Discussion
In the current study, the mean ROP score was significantly higher in
prematures with severe (type -1) ROP compared to the no-ROP group and
type-2 ROP group. The adjusted cut-off points of the model had a high
sensitivity in the prediction of ROP, particularly in type-1 ROP with a
sensitivity of 100% and a specificity of 83%. NPV of the model was
100% for infants would not develop severe ROP (type-1 ROP). The
ROPScore model would provide a significant reduction in the number of
examinations in patients with any stage of ROP and severe type -1 ROP.
The models based on scoring algorithms are thought to be a useful method
to predict the development and severity of ROP, especially in type-1
ROP. Various scoring algorithms consisting of WINROP, PINT-ROP, CHOP-ROP
and CO-ROP scores were established based on the clinical risk factors,
such as postnatal weight gain, serum insulin-like growth factor levels,
birth weight, receiving cardiopulmonary support in
NICU.(9, 11, 17, 18) However, Fierson et
al.(16) stressed that using these algorithms as a ROP
screening method and adapting them to the international population is
still controversial.
In the present study with arranged cutoff points, the sensitivity of
ROPScore was 88.5% for any stage of ROP and 100% for type-1 ROP. In
consistent with our study, Eckert et al.(9) stated
96% sensitivity and Lucio et al.(10) also reported
95.4% sensitivity for type-1 ROP with their adjusted cutoff points.
Another study conducted by Piermarocchi et al.(11)revealed 100% sensitivity in the ROPScore for type-1 ROP.
The negative predictive value (NPV) determined in the current study
implied that the likelihood in which a premature with a ROPscore under a
cutoff point of 14.9 might not develop severe type-1 ROP was 100%,
whilst the probability of the same infants were not expected to be
develop any stage of ROP was 74.1%. In compliance with our study,
Cagliari et al.(19) reported 100% of NPV and 22% PPV
for the prediction of ROP with a cutoff point of 14.5 using ROPScore
algorithm. In addition, Lucio et al.(10) stated
similar NPV and PPV’s in premature infants with cutoff ROPscore ≥16.6.
All detailed comparisons of the studies with their adjusted cutoff
points for the ROPScore were summarized in table 3. The low GA and BW
are the two major risk factors for development of ROP, which are used in
numbers of ROP algorithms.(19-22) Our analysis
revealed that the mean AUC for ROPScore was significantly higher than
that of for BW and GA in type-1 ROP. Eckert et al.(9)also reported significant the mean AUCs in terms of ROP score, BW and GA
both in any stage ROP and type-1 ROP.
In the present study, ROPScore algorithm hypothetically would provide a
reduction in the number of examinations by 17.5% for any stage of ROP
and 70.9 % for the infants with type-1 ROP (Table 3). Previous studies
analyzing the ROPScore model defined various percentages related to
decline in the total number of infants needing screening examination for
ROP.(9,10,18-22) The discrepancies among the studies
in terms of specificity, sensitivity, NPV, PPV and the total estimated
reduction rate in screening examinations could be attributed to several
confounding factors of the cohorts, such as diversity in patient
demographics, health care systems and clinical courses. (Sepsis, chronic
lung disease, interventricular hemorrhage, etc.)
Finally, a number of potential limitations need to be considered. First,
as the current study had a small sample size, further studies with a
large multi-centered cohort with being stratified confounding factors is
required to determine the validity of the ROPScore. Second, the ROPScore
model may miss aggressive posterior ROP (APROP) cases characterized by
rapid progression in the very early weeks of life. Lastly, the algorithm
includes only preterm infants of ≤30 weeks and/or ≤1500 g, whereas
several studies have shown that older and larger preterm infants may
also have any type of ROP or severe ROP, particularly in underdeveloped
and developing countries.(1, 23, 24)
In conclusion
To the best of our knowledge, this is the first study investigating the
efficacy of ROPScore algorithm in prediction of any stage and severe
type -1 ROP in our population. The mean ROPScore was significantly
higher in the infants with type -1 ROP. The ROPScore of was
significantly higher than the AUCs for GA and BW in detection of severe
ROP. The sensitivities of ROPScore algorithm based on adjusted cut-off
values were 100% and 88.5 for severe (type -1) ROP and any stage of ROP
respectively. The negative predictive value of ROPScore was 100%. The
mean AUCs value for ROPScore was higher than that of BW and GA. Our
study shows that ROPScore scoring model is a useful method to early
identify the premature infants having high risk for development of ROP,
particularly in type-1 ROP requiring prompt intervention to prevent
permanent loss of vision.