnosis and severity of the condition. Because birtheight and gestational age comprise 2 of the 6 CRIB scoreariables and have previously been shown to be indepen-ent predictors of threshold ROP, we also calculated aRIB subscore (CRIBSUB) that was composed of thether 4 variables and included it in our analysis.
ye Examination for ROP
he results of eye examinations for ROP were recordedccording to the International Classification for Retinop-thy of Prematurity.22 At UHC and CCHMC, all prema-ure infants weighing 401 to 1250 g at birth received annitial screening eye examination by a pediatric ophthal-
ologist at 5 to 6 weeks after birth or 32 weeks of post-enstrual age, whichever is later. Subsequent eye exami-
ations were performed at intervals of 1 to 6 weeks de-ending on the location and severity of ROP. Typically,atients were examined at 2-week intervals. If they devel-ped early prethreshold ROP, they were examined weeklyntil the ROP either regressed or progressed to thresholdr severe prethreshold ROP that warranted surgical treat-ent. When immature vessels or mild ROP located in
one 3 was found, a follow-up examination at 4 to 6 weekssually was scheduled. Infants who developed ROP war-anting surgery were treated with laser photocoagulationnless they were too ill to undergo surgery. Severe pre-hreshold ROP was defined as zone 1 or zone 2 prethresh-ld ROP which, in the judgment of the screening oph-halmologist, indicated a very high risk of poor outcomes.his meant (A) zone 1 or posterior zone 2, stage 1 or stageROP with “plus” disease or (B) zone 1 or posterior zone
, stage 3 ROP with “plus” disease in which the extent ofnvolvement as measured by clock hours of neovascular-zation did not quite satisfy the criteria for threshold dis-ase. Decisions by ophthalmologists in this study to sur-ically treat severe prethreshold ROP patients predate thendings and recommendations of the Early Treatment foretinopathy of Prematurity study.23
Because we anticipated that not all infants would haveomplete medical records (ie, those documenting a com-lete sequence of eye examinations showing full vascular-zation of the retina to the ora serrata), we developed someriteria for determining whether a particular infant’s se-uence of eye examinations was sufficient to conclude thathey did or did not develop ROP warranting surgery.bviously, eyes that developed ROP warranting surgery,
onfirmed by documentation of surgery or extenuatingirc*mstances such as the infant being too ill for surgery,
ad an adequate sequence of eye examinations. To con- w
lude that other eyes did not develop ROP warrantingurgery, they were required to have records that docu-ented an initial eye examination with follow-up exami-
ations at typical intervals and that also met 1 of 2 sets ofriteria: (1) At the final ROP examination available, theetina was completely mature, ROP of any stage was inone 3, postmenstrual age �44 weeks had been reached, orhe examining ophthalmologist indicated a finding ofhronic peripheral avascularity or requested a follow-upnterval of �4 weeks associated with a sequence of eyexaminations that was consistent with this last diagnosis.2) The eye had zone 2 ROP with 2 or more consecutivexaminations showing regression of ROP from prethresh-ld to “any” ROP, from stage 2 to stage 1 ROP, or fromany” to “no” ROP with the last eye examination occur-ing at postmenstrual age �40 weeks. Only a few eyeselonged to this second category because of missingecords at the end of the infants’ sequence of eye exami-ations. On the basis of the available records and afore-entioned criteria, 2 infants each had only 1 eye that
ualified as receiving an adequate sequence of eye exami-ations, although the excluded fellow eye probably nevereveloped ROP warranting surgery. Hence, the totalumber of eyes for the 299 infants in this study was 596yes instead of 598 eyes.
tatistical Analysis
tatistical analysis was performed using SPSS 12.0 (SPSS,nc., Chicago, IL). For risk of ROP warranting surgery,he effects of the basic prognostic variables used in CRYO-OP were determined using multiple logistic regression
nalysis. Chi-square analysis was performed during blockntry to determine whether gender, the CRIB score, andhe CRIBSUB score improved the basic model of risksing the CRYO-ROP variables. Finally, multiple logisticegression analysis was performed to examine the predic-ive value of the basic CRYO-ROP variables, gender, theRIB score, and the CRIBSUB score in various risk mod-
ls of ROP warranting surgery. Receiver operating char-cteristics (ROCs) were calculated to determine the modelith the greatest predictive accuracy.24 Tests for multicol-
inearity also were performed to identify possibly impre-ise parameter estimates. Multiple logistic regression waserformed to verify that the CRIB score predicted neona-al or in-hospital mortality and to determine whether racend gender were predictors for survival in our population.
ultiple linear regression analysis was performed to de-ermine whether race and gender were predictors ofigher CRIB scores.
ESULTS
n initial 503 infants admitted with birth weights of 401-250 g were identified using the NICHD Neonatal Net-ork registry after 3 infants were excluded because of
mbiguous genitalia (Figure 1). In addition, 146 infants
ere excluded because an adequate sequence of eye exam-
ipbteRChRetsppksg(t
o1obwnpcfistoo
Hbwtlagaw
gfbabggIag5hsc
brihvmatmic
ttittc3o2alSSsu0�r
Ftc
Journal of AAPOSVolume 10 Number 3 June 2006256 Yang, Donovan, and Wagge
nations either was not performed or could not be com-leted because of early death or could not be verifiedecause of incomplete records or failure to follow-up. Ofhe remaining 357 infants, 58 infants (including 1 infant [2yes] who received laser photocoagulation for thresholdOP) were excluded because of inadequate data for theRIB score. Of the remaining 299 infants (596 eyes) whoad both an adequate sequence of eye examinations forOP and a CRIB score, 264 infants (526 eyes; 2 infantsach with only 1 qualifying eye) did not receive surgicalreatment whereas 35 infants (66 of 70 eyes) received laserurgery for ROP warranting surgery (5 eyes for severerethreshold ROP and 61 eyes for threshold ROP). Threeatients (5 eyes) treated with laser photocoagulation werenown to develop unfavorable retinal outcome or requirecleral buckling for progression of disease. Included in theroup of infants with ROP warranting surgery was 1 infant2 eyes) who developed threshold ROP bilaterally but wasoo ill to undergo laser treatment and subsequently died.
Table 2 summarizes the clinical characteristics of theverall population of 503 infants with birth weight 401-250 g and the 299 infants who met the final study criteriaf adequate eye examinations and CRIB scores. Althoughlack infants were adequately represented, our populationas not otherwise racially diverse. Of the 9 nonblack,onwhite infants in the overall population, 5 were His-anic white, 1 was Asian/Pacific Islander, and 3 werelassified as “other.” Only 7 of these infants were in thenal study group, and none developed ROP warrantingurgery. The percentage of infants who were born insidehe study hospital was approximately 96% for both theverall population and the study infants. The percentages
503 infants (401 – 1250 g birth weight)
Excluded for: Inadequate eye exams for ROP - 146
Died before eye exams began -118 Died before eye exams completed - 5 Died ≤ 28 days - 118 Died > 28 days - 5
Incomplete records/lost to follow-up - 23
357 infants (712 eyes) Adequate eye exams for ROP
Excluded for incomplete CRIB - 58 Base excess - 30 Maximum appropriate FiO2 - 23 Base excess and Max FiO2 - 1
Multiple - 6 Surgery for ROP – 1 infant (2 eyes)299 infants (596 eyes)
CRIB score available
264 infants (526 eyes) No surgery
35 infants (66 of 70 eyes)
Surgery for ROP
IG 1. Inclusion/exclusion criteria. The number of eyes is not exactlywice that of infants because two infants had only one eye that metriteria for an adequate sequence of eye examinations.
f singletons and boys in both populations were similar. p
owever, there were a slightly higher percentage of non-lack infants in the study population (64%) as comparedith the overall population (58%). Compared with pa-
ients who did not receive surgery, infants who underwentaser surgery were more likely to be from a multiple birthnd had lower birth weight, younger gestational age,reater CRIB, and greater CRIBSUB scores. Nonblacknd male infants were also more likely to develop ROParranting surgery (Tables 2 and 3).To determine the risk factors for ROP warranting sur-
ery, we performed multiple logistic regression using dif-erent variable combinations in several models of risk (Ta-le 4). In all models, inborn status initially was eliminateds a significant predictive factor (P � 0.998). The otherasic CRYO-ROP variables are in Model 1: birth weight,estational age, multiple birth, and race. Model 2 addsender to Model 1. Model 3 adds CRIB score to Model 2.n Model 4, the variables birth weight and gestational agere removed to avoid duplication with the birth weight andestational age components of the CRIB score. In Model, birth weight and gestational age at birth are included;owever, the CRIB score is replaced by the CRIBSUBcore, which lacks the birth weight and gestational ageomponents of the CRIB score.
We performed the multiple logistic regression analysisy treating the outcome for each eye of the infant sepa-ately, although the predictive variables characterized eachnfant. Our rationale was that although both eyes tend toave similar outcomes, a small percentage of infants de-eloped ROP warranting surgery in only one eye, whichay indicate intermediate risk for certain predictive vari-
bles. Therefore, performing the analysis using eyes ratherhan infants should increase the predictive accuracy of theodels. However, even when the analysis was based on
nfant outcome, the predictive variables remained statisti-ally significant (data not shown).
In all of the models containing birth weight and gesta-ional age at birth, there was an 11% to 32% reduction inhe likelihood of ROP warranting surgery for each 100-gncrease in birth weight and a 37% to 43% reduction inhe likelihood of ROP warranting surgery for each addi-ional week of gestational age. Multiple birth status in-reased the odds of ROP warranting surgery by 3.1- to.8-fold. Nonblack race and male gender increased thedds of ROP warranting surgery by 2.9- to 4.3-fold and by.3- to 3.3-fold, respectively. In Models 3 and 4, eachdditional point on the CRIB score scale increased theikelihood of ROP warranting surgery by 21% to 37%.imilarly, in Model 5, each 1-point increase in the CRIB-UB score increased the likelihood of ROP warrantingurgery by approximately 26%. Block-entry analysis eval-ated by the �2 showed that the addition of gender (P �.009), CRIB score (P � 0.0001), and CRIBSUB score (P
0.00001) to the basic CRYO-ROP model (Model 1)esulted in statistically significant improvements to model
redictive capacity. ROC curves, which examine the rela-
tgeaa(sw
iatqrrnbitaoceomAmatt9itF0
Oms0s(UssTr
mSensgomsVcacMecpC
D
ThwrCCosdw
T
NBGBSMNCC
Tc
MFT
Journal of AAPOSVolume 10 Number 3 June 2006 Yang, Donovan, and Wagge 257
ionship between the specificity and the sensitivity of anyiven model, were plotted for all 5 models of risk. How-ver, only Models 1, 3, and 5 are shown (Figure 2). Therea under the curve was calculated and showed excellentccuracy of prediction greater than 0.85 for all 5 modelsTable 5). However, Model 5, which uses the CRIBSUBcore instead of CRIB score, was the most accuratehereas Models 1 and 4 were the least accurate.Because the CRIB score initially was developed for use
n the first 12 hours of life rather than the first 12 hours ofdmission to the NICU, we then restricted our study tohe 574 eyes of 288 infants (2 infants each with only 1ualifying eye) who were inborn. After reanalysis, theesults essentially were unchanged (data not shown). Theesults also were unchanged after including the one girl, aonblack infant (2 eyes) who underwent surgery for ROPut was missing a CRIB score (Figure 1). Two hypothet-
cal CRIB scores were calculated for this infant by usinghe infant’s birth weight and gestational age subscores anddding the lowest or highest possible subscores for thether 4 components of the CRIB score. We also wereoncerned that the exclusion of 58 infants with adequateye examinations but who lacked CRIB scores could alterur results. Most of these infants were missing data foraximum base excess and/or maximum appropriate FiO2.lthough we could not make any assumptions concerningissing base excess values, we could make some reasonable
ssumptions concerning maximum appropriate FiO2 inhe typical situation. If the O2 saturation associated withhe maximum FiO2 was higher than the upper limit of5% that defined the appropriate range, then the theoret-cal FiO2 necessary to lower the O2 saturation to the 88%o 95% range should have been lower than the recordediO2. For example, let us assume the maximum FiO2 is
able 2. Clinical characteristics of all infants, study infants, and study in
Characteristics All patients
o. of infants 503irth weight (g) (mean � SD) 873 � 222estational age (wk) (mean � SD) 26.7 � 2.5orn in the study hospital (%) 96.4ingleton births (%) 75.5ale (%) 49.7onblack race (%) 58.1RIB score (mean � SD) –RIBSUB score (mean � SD) –
able 3. Raw incidence of ROP warranting surgery in 596 study eyesategorized by race and gender (eyes treated/eyes in group)
Black Nonblack Total by gender
ale 7.3% (8/110) 19.8% (36/182) 15.1% (44/292)emale 7.5% (8/107) 7.1% (14/197) 7.2% (22/304)otal by race 7.4% (16/217) 13.2% (50/379)
.95 and the corresponding O2 saturation is 100%. If the e
2 saturation had instead been 88% to 95%, then theaximum appropriate FiO2 necessary to achieve that O2
aturation level should theoretically have been less than.95. The FiO2 would probably have been in either theame category (0.91-1.00) or the next lower category0.81–0.90) but probably not 2 categories lower (Table 1).
sing these assumptions, we generated 2 alternate CRIBcores for the 22 (of 58) infants who were missing a CRIBcore as the result of an inappropriate maximum FiO2.he use of alternate CRIB scores did not change the
esults significantly (data not shown).Race and gender were highly significant in each of the
odels, and the inclusion of the CRIB score or the CRIB-UB score resulted in a higher, not lower, likelihood ofyes developing ROP warranting surgery associated withonblack race and male gender. Both of these findingsuggest that very little interaction exists between race,ender, and the illness severity scores in predicting ROPutcome. To confirm these initial impressions, tests forulticollinearity were performed to assess the presence of
ubtle interaction between the variables in each model.ariance proportions and tolerance inflation factors were
alculated and revealed no significant interaction betweenny of the predictive factors, with the exception of aorrelation between the CRIB score and birth weight in
odel 3 (data not shown). However, this interaction wasxpected because the CRIB score included a birth weightomponent, and a loss of statistical significance in theredictive value of birth weight was observed when theRIB score was added in Model 3 (Table 4).
ISCUSSION
his study shows that nonblack race, male gender, andigher CRIB illness severity scores are predictors of ROParranting surgery. When Schaffer and colleagues3 first
eported a racial difference in threshold ROP in theRYO-ROP study, they urged caution because theRYO-ROP study was not specifically designed to dem-nstrate such a difference. Saunders and colleagues2 con-idered a number of explanations for the apparent racialisparity in threshold ROP; however, they stated theyere unaware of any racial disparity in health that could
ategorized by need for laser surgery to treat ROP
tudy patients
Study patients
No surgery Surgery
299 264 35915 � 193 936 � 187 752 � 15327.1 � 2.1 27.3 � 2.0 25.1 � 1.6
96.3 95.8 10072.6 74.6 57.148.8 46.6 65.763.5 61.7 77.1
4.6 � 4.0 4.1 � 3.6 8.5 � 4.11.8 � 2.5 1.5 � 2.4 3.5 � 3.2
fants c
S
xplain the observed result. Nevertheless, they speculated
tsstoCricb
tTip
stuNsosttsctCwsr
T
B
G
M
R
G
C
C
Ca
Fomttusswn
Ts
12345Csg
Journal of AAPOSVolume 10 Number 3 June 2006258 Yang, Donovan, and Wagge
hat if the sickest black premature infants had failed tourvive until the time of eye examination for ROP, aelection bias may have favored the healthier black prema-ure infants who survived and developed less-severe formsf ROP. However, the natural history portion of theRYO-ROP study reported a lower neonatal mortality
ate (NMR) among black infants as compared with whitenfants, even when the results were adjusted for potentialonfounding factors.8 Therefore, if a selection bias had
able 4. Comparison of different models of risk of ROP warranting surge
Model 1 CRYO-ROP variables
Model 2ROP �
irth weight (each 100-g increase) 0.71 [0.56–0.90]P � 0.005
0.68 [0.5P � 0.00
estational age (wks) 0.57 [0.45–0.71]P � 0.00005
0.60 [0.4P � 0.00
ultiple birth (yes � 1, no � 0) 3.55 [1.85–6.82]P � 0.0005
3.28 [1.7P � 0.00
ace (nonblack � 1, black � 0) 2.90 [1.43–5.85]P � 0.005
2.92 [1.4P � 0.00
ender (male � 1, female � 0) – 2.26 [1.2P � 0.01
RIB score – –
RIBSUB score – –
RYO-ROP variables: birth weight, gestational age, multiple birth, and race; inborn stge.
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
1 - Specificity
Se
nsi
tiv
ity
Model 1: CRYO-ROP
Model 3: CRYO-ROP +gender + CRIB
Model 5: CRYO-ROP +gender + CRIBSUB
Random Coin Toss
IG 2. Receiver operator characteristic (ROC) curves for predictionf ROP warranting surgery by different multiple logistic regressionodels. Diagonal indicates a predictive model based on random
oss of a coin; area under the curve for this model is 0.5. A modelhat predicts perfectly has an area under the curve of 1.0. Model 1ses the CRYO-ROP variables. Model 3 adds gender and the CRIBcore to Model 1. Model 5 uses CRIBSUB score instead of the CRIBcore and was the most accurate. CRYO-ROP variables: birtheight, gestational age, multiple birth, and race; inborn status was
ot included because it was highly insignificant.
een present, it should have resulted in fewer cases of R
hreshold ROP among white rather than black infants.he lower NMR among black premature infants may
nstead suggest that they generally are healthier than whiteremature infants and therefore develop less severe ROP.
Unlike Tadesse and colleagues,13 who selected Candidaepsis and similar duration of oxygen exposure as indica-ive of similar illness severity between racial groups, wesed an illness severity score. The CRIB and the Score foreonatal Acute Physiology (SNAP) both measure illness
everity in early neonatal life.16,25 They correlate well withther markers of severity of illness and have been useduccessfully to predict neonatal mortality, in-hospital mor-ality, and certain neonatal morbidities.16-18,26-29 One fac-or that led us to use the CRIB score instead of the SNAPcore was the need for arterial blood gas samples to cal-ulate the latter. Because arterial blood gases are not rou-inely collected for premature infants in every NICU, theRIB score seemed more easily obtainable and moreidely applicable. Because gender differences in illness
everity and neonatal mortality rate previously have beeneported, we also included gender in our models of risk for
6 study eyes
-r
Model 3 CRYO-ROP � gender
� CRIB
Model 4 model3 without BW,
GA
Model 5 CRYO-ROP � gender
� CRIBSUB
0.89 [0.69–1.16]P � 0.401
– 0.68 [0.54–0.87]P � 0.005
0.63 [0.50–0.79]P � 0.0001
– 0.61 [0.49–0.77]P � 0.00005
3.46 [1.77–6.76]P � 0.0005
3.10 [1.66–5.80]P � 0.0005
3.83 [1.94–7.53]P � 0.0005
4.32 [1.93–9.67]P � 0.0005
3.48 [1.65–7.34]P � 0.005
4.23 [1.91–9.38]P � 0.0005
2.68 [1.39–5.16]P � 0.005
3.29 [1.75–6.17]P � 0.0005
2.77 [1.43–5.36]P � 0.005
1.21 [1.10–1.34]P � 0.0001
1.37 [1.27–1.47]P � 1 � 10�16
–
– – 1.26 [1.14–1.39]P � 0.00001
s not included because it was highly insignificant; BW, birth weight; GA, gestational
able 5. Performance of different models of risk for ROP warrantingurgery
ModelArea underROC curve SE
. CRYO-ROP variables (CRYO-ROP) 0.859 0.021
. CRYO-ROP � gender 0.873 0.019
. CRYO-ROP � gender � CRIB 0.884 0.019
. CRYO-ROP � (BW,GA) � gender � CRIB 0.850 0.025
. CRYO-ROP � gender � CRIBSUB 0.892 0.017RYO-ROP variables: birth weight, gestational age, multiple birth, and race; inborntatus was not included because it was highly insignificant; BW, birth weight; GA,estational age.
ry in 59
CRYOgende
4–0.86]5
8–0.75]001
0–6.34]05
4–5.92]5
1–4.21]
atus wa
OP warranting surgery.8,11,30
bRnccit
tti4embCetccmwwppTg
a4wgdtfw2smutraTt
sidbgaeBd
gtepaswcria
pmwtsbabpsswNihitvfctitol
woC5RRewRete
oprw
Journal of AAPOSVolume 10 Number 3 June 2006 Yang, Donovan, and Wagge 259
Although the basic CRYO-ROP prognostic variables ofirth weight, gestational age, and multiple birth predictedOP warranting surgery in our models, inborn status didot. However, our sample was small, and a higher per-entage of our infants was born inside the study hospital asompared with CRYO-ROP (96% vs, 82%).1 This dispar-ty may be the result of different regional prenatal prac-ices or referral patterns.
Because the CRIB score includes birth weight and ges-ational age as 2 of its 6 components, it was not surprisinghat the CRIB score and the birth weight variable were notndependent of each other and interacted (Model 3, Table). However, gestational age remained highly predictiveven with the CRIB score present in the same model. Thisay not be entirely unexpected because the 4 ranges of
irth weight and their corresponding subscores within theRIB score (Table 1) may provide nearly as graded an
stimate of increased risk of ROP warranting surgery ashe continuous birth weight variable. By contrast, the 2ategories of gestational age within the CRIB score asompared with the continuous gestational age variableay not allow adequate differentiation of risk for ROParranting surgery. In addition, the CRIBSUB score,hich excludes the birth weight and gestational age com-onents of the CRIB score, was a highly significant inde-endent, predictive factor for ROP warranting surgery.his suggests that the CRIB score is not simply a surro-ate for birth weight and gestational age.
The increased likelihood of ROP warranting surgeryttributed to nonblack race in our study (odds ratio 2.9 to.3) is slightly higher than in CRYO-ROP (odds ratio 2.9),hereas the increased likelihood of ROP warranting sur-ery among males was not expected because CRYO-ROPetected no gender difference in threshold ROP.1,3 Al-hough our results could have been biased by missing datarom 23 infants who were lost to follow-up (of whom 58%ere female and 54% were black), we believe most of these3 infants were at low risk for ROP warranting surgery, aseveral infants had normal examinations at �6 to 12onths of age and any infant that developed poor vision or
nfavorable outcome would likely have been referred backo our clinic. The majority of these patients also had lowisk characteristics with birth weights �1050 g, gestationalge �29 weeks, and low CRIB scores (mean � 3.0).herefore, their inclusion would probably not have altered
he results.Alternatively, because our study was not a masked pro-
pective study, the examiners were aware of the race of thenfant and may have been biased by the known racialisparity in ROP. However, this hypothesis seems unlikelyecause an initial underreporting of ROP warranting sur-ery should result in a number of eyes progressing to poornatomic outcome even if one assumes that 40% of theyes that reach threshold status will regress spontaneously.y contrast, a gender difference in threshold ROP was not
etected in CRYO-ROP, and recent reports that suggest a 8
ender disparity in severe ROP were published subsequento the time period covered by our study.31,32 Therefore,xaminer bias regarding gender should not have beenresent in our study. Although gender bias in the medicalnd ophthalmic care of patients has been reported, iteems unlikely in the diagnosis and treatment of ROParranting surgery given the potentially devastating out-
ome from untreated disease. Overreporting of ROP war-anting surgery among white and among male prematurenfants is possible but seems unlikely considering the risksnd visual consequences associated with laser therapy.
The CRIB score and CRIBSUB score were verified asredictive factors for neonatal mortality and in-hospitalortality in our population (data not shown). However,e found no gender or racial difference in survival or in
he CRIB score (data not shown), although race has beenhown to predict SNAP scores29 and the CRIB score haseen modified for gender disparities.33 The lack of racialnd gender disparities in mortality rates in our study maye attributed to (1) the smaller sample size in the survivalortion of our study as compared with CRYO-ROP, (2) akewed referral pattern that resulted in different illnesseverities among racial and gender groups as comparedith other published studies, or (3) a temporal change inMR among racial and gender groups. This last possibil-
ty is intriguing. It has been reported that survival ratesave been improving for both black and white premature
nfants over the last several decades. Inexplicably, however,he gap between the races has been narrowing, with sur-ival improving more rapidly among white premature in-ants.34,35 Possible explanations for this finding may in-lude differences in the medical care received by prema-ure infants, a racial difference in response to the samemprovements in care, or the existence of an inherent limito improvements in survival resulting in a diminishing ratef improvement among blacks, who already approach thatimit.
The incidence of ROP warranting surgery in our studyas greater than that found in CRYO-ROP. In the groupf 299 study infants with adequate eye examinations andRIB scores, the incidence was 11.1% (66 treated eyes/96 total eyes) as compared with 5.5% found in CRYO-OP. (Note: the incidence of threshold ROP in CRYO-OP was 6% for infants and 5.5% for all natural history
yes.)1,3 Even when the number of eyes in the denominatoras expanded to include the 712 eyes that had adequateOP examination (but not necessarily CRIB scores) andyes treated for severe prethreshold ROP were excluded,he percentage of eyes treated (8.6%; 61 treated eyes/712yes) was still higher than that reported in CRYO-ROP.
This higher incidence of ROP warranting surgery atur institution may simply reflect risk differences in ouratient population. For example, our population was lessacially diverse, with a lower percentage of nonblack, non-hite infants as compared with CRYO-ROP (2.3% vs.
.7%).1 However, this difference should have resulted in
fia
mIwsCltmoRCmboabTpnhhgpcg
iinaegwaiwuofiAogr
dgmpatct
faosfgoR
sH
1
1
1
1
1
1
1
1
Journal of AAPOSVolume 10 Number 3 June 2006260 Yang, Donovan, and Wagge
ewer cases with ROP warranting surgery because thencidence of threshold ROP in CRYO-ROP was highestmong nonblack, nonwhite infants.1
Alternatively, the incidence of ROP warranting surgeryay be higher because our patients are more severely ill.
nitially, this conclusion seemed unlikely. The mean birtheight and gestational age of infants in our study is
lightly, but not significantly, lower as compared withRYO-ROP (Table 2).1 The NMR in our study also is
ower than CRYO-ROP (23.5% vs 28.7%).1 However,his last finding is difficult to interpret. If lower NMR is aarker of lower illness severity, as we initially postulated,
ne would expect the percentage of eyes progressing toOP warranting surgery in our study to be smaller than inRYO-ROP, which we did not observe. However, we areore than a decade removed from CRYO-ROP, and it has
een suggested that advancements in postnatal treatmentsf premature infants that improve neonatal survival, suchs surfactant use, may not result in a concomitant decrease,ut rather an increase, in the incidence of severe ROP.36
his may reflect the complex interaction between im-roved care and illness severity. Although advances ineonatal care have stabilized and improved the neonatalealth of premature infants who in previous years wouldave died, these recent survivors may nevertheless havereater underlying illness severity than the average ofremature infants �1251 g birth weight and thereforeontribute to a higher incidence of ROP warranting sur-ery.
Our finding of a higher risk of ROP warranting surgeryn eyes of male infants may be similarly explained, as theres evidence that the greater male vulnerability to certaineonatal morbidities may be exacerbated as survivalmong male infants improves.12 Therefore, one mightxpect this improvement in male survival to translate intoreater illness severity among surviving male as comparedith female infants. Likewise, an improvement in survival
mong white infants may be expected to result in greaterllness severity among surviving white infants. However,e found no gender or racial differences in illness severitysing the CRIB score. Perhaps, the use of the SNAP scorer the measurement of chronic illness severity beyond therst 12 hours after birth would yield different results.lternatively, advances in care may have eliminated genderr racial differences in illness severity altogether whereasender and racial differences in ocular morbidity, ie ROP,emain.
Indeed, the finding that race and gender are indepen-ent of illness severity in predicting ROP warranting sur-ery suggests that other physiologic or genetic differencesay be involved. The potential protective effect of ocular
igmentation against light-induced retinal vascular dam-ge was previously considered a possible explanation forhe racial difference in ROP2,15; however, more recentlinical trials failed to demonstrate a lower incidence of
hreshold ROP among infants whose eyes were shielded
rom light.37,38 Thus it remains unclear how race protectsgainst ROP. Perhaps gender differences in systemic39,40
r local levels of angiogenic factors, or a differential re-ponse to these factors, may yet be discovered that accountor the gender disparity in ROP. Finally, the inclusion ofender as well as the CRIB illness severity score in modelsf risk for ROP41,42 may improve our ability to predictOP outcome.
We thank Judy Bean, PhD, for her helpful comments regardingtatistical analysis and Jillian Ellis, Frances Bowman, and Robertutchins, MD, for their assistance in supplemental data collection.
References1. Palmer EA, Flynn JT, Hardy RJ, Phelps DL, Phillips CL, Schaffer
DB, et al. The Cryotherapy for Retinopathy of Prematurity Coop-erative Group. Incidence and early course of retinopathy of prema-turity. Ophthalmology 1991;98:1628-40.
2. Saunders RA, Donahue ML, Christmann LM, Pakalnis AV, Tung B,Hardy RJ, et al. Racial variation in retinopathy of prematurity. ArchOphthalmol 1997;115:604-8.
3. Schaffer DB, Palmer EA, Plotsky DF, Metz HS, Flynn JT, Tung B,et al. Prognostic factors in the natural course of retinopathy ofprematurity. Ophthalmology 1993;100:230-7.
4. Shohat M, Reisner SH, Krikler R, Nissenkorn I, Yassur Y, Ben-SiraI. Retinopathy of prematurity: incidence and risk factors. Pediatrics1983;72:159-63.
5. Gunn TR, Easdown J, Outerbridge EW, Aranda JV. Risk factors inretrolental fibroplasia. Pediatrics 1980;65:1096-100.
6. Alexander GR, Tompkins ME, Altekruse JM, Hornung CA. Racialdifferences in the relation of birth weight and gestational age toneonatal mortality. Public Health Rep 1985;100:539-47.
7. Alo CJ, Howe HL, Nelson MR. Birth-weight–specific infant mor-tality risks and leading causes of death: Illinois, 1980-1989. Am J DisChild 1993;147:1085-9.
8. Phelps DL, Brown DR, Tung B, Cassady G, McClead RE, PurohitDM, et al. 28-day survival rates of 6676 neonates with birth weightsof 1250 grams or less. Pediatrics 1991;87:7-17.
9. Fujikura T, Froehlich LA. The influence of race and other factors onpulmonary hyaline membranes. Am J Obstet Gynecol1966;95:572-8.
0. Kavvadia V, Greenough A, Dimitriou G, Hooper R. Influence ofethnic origin on respiratory distress syndrome in very prematureinfants. Arch Dis Child Fetal Neonatal Ed 1998;78:F25-8.
1. Effer SB, Moutquin JM, Farine D, Saigal S, Nimrod C, Kelly E, etal. Neonatal survival rates in 860 singleton live births at 24 and 25weeks gestational age: a Canadian multicentre study. Br J ObstetGynecol 2002;109:740-5.
2. Wells JCK. Natural selection and sex differences in morbidity andmortality in early life. J Theor Biol 2000;202:65-76.
3. Tadesse M, Dhanireddy R, Mittal M, Higgins RD. Race, Candidasepsis, and retinopathy of prematurity. Biol Neonate 2002;81:86-90.
4. Verma A, Okun NB, Maguire TO, Mitchell BF. Morbidity assess-ment index for newborns: a composite tool for measuring newbornhealth. Am J Obstet Gynecol 1999;181:701-8.
5. Yang MB. Ethnic variation in the incidence and severity of retinop-athy of prematurity: possible explanations including racial disparityin illness severity. Int Ophthalmol Clin 2003;43:91-103.
6. International Neonatal Network: CRIB Investigator’s Group. TheCRIB (clinical risk index for babies) score: a tool for assessing initialneonatal risk and comparing performance of neonatal intensive careunits. Lancet 1993;342:193-8.
7. De Courcy-Wheeler RHB, Wolfe CDA, Fitzgerald MS, Goodman
JDS, Gamsu HR. Use of the CRIB (clinical risk index for babies)
1
1
2
2
2
2
2
2
2
2
2
2
3
3
3
3
3
3
3
3
3
3
4
4
4
Journal of AAPOSVolume 10 Number 3 June 2006 Yang, Donovan, and Wagge 261
score in prediction of neonatal mortality and morbidity. Arch DisChild 1995;73:F32-6.
8. Eriksson M, Bodin L, Finnstrom O, Schollin J. Can severity-of-illness indices for neonatal intensive care predict outcome at 4 yearsof age? Acta Paediatr 2002;91:1093-100.
9. Coscia A, Prandi G, Borgione S, Leone A, Nicocia M, Mombro M,et al. CRIB score: mortality, morbidity, and long-term neurologicdevelopment. Acta Biomed Ateneo Parmense 2000;71(Suppl 1):637-40.
0. Allegaert K, Casteels I, Cossey V, Devlieger H. Retinopathy ofprematurity: any difference in risk factors between a high and lowrisk population? Eur J Ophthalmol 2003;13:784-8.
1. Lund GC, Green D, Browne R, Ackerman NB. New CRIB score:one score for all NICU admissions. Pediatr Res 1997;41:162A.
2. The Committee for the Classification of Retinopathy of Prematurity.An international classification of retinopathy of prematurity. ArchOphthalmol 1984;102:1130-4.
3. Early Treatment for Retinopathy of Prematurity CooperativeGroup. Revised indications for the treatment of retinopathy of pre-maturity: results of the early treatment for retinopathy of prematu-rity randomized trial. Arch Ophthalmol 2003;121:1684-96.
4. Hanley JA, McNeil BJ. A method of comparing the areas underreceiver operating characteristic curves derived from the same cases.Radiology 1983;148:839-43.
5. Richardson DK, Gray JE, McCormick MC, Workman K, GoldmannDA. Score for neonatal acute physiology: a physiologic severity indexfor neonatal intensive care. Pediatrics 1993;91:617-23.
6. Richardson DK, Phibbs CS, Gray JE, McCormick MC, Workman-Daniels K, Goldmann DA. Birth weight and illness severity: inde-pendent predictors of neonatal mortality. Pediatrics 1993;91:969-75.
7. Richardson DK, Shah BL, Frantz ID, Bednarek F, Rubin L, Mc-Cormick MC. Perinatal risk and severity of illness in newborns at 6neonatal intensive care units. Am J Public Health 1999;89:511-6.
8. Wilson A, Gardner MN, Armstrong MA, Folck BF, Escobar GJ.Neonatal assisted ventilation: predictors, frequency, and duration ina mature managed care organization. Pediatrics 2000;105:822-30.
9. Berman S, Richardson DK, Cohen AP, Pursley DM, Lieberman E.Relationship of race and severity of neonatal illness. Am J ObstetGynecol 2001;184:668-72.
0. Stevenson DK, Wright LL, Lemons JA, Oh W, Korones SB, PapileLA, et al. Very low birth weight outcomes of the National Instituteof Child Health and Human Development Neonatal Research Net-work, January 1993 through December 1994. Am J Obstet Gynecol
1998;179:1632-9.
1. Darlow BA, Hutchinson JL, Henderson-Smart DJ, Donoghue DA,Simpson JM, Evans NJ; Australian and New Zeland Neonatal Net-work. Prenatal risk factors for severe retinopathy of prematurityamong very preterm infants of the Australian and New ZealandNeonatal Network. Pediatrics 2005;115:990-6.
2. Early Treatment for Retinopathy of Prematurity CooperativeGroup. The incidence and course of retinopathy of prematurity:findings from the early treatment for retinopathy of prematuritystudy. Pediatrics 2005;116:15-23.
3. Parry G, Tucker J, Tarnow-Mordi W , for the UK Neonatal StaffingStudy Collaborative Group. CRIB II: an update of the clinical riskindex for babies score. Lancet 2003;361:1789-91.
4. Allen MC, Alexander GR, Tompkins ME, Hulsey TC. Racial dif-ferences in temporal changes in newborn viability and survival bygestational age. Paediatr Perinatal Epidemiol 2000;14:152-8.
5. Demissie K, Rhoads GG, Ananth CV, Alexander GR, Kramer MS,Kogan MD, et al. Trends in preterm birth and neonatal mortalityamong blacks and whites in the United States from 1989 to 1997.Am J Epidemiol 2001;154:307-15.
6. Repka MX, Hardy RJ, Phelps DL, Summers CG. Surfactant pro-phylaxis and retinopathy of prematurity. Arch Ophthalmol 1993;111:618-20.
7. Reynolds JD, Hardy RJ, Kennedy KA, Spencer R, van Heuven WA,Fielder AR. Lack of efficacy of light reduction in preventing reti-nopathy of prematurity. Light Reduction in Retinopathy of Prema-turity (LIGHT-ROP) Cooperative Group. N Engl J Med 1998;338:1572-6.
8. Phelps DL, Watts JL. Early light reduction for preventing retinop-athy of prematurity in very low birth weight infants. CochraneDatabase Syst Rev 2001;(1):CD000122.
9. Malamitsi-Puchner A, Tziotis J, Tsonou A, Protonotariou E,Sarandakou A, Creatsas G. Changes in serum levels of vascularendothelial growh factor in males and females throughout life. J SocGynecol Invest 2000;7:309-12.
0. Wheeler T, Evans PW, Anthony FW, Godfrey KM, Howe DT,Osmond C. Relationship between maternal serum vascular endothe-lial growth factor concentration in early pregnancy and fetal andplacental growth. Hum Reprod 1999;14:1619-23.
1. Hardy RJ, Palmer EA, Schaffer DB, Phelps DL, Davis BR, CooperCJ. Outcome-based management of retinopathy of prematurity.JAAPOS 1997;1:46-54.
2. Hardy RJ, Palmer EA, Dobson V, Summers CG, Phelps DL, QuinnGE, et al. Risk analysis of prethreshold retinopathy of prematurity.
Arch Ophthalmol 2003;121:1697-701.
