Prediction model to predict type 1 retinopathy of prematurity using gestational age and birth weight (PW-ROP)

To develop a prediction model for type 1 retinopathy of prematurity (ROP) from an Asian population.

This retrospective cohort study included 1043 premature infants who had ROP screening in a tertiary hospital in Hong Kong from year 2006 to 2018. The ROP prediction model was developed by multivariate logistic regression analyses on type 1 ROP. The cut-off value and the corresponding sensitivity and specificity were determined by receiver operating characteristic curve analysis. A validation group of 353 infants collected from another tertiary hospital in another region of Hong Kong from year 2014 to 2017 was used for external validation.

There were 1043 infants in the study group. The median gestational age (GA) was 30 weeks and 1 day and median birth weight (BW) was 1286 g. The prediction model required only GA and BW as parameters (prematurity-birth weight ROP (PW-ROP)). The area under curve value was 0.902. The sensitivity and specificity were 87.4% and 79.3%, respectively. Type 1 ROP developed in 0.9%, 17.4% and 50% of infants with PW-ROP scores<0, between 0 and <300, and ≥300 respectively (p<0.001). On external validation, our prediction model correctly predicted 95.8% of type 1 ROP (sensitivity=95.8%, specificity=74.8%) in the validation group.

The PW-ROP model is a simple model which could predict type 1 ROP with high sensitivity and specificity. Incorporating this model to ROP examination would help identify infants at risk for ROP treatment.