Abstract
Purpose :
Prior risk models for retinopathy of prematurity (ROP) have been based on risk factors assessed at a single time point, even though risk factors can change over time. We sought to develop and validate a dynamic prediction model for treatment-requiring ROP (TR-ROP) using time-updated risk factors in two large, broad-risk cohort studies of premature infants.
Methods :
Secondary analyses were performed using data from the G-ROP-1 study (retrospective, 7483 infants from 29 hospitals, 2006-2012) for model development and data from the G-ROP-2 study (prospective, 3980 infants from 41 hospitals, 2015-2017) for model validation. The outcome was development of TR-ROP (e.g., type 1 or treated ROP). Risk factors considered for the prediction model included birth weight [BW], gestational age [GA], gender, race, oxygen use, Sepsis/NEC, thrombocytopenia, maternal or donor milk feedings, weight gain rate, a time-updating previously validated ROP severity score, and postmenstrual age (PMA). Univariable and multivariable repeated measures logistic regression models were performed to identify predictive risk factors. Area under the ROC curve (AUC) was calculated to evaluate the prediction model performance.
Results :
TR-ROP occurred in 524 (7.0%) of G-ROP-1 infants (median GA 28 wks; median BW 1099g) and 256 (6.4%) of G-ROP-2 infants (median GA 28 wks; median BW 1072g). Using G-ROP-1 data, a dynamic prediction model (Table 1) was developed that included BW (p<0.001), race (p<0.001), week of first enteral feeds during the first 6 weeks of life (p<0.001), time-updating thrombocytopenia (p=0.004), NEC/sepsis (p=0.004), slow weight gain (p=0.004), ROP severity score (p<0.001), PMA, and interaction term between ROP severity score and PMA (p<0.001). This prediction model had an AUC of 0.88, which was significantly higher than a prediction model based on BW and GA only (AUC=0.77, p<0.0001). A prediction model that included significant predictors (BW, GA, race, ROP severity score, PMA, and interaction between ROP severity score and PMA) that were available in both the G-ROP-1 and G-ROP-2 datasets had an AUC of 0.86 in G-ROP-1 and 0.87 in G-ROP-2 (Table 2).
Conclusions :
Our developed and validated dynamic prediction model that considers time-updating ROP severity score and other traditional and novel risk factors may provide a tool for dynamically identifying high risk premature infants that are likely to develop TR-ROP.
This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.