ROC Analysis and Establishing ROPScore Accuracy (Sensitivity and
Specificity)
At first, the accuracy of the model regarding to predict onset and
severity of ROP was determined by comparing the area under curves (AUC)
for the ROPScore, GA and BW to each other. Due to the unavailability of
analysis in SPSS software, the comparison of data for AUCs was performed
manually using a formulation described by Hanley et
al.(15) The optimal cutoff points of the continuous
values for sensitivity and specificity have been tailored according to
this formula.
In the present study, the standard cutoff values were 11 for any stage
of ROP and 14.5 for type-1 ROP according to previously described by
Eckert et al.(9) To maintain optimized sensitivity,
but also to optimize specificity (i.e., to decrease false positives),
the best cutoff points were arranged as in several similar studies, and
we set a customized value of 12.3 for any stage ROP and 14.9 for type-1
ROP.(10,11) Positive predictive values (PPVs; the
likelihood of a preterm baby with a score above the cut-off point,
indicating that any stage of ROP or type-1 ROP would develop) and
negative predictive values (NPVs; the likelihood of a preterm infant
with a score below the cutoff point, indicating that any stage of ROP or
type-1 ROP would not develop) were also obtained with 95% confidence
intervals (CIs) and then compared to each other for every predictors.