Acute concomitant esotropia in children

screening eventually develop ROP requiring treatment. This has led to a number of proposed models to improve the efficiency of ROP screening. The CHOP postnatal weight gain, birthweight and gestational age ROP risk model is one of these models (Binenbaum et al. 2012). The CHOPROP model uses rate of weight gain in addition to gestational age and birthweight. The CHOP-ROP authors studied 524 infants and used logistic regression to determine a statistical model to predict risk and calculate a risk score. There are specific alarm cutoffs and different coefficients depending on the use of daily or weekly weights (Binenbaum et al. 2012). The CHOP-ROP model had 100% sensitivity in the recently published article by Piermarocchi et al. Our group also sought to validate the CHOP-ROP model of predicting severe ROP in a Colorado cohort of infants at risk of ROP. We performed a retrospective review of 1,309 infants who met 2013 ROP screening criteria in the years 2006– 2014. Infants missing all 4 weekly weights (n = 124) and also those without a final ROP outcome (n = 7) were excluded. Retinopathy of Prematurity (ROP) was defined according to the International Classification of ROP with maximumstageand lowest zoneofROP in the worse eye. We used the Early Treatment for Retinopathy of Prematurity (ETROP) study to define type 1 and type 2 ROP (Early Treatment For Retinopathy Of Prematurity Cooperative Group, 2003). We defined severe ROP as ETROP type1or type 2ROPand low-grade ROP as ROP that is neither type 1 nor type 2. The CHOP model was applied using weekly weights. Our goal was to evaluate the sensitivity, specificity as well as negative predictive value (NPV) and positive predictive value (PPV) of the CHOPROP model in our cohort. Our cohort had 70 babies (5.9%) with type 1 ROP, 48 (4.1%) with type 2 ROP, 280 (23.8%) with low-grade ROP and 780 babies (66.2%) with no ROP. Application of the CHOP-ROP model to our infants would have triggered screening for 542 infants and reduced the number of infants screened compared to 2013 national criteria by 54%. The CHOP-ROP model had a sensitivity for severe ROP of 97.5% (95% CI: 92.8–99.5%) and specificity for no ROP of 72.2% (95% CI: 68.9– 75.3%). Negative predictive value (NPV) of no ROP and severe ROP was 88.5% and 99.5%, respectively. The PPV of severe and any ROP was 21.2% and 60.0%. However, three infants with severe ROP (one with type 1 and two with type 2) did not set off an alarm. There were some limitations to the study. We had some missing data, and it was limited to a Colorado population, which could affect generalizability of the results. In our Colorado cohort of infants, the CHOP-ROP model had very high sensitivity for detecting type 1 and type 2 ROP. However, three infants with severe ROP were not detected, leaving the important question – what amount of risk of missing a baby with severe ROP is acceptable? In other words, as we look to adopt a more efficient ROP screening model, what is the acceptable lower end of the 95% confidence interval? Further validations with larger and more diverse patient populations are needed to find an ideal ROP screening algorithm before implementation.