Systematic review and meta-analysis of risk prediction models for retinopathy of prematurity in preterm infants

Li Li ¹, Yanlin Gao ¹, Wei Chen ¹

⁰Tianjin Key Laboratory of Ophthalmology and Vision Science, Tianjin Eye Hospital, Tianjin 300020, P.R. China.

Biomedical Reports +

|

November 12, 2025

View abstract on PubMed

Summary

Risk prediction models for retinopathy of prematurity (ROP) show good potential for clinical screening in preterm infants. However, moderate bias and limited validation hinder their widespread use.

Area Of Science

Neonatal Ophthalmology
Medical Informatics
Public Health

Background

Retinopathy of prematurity (ROP) is a leading cause of blindness in preterm infants due to abnormal retinal vascular development.
Existing ROP risk prediction models vary significantly in methodology and performance, limiting their clinical applicability.
Reliable ROP risk prediction is essential for optimizing screening and intervention strategies in neonatal care.

Purpose Of The Study

To systematically review and meta-analyze the effectiveness of current risk prediction models for ROP in preterm infants.
To evaluate the discriminative ability and identify limitations of existing ROP prediction models.
To provide evidence-based recommendations for developing more robust and generalizable ROP screening tools.

Main Methods

Systematic literature search of PubMed, Cochrane Library, Web of Science, and Embase databases.
Included 28 studies (2009-2025) with 72,991 preterm infants, assessing risk of bias using PROBAST.
Performed meta-analysis, heterogeneity testing, subgroup, and sensitivity analyses to evaluate model performance (pooled AUC=0.87).

Main Results

All ROP risk prediction models exhibited a moderate risk of bias, often due to retrospective designs and inadequate confounding factor control.
The pooled discriminative ability (AUC) was 0.87, but significant heterogeneity (I²=99.2%) was observed across models and regions.
Subgroup analyses revealed high heterogeneity in traditional statistical and machine learning models, and regional variations in Asia and North America + Europe.

Conclusions

Current ROP risk prediction models demonstrate good clinical potential and discriminative ability for screening preterm infants.
Moderate risk of bias and insufficient validation (especially external) limit the generalizability of existing models.
Future research should focus on prospective designs, larger sample sizes, and rigorous external validation to enhance ROP model accuracy and universality.

Keywords:

meta-analysis prediction model preterm infant retinopathy of prematurity systematic review