Risk factor ranking of retinopathy of prematurity using classification and regression tree analysis

Numerous risk factors for ROP have been identified, with gestational age and birth weight being the most established. Ongoing improvements in neonatal care may change the critical thresholds for these factors associated with ROP development. We applied a classification and regression tree (CART) analysis to predict the likelihood of developing ROP using a contemporary database.

This cross-sectional study included 1296 premature infants who completed ROP screening by pediatric ophthalmologists from a children's hospital. Relevant infant and maternal parameters were obtained from medical records. CART analysis with the recursive partitioning package was used to predict ROP incidence, treatment need, and plus disease. The dataset was randomized, with 80% for training and 20% reserved for testing the model's accuracy. A 10-fold cross-validation was conducted to assess the model's generalizability.

The mean gestational age was 27.7 ± 2.5 weeks, and the birth weight was 1017 ± 346.9 g; 538 infants developed ROP. The CART analysis produced a predictive model for ROP incidence with a mean absolute error of 0.25. The critical thresholds identified for ROP development were gestational age <27.2 weeks and birth weight <775 g. Two models for ROP progression showed that age <25 weeks and weight <569 g were linked to the need for ROP treatment, and age <24 weeks and non-Black maternal race were associated with plus disease.

Our CART analysis provides updated clinical thresholds for ROP development and progression in a modern U.S. preterm population. Prospective validation across diverse populations will further enhance the model's utility and generalizability.