Related Experiment Video
Updated: Aug 6, 2026
Monitoring Dynamic Growth of Retinal Vessels in Oxygen-Induced Retinopathy Mouse Model
Published on: April 2, 2021
Corrected gestational age-specific prediction models for identifying treatment-requiring retinopathy of prematurity
Sao Sugimoto1, Masako Nagahara1, Kentaro Hayashi1
1Department of Ophthalmology, The University of Tokyo Hospital, Tokyo, Japan.
Insights
Developing corrected gestational age (CGA)-specific models for retinopathy of prematurity (ROP) shows evolving risk factors. These models can help identify treatment-requiring ROP in preterm infants as they grow.
Area of Science:
- Neonatalogy
- Ophthalmology
- Biostatistics
Background:
- Retinopathy of prematurity (ROP) is a leading cause of blindness in preterm infants.
- Early identification of treatment-requiring ROP is crucial for timely intervention.
- Current prediction models may not fully capture the dynamic nature of ROP risk.
Purpose of the Study:
- To develop and validate corrected gestational age (CGA)-specific prediction models for treatment-requiring ROP.
- To utilize routinely available systemic factors for ROP risk prediction.
- To assess how predictor combinations and model performance change over time with advancing CGA.
Main Methods:
- Retrospective cohort study of preterm infants meeting ROP screening criteria.
- Development of multivariable logistic regression models at specific CGAs (28, 30, 32, 34 weeks).
- Optimal predictor sets identified using Akaike information criterion; model performance evaluated by cross-validation and AUC.
Main Results:
- Predictor combinations varied significantly across CGAs, indicating evolving risk factors.
- Growth and inflammatory indices were consistently important predictors.
- Model discrimination improved with advancing CGA, with AUCs ranging from 0.650 to 0.842.
Conclusions:
- CGA-specific prediction models can characterize the changing risk of treatment-requiring ROP.
- Systemic factors play a dynamic role in ROP development.
- Prospective multicenter validation is needed for clinical implementation.
Purpose:
To develop and validate time-updated, corrected gestational age (CGA)-specific prediction models for identifying treatment-requiring retinopathy of prematurity (ROP) using routinely available systemic factors, and to assess longitudinal changes in predictor combinations and model performance with advancing CGA.
Study Design:
Retrospective cohort study.
Methods:
Preterm infants who met institutional ROP screening criteria from a single tertiary center were included, excluding those with aggressive ROP. Multivariable logistic regression models were independently developed at CGA 28, 30, 32, and 34 weeks using routinely available systemic variables summarized from birth to each CGA. For each CGA-specific analysis, infants treated before the target CGA were excluded. Exhaustive subset selection based on the Akaike information criterion identified CGA-specific optimal predictor sets. Model performance was evaluated using stratified five-fold cross-validation with out-of-fold predictions, and discrimination was assessed by the area under the receiver operating characteristic curve (AUC).
Results:
Predictor combinations selected by the optimal models differed across CGAs, indicating longitudinal changes in the predictive relevance of systemic factors. Growth-related and inflammatory indices were repeatedly selected, while overall predictor composition evolved with advancing CGA. Discrimination improved at later CGAs, with cross-validated AUCs of 0.650, 0.716, 0.809, and 0.842 at CGA 28, 30, 32, and 34 weeks, respectively.
Conclusion:
These findings suggest that CGA-specific prediction models may help characterize the evolving risk of treatment-requiring ROP. However, prospective multicenter validation and further refinement are required before clinical application.
