A New Risk-Based Screening Criterion for Treatment-
Demanding Retinopathy of Prematurity in Denmark
WHAT’S KNOWN ON THIS SUBJECT: Efforts are made worldwide
to determine the most effective and safe criterion for retinopathy
of prematurity screening. Combining gestational age at delivery
and birth weight limits is the preferred method to effectiveness
and safety measured by the capability to detect advanced
retinopathy of prematurity cases.
WHAT THIS STUDY ADDS: This study shows how advanced
statistical models are helpful tools for evaluating of the suitability
of screening criteria for implementation. It also presents a
criterion based on the estimated risk of developing treatment-
demanding retinopathy of prematurity. This criterion is more
effective and safer than any conventional screening criteria.
abstract
OBJECTIVE: The aim of this study was to uncover the most effective and
safe criterion to implement for retinopathy of prematurity screening in
Denmark.
METHODS: This retrospective national cohort study is based on data
from 3 national registers. These registers provided on infants treated
for retinopathy of prematurity, infants in need of treatment but missed
by the present screening program, and the candidate neonates for
advanced retinopathy of prematurity development A nonlinear logistic
regression model was fitted to the data, and various screening criteria
were evaluated.
RESULTS: During the study period (2002–2006), 116 infants were
treated for retinopathy of prematurity, no treatment-demanding reti-
nopathy of prematurity infants were missed by the screening program,
and 182 premature infants were candidates for developing treatment-
demanding retinopathy of prematurity. Screening criteria combining
gestational age at delivery and birth weight limits and new risk-based
criteria were compared with regards to their effectiveness. The risk-
based criteria were the most effective. Use of the 0.13% risk-based
criterion to define the population to be screened resulted in the detec-
tion of all treated infants in the study period and 17.4% fewer infants to
screen. The model predicted this criterion to result in 1 missed case of
treatment-demanding retinopathy of prematurity every 11 years and 1
case of blindness every 18 years in Denmark.
CONCLUSIONS: Screening criteria based on risk estimates of develop-
ing treatment-demanding retinopathy of prematurity are the most ef-
fective for retinopathy-of-prematurity screening. The risk-based crite-
rion of 0.13% can safely be implemented for future retinopathy-of-
prematurity screening in Denmark. Pediatrics 2011;127:e598–e606
e598 SLIDSBORG et al
AUTHORS: Carina Slidsborg, MD,a Julie Lyng Forman,
MSc, PhD,b Steen Rasmussen, MSc,c Hanne Jensen, MD,
PhD,a,d Kamilla Rothe Nissen, MD, PhD,a Peter Koch
Jensen, MD, Dr Med Sci,a Regitze Bangsgaard, MD,a
Hans Callø Fledelius, MD, Dr Med Sci,a Gorm Greisen, MD,
Dr Med Sci,e and Morten la Cour, MD, Dr Med Scia
aDepartment of Ophthalmology, Copenhagen University Hospital,
Glostrup Hospital, Copenhagen, Denmark; bDepartmant of
Biostatistics, University of Copenhagen, Copenhagen, Denmark;
cNational Board of Health, Copenhagen, Denmark; dEye Clinic,
Kennedy Center, Glostrup, Denmark; and eDepartment of
Neonatology, Copenhagen University Hospital, Rigshospitalet,
Denmark
KEY WORDS
screening, preterm delivery, retinopathy of prematurity,
threshold disease, retinal ablation therapy, expected missed
treatment-demanding ROP cases, risk-based criterion,
conventional screening criteria
ABBREVIATIONS
ROP—retinopathy of prematurity
T-ROP— threshold retinopathy of prematurity
GA—gestational age at delivery
BW—birth weight
PTROP—prethreshold retinopathy of prematurity
TD-ROP—treatment-demanding retinopathy of prematurity
NRBVIC—National Registry for Blind and Visually Impaired
Children
CI—confidence interval
www.pediatrics.org/cgi/doi/10.1542/peds.2010-1974
doi:10.1542/peds.2010-1974
Accepted for publication Nov 19, 2010
Address correspondence to Carina Slidsborg, MD, Department
of Ophthalmology, Glostrup Hospital, Nordre Ringvej 57, 2600
Glostrup, Denmark. E-mail: carinaslidsborg@hotmail.com.
PEDIATRICS (ISSN Numbers: Print, 0031-4005; Online, 1098-4275).
Copyright © 2011 by the American Academy of Pediatrics
FINANCIAL DISCLOSURE: The authors have indicated that they
have no personal financial relationships relevant to this article
to disclose.

Retinopathy of prematurity (ROP) is
one of the main causes of visual im-
pairment in premature infants.1 The vi-
sual impairment associated with ROP
is lifelong and has a great impact on
the individual’s quality of life.2 In 1988,
the US Cryotherapy for ROP multi-
center study showed that poor struc-
tural and functional retinal outcome
caused by threshold ROP (T-ROP) can
be reduced if cryotherapy is per-
formed.3 Subsequently, screening cri-
teria were implemented worldwide. To
date, the most commonly used screen-
ing criteria are those based on gesta-
tional age at delivery (GA) and the
birth-weight (BW) limits. They use sin-
gle values of GA or BW or both in con-
junction (AND criteria, OR criteria).
Such screening criteria are herein la-
beled conventional screening criteria.
Unnecessary screening should be
avoided because the screening exami-
nations are painful and potentially haz-
ardous to infants.4–9 Hence, screening
criteria need to be reviewed periodi-
cally. An effective and safe criterion
keeps the screening load low while de-
tecting all advanced ROP cases. Re-
cently, several studies of highly devel-
oped countries showed that infants
who develop T-ROP are very young and
small, and new screening criteria
were suggested accordingly.10–13 Re-
cent Danish studies showed a similar
trend, and one study additional dem-
onstrated that older T-ROP infants
also are born small for gestational
age.14–17 All considered, in Denmark,
larger infants seems to be screened
unnecessarily.
This study aims to specify an effective
and safe screening criterion to imple-
ment in Denmark. In Denmark, the in-
tention was to maintain treatment on
classical threshold disease (T-ROP).3,18
Nevertheless, some of the treated in-
fants in the present study were treated
at prethreshold ROP (PTROP), in accor-
dance with the recommendations of
PEDIATRICS Volume 127, Number 3, March 2011
the early treatment for ROP random-
ized trial.18 Therefore, the infants who
were treated will be referred to as in-
fants with treatment-demanding ROP
(TD-ROP) instead of using the T-ROP
designation. We define effectiveness
as the screening load associated with
the number of expected missed TD-ROP
cases. The number of expected missed
TD-ROP cases is an independent indica-
tor of a criterion’s predicted safety. We
developed a nonlinear logistic regres-
sion model that estimated the risks of
developing TD-ROP for each candidate
neonate. These risk estimates were
used to calculate the number of ex-
pected missed TD-ROP cases resulting
from a given criterion. The new type of
criteria that emerged from isolines of
estimated risk of TD-ROP (risk-based
criterion) is, when based on a sensible
statistical model and a sufficiently
large sample, the most effective of all
criteria.
MATERIALS AND METHODS
The Register Data
We present a retrospective, register-
based cohort study for premature in-
fants born in Denmark in the period
2002–2006. The data were obtained
from 3 different national registers.
Data from these registers were unam-
biguously linked to specific infants
through their individual civil registra-
tion number.
In the National Birth Register of the Na-
tional Board of Health, we obtained
birth information on all candidate neo-
nates for advanced ROP development.
Thus, inclusion criteria (GA ⬍32 weeks
or BW ⬍1750 g) of the search resulted
in 4478 infants alive at the fifth postna-
tal week. We excluded infants who
were registered with extreme values,
which were most likely typing errors
and not real values of preterm infants.
Consequently, we excluded the follow-
ing infants: 46 infants with unknown GA
or BW; 2 infants with a GA of less than
ARTICLES
20 weeks; 5 infants with a GA of more
than 45 weeks; 65 infants with a BW of
more than 3 SDs; and 178 infants with a
BW of less than 6 SDs of that expected
for a given GA.19 None of the excluded
infants had received treatment for
ROP. The resulting population con-
sisted of 4182 infants. The primary
screening criterion of a GA of less than
32 weeks was actually used for the en-
tire study period.20 A survey clarified
that the secondary criterion of a BW of
less than 1750 g was rarely applied.
Thus ⬃1305 of the study sample
(31.2%) was probably not screened for
ROP.
The registry of the National Univer-
sity Hospital, Rigshospitalet, pro-
vided information on all infants
treated for ROP in Denmark. Three
treated infants, born by mothers
with a permanent address outside
Denmark at childbirth, were ex-
cluded from the study population be-
cause such infants were not pre-
sented in the National Birth Registry.
The remaining sample consisted of
116 treated infants.
The National Register of Blind and
Visually Impaired Children (NRBVIC)
was searched on August 6, 2009, to
identify advanced ROP cases missed
by the present screening program.
The law mandates reporting to this
registry if a child is suspected of be-
ing visually impaired (ie, best cor-
rected visual acuity ⱕ6/18 [0.33] in
the better eye). Such children are ex-
amined by a specialist pediatric oph-
thalmologist, who records into the
registry the most likely underlying
cause. In the NRBVIC, 11 children
were registered with a primary diag-
nosis of sequels to ROP, and all had
retinal treatment during the neona-
tal period. Thus, no screening fail-
ures were found. The 116 infants iden-
tified as treated infants could be linked
to the National Birth Registry data set by
the civil registration number.
e599
Statistical Methods
The risk of developing TD-ROP depend-
ing on GA and BW was analyzed by
means of a nonstandard polynomial lo-
gistic regression model. Because a few
of the treated eyes did not reach T-ROP
before treatment, the term TD-ROP is
used here instead of T-ROP. The out-
come of interest was TD-ROP in at least
1 eye. The majority of infants with a GA
of more than or equal to 32 weeks
were not screened; therefore, the only
obtainable information for these in-
fants were whether they were later
registered in the NRBVIC as blind (ie,
visual acuity ⱕ20/200 in the better
eye) from ROP sequels. Thus, a non-
standard logistic regression model for
the outcome of TD-ROP/blindness was FIGURE 1
developed according to Bayes’ theorem The study cohort of ROP candidate neonates are represented by full, black circles. A minority of infants
above 32 weeks was screened (gray shade). The red circles represent the treated infants. The upper
that the probability of blindness is part of the figure show the gestational age and birth weight values of the infants, whereas the lower
proportional to the probability of TD- part of the figure presents the infants when the birth weights are transformed into a z scores
ROP: P(Blindness) ⫽ P(TD-ROP) ⫻ (standard normal deviate), being a measure of how much the infants’ birth weights deviate from the
expected birth weight expressed in SDs.
P(blindness TD-ROP). We assumed the
factor of proportionality, P(blindness TD-
ROP), to be constant and independent
deviates) that measure BW relative to cases are not severely biased as a
of the prematurity level; based on the
what is expected for a particular GA result of being evaluated on the
Cryotherapy for ROP Study21, the prob-
before the analysis.19 The z score is de- same data that were used for esti-
ability is between 0.377 and 0.587.
fined as (BW ⫺ MGA)/SD, where MGA is mating the risks.22,23 Confidence in-
The choice of the higher value,
the expected BW of an infant for a given tervals (CIs) were computed by boot-
P(blindness TD-ROP) ⫽ 0.587, results
GA, and SD is the standard deviation of strapping, from which 1000 full
in a conservative model that possibly
the expected BW.19 Fig 1 presents the samples were drawn from the data
overestimates the risk of blindness in
candidate neonates before and after with replacement.24
infants with a GA of less than 32 weeks
the BWs are transformed into z scores. Because of the nonstandard outcome
in case they are not treated. Given that
In addition, note that as the risk of TD- (TD-ROP/blindness), special programs
the implied screening loads and ex-
ROP was found to decrease faster with for the analysis were developed at the
pected numbers of missed TD-ROP cases
increasing GA than what is plausible in Department of Biostatistics, University
were computed from these estimated a simple logistic regression, polyno-
risks, these also are conservative. Inclu- of Copenhagen, within the framework
mial terms of GA was included in the of R Statistical software version 2.8.0.25
sion of an outlying PTROP infant in the model up to the fifth order. The inclu-
model increases the risk estimates mar- sion of even higher-order terms did RESULTS
ginally. We note that the screening load, not improve the fit of the model. In the
and therefore the effectiveness of differ- lack of prospective data, we used inter- Estimates of Risk
ent screening criteria, is sensitive to the nal validation methods (regular boot- In Fig 2, the estimated risks of develop-
distribution of BW and GA. strapping) as recommended by Stey- ing TD-ROP projected on the explana-
The statistical model is described in erberg et al. [22] to verify that the tory variables surface are shown. Dif-
detail in Appendix 1. Note that as the 2 prognostic model is optimal with re- ferent colors of the estimated risk
explanatory variables, GA and BW, are spect to mean-squared prediction er- surface represent different risk zones.
highly collinear (Fig 1), BW was trans- rors and that the estimated screening The risk estimates decrease with
formed to z scores (standard normal loads and expected missed TD-ROP larger GA and BW. The infants born less

e600 SLIDSBORG et al

Risk 1: 0.89%
2000
Birth weight (g)
1500
LGA by +2SD
1000
24 26
FIGURE 2
The risk card represents the estimated risks of developing TD-ROP for all combinations of gestational age and birth weight. The stippled lines depict the
limits according to which data points are classified as large for gestational age, average for gestational age, and small for gestational age. Three crucial
risk-based isolines of 0.07%, 0.13%, and 0.89% also are presented. The risk-based isolines are defined by an identical risk of developing T-ROP for the all the
points (gestational age and birth weight) on the line. To the left of the risk-based isolines, the risk of developing TD-ROP is higher than on the line and to the
right of the line the risk is lower than on the line. These risk-based isolines can be used to delineate an area representing the population to be screened.
premature, being either average for
gestational age or large for gesta-
tional age, are at the lowest risk of de-
veloping TD-ROP. The infants born
small for gestational age are at high-
est risk. Figure 2 also represents the
treated sample, including the largest
T-ROP infant with an estimated risk of
0.89% and the PTROP outlier with an
estimated risk of 0.13%. All candidate
neonates also are shown, and risk es-
timates extrapolated beyond these
data points are not trustworthy. Three
crucial risk-based isolines of 0.07%,
0.13%, and 0.89% also are presented.
They are means of the estimated risk
for 2 genders because the gender-
specific lines were practically insepa-
rable. The risk-based isolines are de-
fined by an identical risk of developing
TD-ROP for all points (GA and BW) on
the line. To the left of the risk-based
PEDIATRICS Volume 127, Number 3, March 2011
0.13% 0.07%
SGA by −2SD
28 30
Gestational age (wk)
isolines, the risk of developing TD-ROP
is higher than on the line, and to the
right of the risk-based isolines the risk
is lower than on the line. The risk-
based isolines can be used to delineate
an area representing the population to
be screened. Therefore, the isolines
can be used as screening criteria with
which infants with risk estimates
equal to or higher than a given risk
level are all screened for ROP.
Figure 3 presents the effectiveness of
risk-based criteria and conventional
screening criteria. The reported val-
ues are normalized to 1000 candidate
neonates. When a dual criterion re-
sulted in several screening loads for a
given number of missed TD-ROP cases,
the lowest one was reported.
Table 1 presents the screening load,
the expected number of missed TD-ROP
ARTICLES
16%
8%
4%
2%
1%
0.5%
0.25%
0.12%
0.06%
0.03%
32
cases, and the expected number of
blind children resulting from different
risk-based criteria. The reported val-
ues are normalized to 1000 candidate
neonates. All these criteria result in
fewer infants to screen than the
present screening criterion, while de-
tecting all treated infants of the 2002–
2006 study cohort.
Evaluation of Screening Criteria
Table 2 presents various risk-based
and conventional screening criteria.
The reported values are normalized to
1000 candidate neonates. The criteria
expected to result in 0.05 missed TD-
ROP cases would imply 1 missed TD-
ROP case every 22 years or 1 infant be-
coming legally blind every 37 years.
The currently used criterion of a GA of
less than 32 weeks is exceeded by the
risk-based criterion of 0.071% and the
e601
 The criteria expected to result in ⬃0.9

2.0
missed TD-ROP cases are equal to 1
Risk criterion missed TD-ROP case every 1.2 years or
AND criterion
Expected missed TD-ROP cases

OR criterion 1 infant becoming legally blind every 2

1.5

GA criterion years. The risk-based criterion of

BW criterion
(per 1000 children)

0.89% exceeds all other criteria of sim-

ilar risk because it results in 49%
1.0

fewer infants to screen. The remaining

criteria also provided a large reduc-
tion in screening load. However, the
0.5

high number of expected missed TD-

ROP/blind cases resulting from these
criteria prohibits their implementa-
0.0

tion. The risk-based criterion of 0.89%

300 400 500 600 700
would have detected all treated ROP
Screening load cases except the PTROP outlier. Table 3
(per 1000 children) shows gender-specific GA and BW val-
FIGURE 3 ues corresponding to the isolines of
Effectiveness of the 4 conventional screening criteria and of the risk-based criterion. The lines rep- 0.13%.
resent the lowest possible screening load for a given number of TD-ROP cases. In the case in which a
dual criterion resulted in several screening loads for a given number of missed T-ROP cases, the
lowest one was reported. DISCUSSION
TABLE 1 Risk-Based Criteria That Detect All Treated Infants of the Study Sample and Keep the Efforts are made worldwide to deter-
Screening Load Lower Than the Present Screening Criterion mine the most effective and safe ROP
Risk-Based Screening Expected Missed Expected Missed screening criteria.10,12,26–29 The conven-
Criteria Load per 1000 TD-ROP Cases per Blindness Cases tional screening criteria currently are
Candidate 1000 Candidate per 1000 Candidate
Neonates (CI) Neonates (CI) Neonates (CI)
the main choice.20,26,30–34 However, re-
0.0005 666 (460–935) 0.032 (0.007–0.410) 0.019 (0.004–0.241)
cently 2 new diagnostic screening
0.0006 648 (451–908) 0.042 (0.012–0.444) 0.025 (0.007–0.261) guidelines have been suggested as ef-
0.0007 631 (445–886) 0.053 (0.018–0.476) 0.031 (0.011–0.279) fective alternatives.30,31 All these crite-
0.0008 616 (437–850) 0.064 (0.025–0.509) 0.038 (0.015–0.299)
0.0009 604 (430–823) 0.074 (0.034–0.541) 0.043 (0.020–0.318)
ria resulting from retrospective stud-
0.0010 592 (425–807) 0.086 (0.043–0.572) 0.050 (0.025–0.336) ies are defining effectiveness as a
0.0011 580 (421–790) 0.098 (0.053–0.599) 0.058 (0.031–0.352) screening load associated with the
0.0012 574 (416–775) 0.105 (0.063–0.629) 0.062 (0.037–0.369)
number of detected advanced ROP
0.0013 568 (410–746) 0.113 (0.074–0.661) 0.066 (0.043–0.388)
The risk-based criterion of 0.13% has the lowest screening load. Details about the CIs are further described in the METHODS
cases and safety as the percentage of
section. all advanced ROP cases detected by a
given criterion.10,12,26–29,35–37 In this
study, effectiveness is defined as the
AND criterion of a GA less than or 0.13% exceeds all other criteria with a screening load associated with the
equal to 32.14 weeks and a BW less similar risk because it results in 17.4% number of expected missed TD-ROP
than or equal to 1890 g because they fewer infants to screen. The AND crite- cases, with the expected number of
resulted in minor reductions in the rion of a GA less than or equal to 31.86 missed TD-ROP cases being an inde-
screening loads. All the above- weeks and a BW less than or equal to pendent indicator of safety. To our
mentioned criteria would have de- 1750 g resulted in 13.0% fewer infants knowledge, no previous studies have
tected all infants treated for ROP in to screen. The remaining criteria only evaluated screening criteria based on
the study population. provided minor reductions. The above such measures.
The criteria expected to result in ⬃0.1 criteria, except the BW criterion, would We herein present a statistical model
missed TD-ROP case associated to 1 have detected all infants treated for that estimates the risk of developing
missed TD-ROP case every 11 years or 1 ROP in the study population. The BW TD-ROP for each candidate neonate.
infant becoming legally blind every 18 criterion would have missed the PTROP The number of expected missed TD-
years. The risk-based criterion of outlier. ROP cases associated with a criterion

e602 SLIDSBORG et al
 ARTICLES

TABLE 2 Presents the Various Possible Screening Criteria mainly results from exclusion of many
Criterion Risk GA BW Screening Load Expected Missed Blindness Cases older and larger infants from screen-
Estimate Limits Limits per 1000 TD-ROP Cases per 1000 ing according to the new criterion. On
Candidate per 1000 Candidate
Neonates Candidate Neonates Neonates the basis of recent national and inter-
Risk-based criterion 0.00071 630 0.055 0.032 national experience, such infants no
AND criterion 32 ⫹ 1 1890 662 0.053 0.031 longer develop T-ROP.11,13,15,16,28 This
OR criterion 31 ⫹ 5 1350 684 0.053 0.031 study also confirmed these results. Fi-
GA criterion 31 ⫹ 6 688 0.055 0.032
BW criterion 1760 914 0.053 0.031 nally, statistical uncertainties and the
Risk-based criterion 0.00130 568 0.113 0.066 slightly conservative nature of the risk
AND criterion 31 ⫹ 6 1750 599 0.113 0.066 estimates for the older and larger in-
OR criterion 31 ⫹ 4 1110 614 0.111 0.065
GA criterion 31 ⫹ 5 626 0.104 0.061
fants might also contribute to the dif-
BW criterion 1670 797 0.113 0.066 ference. Although individual risks of
Risk-based criterion 0.00890 353 0.988 0.580 developing TD-ROP can be estimated
AND criterion 31 ⫹ 1 1370 368 0.984 0.577
OR criterion 29 ⫹ 6 1070 387 0.979 0.575
with reasonable accuracy from this
GA criterion 30 ⫹ 2 418 0.879 0.516 moderately sized data set, the esti-
BW criterion 1340 428 0.956 0.561 mated CIs for the expected missed TD-
The expected missed TD-ROP cases for the conventional criteria correspond to one of the risk-based criteria presented. The small ROP/blindness cases are, as expected,
decimal deviates result from the conventional criteria evaluation in full days for GA limits and full grams for BW limits.
relatively broad, reflecting that the
risks are accumulated over a larger
TABLE 3 GA and Gender-Specific BW Values the lowest screening load. However, set of infants (those who would not be
for the Risk-Based Criterion of 0.13% this criterion is associated with an un- screened). In addition, the CIs of the
GA, wk/d Male BW, g Female BW, g acceptablly large number of expected screening load for the risk-based cri-
ⱕ30 ⫹ 5/215 All All missed TD-ROP/blindness cases and teria are wider than for the conven-
30 ⫹ 6/216 1804 1770 was therefore rejected for implemen-
31 ⫹ 0/217 1764 1731
tional criteria because or the fact that
31 ⫹ 1/218 1720 1688 tation. Thus, determining a new the risk-based criteria are based on
31 ⫹ 2/219 1672 1641 screening limit based on the detection estimated risk and thus subject to ad-
31 ⫹ 3/220 1620 1590 of all TD-ROP alone can easily have se- ditional statistical uncertainty. Supple-
31 ⫹ 4/221 1563 1534
31 ⫹ 5/222 1501 1473 rious ophthalmic and medico-legal menting additional data to the model
31 ⫹ 6/223 1434 1408 consequences when such cases are will lead to narrower CIs.
32 ⫹ 0/224 1361 1337
missed soon after. It additional empha- To confirm that the risk-based crite-
32 ⫹ 1/225 1283 1260
32 ⫹ 2/226 1199 1177 sizes the importance of a safety zone rion of 0.13% is safe for future ROP
32 ⫹ 3/227 1108 1088 between the most mature (in this screening in Denmark, a retrospective
32 ⫹ 4/228 1011 993
case) T-ROP infant and the new screen- register search was done to identify
32 ⫹ 5/229 907 891
ⱖ32 ⫹ 6/230 None None ing limit(s). The risk-based screening treated and undetected advanced ROP
Infants with a certain GA should be screened if the BW is criterion of 0.13% detects all treated cases born between 1990 and 2009.
equal to or less than the values given in the gender-
cases on our year study sample and The risk-based criterion would nar-
specific column. Infants born at a GA 30 ⫾ 5 or less should
be screened irrespective of BW. None of the infants born at reduces the screening load by 17.4% rowly have missed advanced ROP case
GA 32 ⫾ 6 or more should be screened irrespective of BW.
infants. The oldest and largest T-ROP born in 1993 with a GA of 31 ⫾ 0 weeks
infant identified in the study period and a BW of 1920 g. Neonates develop-
is calculated from these risk esti- had a GA of 30.86 and a BW of 1150 g. ing advanced ROP were at that time
mates. Our study shows that among Thus, because the treatment criterion born older and larger at delivery, and
conventional criteria, those combining in Denmark has remained the classical such infant also went undetected by
GA and BW limits are the most effec- T-ROP, this criterion provides a reason- the current screening program. Thus,
tive, although they are inferior to the able safety zone to the nearest T-ROP because the risk-based criterion of
criteria based on the estimated risk of infant (risk estimate of 0.89%).3 The 0.13% would detect all treated infants
developing TD-ROP (the risk-based cri- number of expected missed TD-ROP/ of the study sample, this criterion is
terion). The risk-based criterion of blindness cases associated with this safer today than the current criterion
0.89% is the most effective criterion criterion is larger than for the cur- was in the early 1990s.
because it detected all T-ROP infants rently used criterion. The predicted The criterion of a GA less than or equal
during the study period and results in difference between the 2 criteria to 32 weeks and a BW less than or

PEDIATRICS Volume 127, Number 3, March 2011 e603

TABLE 4 Screening Guidelines in Some Highly Developed Countries
Country Primary Criterion Secondary Criterion Threshold for Screening Expected Missed Expected Blindness
Laser Load per 1000 TD-ROP Cases per Cases per 1000
Candidate 1000 Candidate Candidate
Neonates Neonates Neonates
United Kingdom34 GA ⬍32 wk or BW ⱕ1500 g None HR-PTROP 801 (789–814) 0.014 (0.000–6.244) 0.008 (0.000–3.665)
Netherlands39 GA ⬍ 32 wk or BW ⬍1500 g Treated for ⬎3 d with T-ROP 796 (783–809) 0.014 (0.000–7.251) 0.008 (0.000–4.256)
FIO2 ⱖ0.4
Norway30 GA ⬍32 wk or BW ⬍1500 g None T-ROP 796 (783–809) 0.014 (0.000–7.251) 0.008 (0.000–4.256)
Denmark38 GA ⬍32 wk BW ⬍1750 g if GA estimate T-ROP 688 (675–702) 0.055 (0.000–15.497) 0.032 (0.000–9.097)
is doubtful or if severe
systemic illness
United States 33,40 GA ⱕ30 wk or BW ⬍1500 g GA ⬎30 wk or BW 1500– HR-PTROP 604 (589–565) 0.253 (0.050–7.361) 0.149 (0.029–4.321)
2000 g with unstable
clinical course
United Kingdom34 GA ⬍31 wk or BW ⱕ1250 g None HR-PTROP 549 (534–565) 0.233 (0.019–10.619) 0.137 (0.011–6.233)
Sweden31 GA ⱕ31 wk None Alternative 530 (515–545) 0.296 (0.016–15.978) 0.174 (0.010–9.379)
threshold*
Finland (personal GA ⱕ30 wk or BW ⬍1250 g Severe bronchopulmonary HR-PTROP 455 (440–470) 0.576 (0.126–11.322) 0.338 (0.074–6.646)
communication) dysplasia, IVH, III–IV,
septicaemia, frequent
blood transfusions,
severe systemic illness
Canada26 BW ⱕ1200 g None T-ROP 314 (299–328) 2.053 (0.915–12.557) 1.205 (0.537–7.371)
The screening load per 1000 infants and the missed TD-ROP/blindness cases per 1000 infants for the Danish conventional screening criterion are presented. The same cost-benefit effects
after implementation of other primary criteria in Denmark also are presented. HR-PTROP— high-risk PTROP.
*Alternative threshold: stage 3 in at least 4 clock hours in zone 2 (with or without plus disease).

equal to 1750 g is, for the same rea-
sons as the risk-based criterion of
0.13%, an equally safe but slightly less
effective implementable alternative.
We recommend continuous monitor-
ing of the effectiveness of the chosen
risk-based screening criterion. Moni-
toring can be done by updating the
model with GA and BW values of new
infants as they arise in the National
Birth Registry. In the case that the risk
profile of advanced ROP changes, de-
pendent on disease presentation in
Denmark, a statistical model should be
fitted to the new data.
Screening criteria of other highly de-
veloped countries would, if imple-
mented in Denmark, result in either
overlooking TD-ROP infants or in exam-
ining too many infants unnecessari-
ly.20,26,30,31,33,34,38 Implementation of ei-
ther of the herein proposed screening
criteria will not alter the relative posi-
tion of Denmark significantly with re-
spect to missed TD-ROP/blindness
cases compared with the other coun-
tries (Table 4). Highly developed coun-
tries with conservative screening
guidelines could save many unneces-
sary examinations if they should imple-
ment either of these criteria. However,
whether such screening criteria that
we have proposed can be directly im-
plemented in these highly developed
countries depends on their indepen-
dent distribution of GA and BW and the
regional risk profile for advanced ROP.
Generally, it is highly recommended to
perform retrospective, or preferably
prospective, studies to validate these
results before implementation. For
countries where treatment is indi-
cated at the threshold level other than
the classical T-ROP, validation of these
results is even more important. In de-
veloping countries with a very differ-
ent demographic GA and BW distribu-
tion and a different risk profile,
implementation of these criteria are
clearly not recommended.39
Many highly developed countries have
a secondary screening criterion imple-
mented as well, as presented in Table
4.20,33,38 According to a national survey,
during the study period, the Danish
secondary criterion was rarely used.
Furthermore, no T-ROP infants were
detected because of it. However, with
implementation of either of the new
screening criteria, this criterion re-
gains importance.
The statistical model was developed by
use of register-based data. Therefore,
the model’s accuracy is highly depen-
dent on data completeness and reli-
ability. First, the treated sample was
collected from the Rigshospitalet data-
base. This list was compared with the
surgery list of the ophthalmology de-
partment, and because these 2 lists
corresponded well, the treated sample
is expected to be complete. Second,
any infants not appropriately referred
for treatment from the local hospitals
were expected to appear in the NRBVIC
if they are visually impaired as a result
of ROP. The registration to the NRBVIC
is considered reliable because regis-
tration of any child suspected of being
visually impaired (best corrected vi-
sual acuity ⱕ6/18 [0.33]) is mandated
by law. In addition, parents of regis-
tered children can obtain economical
support when buying aids for their vi-

e604 SLIDSBORG et al

sually impaired child. Clearly, there
might be a latency time until the visu-
ally impaired child is referred to the
register; however, an earlier survey
showed that ⬃78% of cases are regis-
tered within the first 2 years of life,
and, therefore, the registrations of this
study cohort is considered almost
complete at the search data (data not
shown). Because no extra advanced
ROP cases were identified in the NRB-
VIC, the completeness of the treatment
sample is further confirmed. A few TD-
ROP cases, missed by the screening
program, which developed milder bin-
ocular visual impairment or monocu-
lar visual impairment (both not eligi-
ble for registration in the NRBVIC)
were unavailable for this study. How-
ever, because the screening pro-
gram is well implemented in Den-
mark and mostly performed by the
same small group of experienced pe-
diatric ophthalmologists, many such
cases are unlikely. Third, the Na-
tional Birth Registry is considered al-
most complete and reliable because
there were missing or erroneous
data for only 6.61% of the candidate
neonates. The registrations of GA
and BW values on the candidate neo-
nates appear accurate because such
data from all 3 registers corre-
sponded well.
REFERENCES
1. Rudanko SL, Fellman V, Laatikainen L. Visual
impairment in children born prematurely
from 1972 through 1989. Ophthalmology.
2003;110(8):1639 –1645
2. Quinn GE, Dobson V, Saigal S, et al. Health-
related quality of life at age 10 years in very
low-birth-weight children with and without
threshold retinopathy of prematurity. Arch
Ophthalmol. 2004;122(11):1659 –1666
3. Cryotherapy for Retinopathy of Prematurity
Cooperative Group. Multicenter trial of
cryotherapy for retinopathy of prematurity:
preliminary results. Pediatrics. 1988;81(5):
697–706
4. Belda S, Pallas CR, De la Cruz J, Tejada P.
Screening for retinopathy of prematurity: is
it painful? Biol Neonate. 2004;86(3):195–200
PEDIATRICS Volume 127, Number 3, March 2011
CONCLUSIONS
The proposed risk-based criterion of
0.13% or AND criterion of a GA less than
or equal to 31.86 weeks and a BW less
than or equal to 1750 g are both safe
and more effective alternatives to the
presently used screening criterion. We
suggest ongoing evaluation of the suit-
ability of these criteria. Their reevalu-
ation should preferably be performed
by an independent prospective study.
Implementation of the proposed criteria
may also benefit other highly developed
countries. Future similar statistical mod-
els could be refined by including addi-
tional risk factors for advanced ROP.40,41
APPENDIX 1: NONLINEAR LOGISTIC
REGRESSION MODEL
The estimated risk of TD-ROP is given
by the following formula:
P(TD-ROP) ⫽ expit(⫺6.97 ⫺ 1.97
⫻ [GA-31] ⫺ 0.0556 ⫻ [GA-31]2
⫺ 0.0871 ⫻ [GA-31]3 ⫹ 0.0009
⫻ [GA-31]4 ⫹ 0.00077 ⫻ [GA-31]5
⫺ 0.688 ⫻ z ⫺ 0.941 ⫻ z ⫻ [GA-31]),
where GA is given in weeks.
z score ⫽ (BW ⫺ MGA) ⁄ SDGA,
where MGA is the expected value of BW
of an infant for a given GA, and SDGA is
the SD of the given GA.19
5. Laws DE, Morton C, Weindling M, Clark D.
Systemic effects of screening for retinopa-
thy of prematurity. Br J Ophthalmol. 1996;
80(5):425– 428
6. Lim DL, Batilando M, Rajadurai VS. Transient
paralytic ileus following the use of
cyclopentolate-phenylephrine eye drops
during screening for retinopathy of prema-
turity. J Paediatr Child Health. 2003;39(4):
318 –320
7. Lim Z, Tehrani NN, Levin AV. Retinal haemor-
rhages in a preterm infant following
screening examination for retinopathy of
prematurity. Br J Ophthalmol. 2006;90(6):
799 – 800
8. Ogut MS, Bozkurt N, Ozek E, Birgen H, Ka-
zokoglu H, Ogut M. Effects and side effects of
ARTICLES
Marsals gender-specified equations:
MGA for males ⫽ ⫺0.0000019073145
⫻ GA4 ⫹ 0.001140644 ⫻ GA3
⫺ 0.1336265 ⫻ GA2 ⫹ 1.976971 ⫻ GA
⫹ 241.0053,
and
MGA for females ⫽ ⫺0.000002761948
⫻ GA4 ⫹ 0.001744841 ⫻ GA3
⫺ 0.2893626 ⫻ GA2 ⫹ 18.91197 ⫻ GA
⫺ 413.5122,
19
where GA is given in days.
The function expit is defined by
expit(x) ⫽ exp(x)/(1 ⫹ exp[x]).
ACKNOWLEDGMENTS
This work was supported by grants from
the Danish Eye Health Society, the Bagen-
kop Nielsens Myopi and Eye Foundation,
the VELUX Foundation, the Aase and Ejnar
Danielsens Foundation, the Dagmar Mar-
shalls Foundation, the Direktør Jacob
Madsen and Hustru Olga Madsens Foun-
dation, and the P A Messerschmidt and
Hustrus Foundation.
We thank the pediatric ophthalmologists
in charge of ROP Screening in Denmark
for participating in the survey on which
screening criteria were applied at the local
NICUs during the study period.
mydriatic eyedrops in neonates. Eur J Oph-
thalmol. 1996;6(2):192–196
9. Sun X, Lemyre B, Barrowman N, O’Connor M.
Pain management during eye examinations
for retinopathy of prematurity in preterm
infants: a systematic review. Acta Paediatr.
2010;99(3):329 –334
10. Ho SF, Mathew MR, Wykes W, Lavy T, Mar-
shall T. Retinopathy of prematurity: an opti-
mum screening strategy. J AAPOS. 2005;
9(6):584 –588
11. Mathew MR, Fern AI, Hill R. Retinopathy of
prematurity: are we screening too many ba-
bies? Eye (Lond). 2002;16(5):538 –542
12. Wright K, Anderson ME, Walker E, Lorch V.
Should fewer premature infants be
screened for retinopathy of prematurity in
e605
 the managed care era? Pediatrics. 1998;
102(1 pt 1):31–34
13. Goble RR, Jones HS, Fielder AR. Are we
screening too many babies for retinopathy
of prematurity? Eye (Lond). 1997;11(pt 4):
509 –514
14. Fledelius HC. Retinopathy of prematurity in
Denmark: epidemiological considerations
and screening limits. Eur J Ophthalmol.
1996;6(2):183–186
15. Fledelius HC, Dahl H. Retinopathy of prema-
turity, a decrease in frequency and severity:
trends over 16 years in a Danish county.
Acta Ophthalmol Scand. 2000;78(3):
359 –361
16. Fledelius HC, Kjer B. Surveillance for reti-
nopathy of prematurity in a Danish
country: epidemiological experience over
20 years. Acta Ophthalmol Scand. 2004;
82(1):38 – 41
17. Slidsborg C, Olesen HB, Jensen PK, et al.
Treatment for retinopathy of prematurity in
Denmark in a ten-year period (1996 2005): is
the incidence increasing? Pediatrics. 2008;
121(1):97–105
18. Early Treatment for Retinopathy of Prema-
turity Cooperative Group. Revised indica-
tions for the treatment of retinopathy of
prematurity: results of the early treatment
for retinopathy of prematurity randomized
trial. Arch Ophthalmol. 2003;121(12):
1684 –1694
19. Marsál K, Persson PH, Larsen T, Lilja H, Sel-
bing A, Sultan B. Intrauterine growth curves
based on ultrasonically estimated foetal
weights. Acta Paediatr. 1996;85(7):843– 848
20. Fledelius HC, Rosenberg T. Retinopathy of
prematurity: where to set screening limits?
Recommendations based on two Danish
surveys. Acta Paediatr Scand. 1990;79(10):
906 –910
21. Palmer EA, Hardy RJ, Dobson V, et al. 15-year
outcomes following threshold retinopathy
of prematurity: final results from the multi-
center trial of cryotherapy for retinopathy
of prematurity. Arch Ophthalmol. 2005;
123(3):311–318
22. Steyerberg EW, Harrel FE, Borsboom GJJM,
Eijkemans MJCR, Vergouwe Y, Habbema
JDF. Internal validation of predictive
e606 SLIDSBORG et al
models: efficiency of some procedures for
logistic regression analysis. J Clin Epide-
miol. 2001;54(8):774 –781
23. Redelmeier DA, Bloch DA, Hickam DH. As-
sessing predictive accuracy: how to com-
pare Brier score. J Clin Epidemiol. 1991;
44(11):1141–1146
24. Davison ACHDV. Bootstrap Methods and
Their Application. Cambridge, United
Kingdom: Cambridge University Press; 1997
25. R Development Core Team. R: A Language
and Environment for Statistical Computing
[computer program]. Vienna, Austria: R
Foundation for Statistical Computing; 2008.
Available at: www.R-project.org.
26. Lee SK, Normand C, McMillan D, Ohlsson A,
Vincer M, Lyons C. Evidence for changing
guidelines for routine screening for reti-
nopathy of prematurity. Arch Pediatr Ado-
lesc Med. 2001;155(3):387–395
27. Brennan R, Gnanaraj L, Cottrell DG. Retinop-
athy of prematurity in practice. I: screening
for threshold disease. Eye (Lond). 2003;
17(2):183–188
28. Yanovitch TL, Siatkowski RM, McCaffree M,
Corff KE. Retinopathy of prematurity in in-
fants with birth weight ⬎ or ⫽ 1250 grams-
incidence, severity, and screening guideline
cost-analysis. J AAPOS. 2006;10(2):128 –134
29. Ells A, Hicks M, Fielden M, Ingram A. Severe
retinopathy of prematurity: longitudinal ob-
servation of disease and screening implica-
tions. Eye (Lond). 2005;19(2):138 –144
30. Haugen OH, Markestad T. Incidence of reti-
nopathy of prematurity (ROP) in the west-
ern part of Norway: a population-based ret-
rospective study. Acta Ophthalmol Scand.
1997;75(3):305–307
31. Larsson E, Holmstrom G. Screening for ret-
inopathy of prematurity: evaluation and
modification of guidelines. Br J Ophthalmol.
2002;86(12):1399 –1402
32. Schalij-Delfos NE, Cats BP. Retinopathy of
prematurity: the continuing threat to vision
in preterm infants. Dutch survey from 1986
to 1994. Acta Ophthalmol Scand. 1997;75(1):
72–75
33. Section on Ophthalmology, American Acad-
emy of Pediatrics; American Academy of
Ophthalmology; American Association for
Pediatric Ophthalmology, Strabismus.
Screening examination of premature in-
fants for retinopathy of prematurity [pub-
lished correction appears in Pediatrics.
2006;118(3):1324]. Pediatrics. 2006;117(2):
572–576
34. Wilkinson AR, Haines L, Head K, Fielder AR.
UK retinopathy of prematurity guideline. Eye
(Lond). 2009;23(11):2137–2139
35. Palmer EA, Flynn JT, Hardy RJ, et al. Inci-
dence and early course of retinopathy of
prematurity: the Cryotherapy for Retinopa-
thy of Prematurity Cooperative Group. Oph-
thalmology. 1991;98(11):1628 –1640
36. Termote JU, Donders AR, Schalij-Delfos NE,
et al. Can screening for retinopathy of pre-
maturity be reduced? Biol Neonate. 2005;
88(2):92–97
37. Hellström A, Hard AL, Engstrom E, et al. Early
weight gain predicts retinopathy in preterm
infants: new, simple, efficient approach to
screening. Pediatrics. 2009;123(4). Avail-
able at: www.pediatrics.org/cgi/content/
full/123/4/e638
38. Schalij-Delfor NE, for the Dutch Working
Group on ROP. Prematuren retinopathie,
richtlijn voor screening. Van der Velde, Ed.
Richtlijn oogheelkunde. Deventer, Uitgave
Commissie Kwaliteit van het Nederlands
Oogheelkundig Gezelschap; 1999
39. Gilbert C, Fielder A, Gordillo L, et al. Charac-
teristics of infants with severe retinopathy
of prematurity in countries with low, mod-
erate, and high levels of development: impli-
cations for screening programs. Pediatrics.
2005;115(5). Available at: www.pediatrics.
org/cgi/content/full/115/5/e518
40. Fortes Filho JB, Eckert GU, Valiatti FB, Dos
Santos PG, da Costa MC, Procianoy RS. The
influence of gestational age on the dynamic
behavior of other risk factors associated
with retinopathy of prematurity (ROP).
Graefes Arch Clin Exp Ophthalmol. 2010;
248(6):893–900
41. Allegaert K, Casteels I, Cossey V, Devlieger H.
Retinopathy of prematurity: any difference
in risk factors between a high and low risk
population? Eur J Ophthalmol. 2003;
13(9 –10):784 –788