Objective To assess the accuracy with which available retinopathy of prematurity (ROP) predictive models detect clinically significant ROP and to what extent and at what risk these models allow for the reduction of screening examinations for ROP. Methods A literature search of the PubMed and Cochrane Library databases was conducted last on May 1, 2015, and yielded 305 citations. After screening the abstracts of all 305 citations and reviewing the full text of 30 potentially eligible articles, the panel members determined that 22 met the inclusion criteria. One article included 2 studies, for a total of 23 studies reviewed. The panel extracted information about study design, study population, the screening algorithm tested, interventions, outcomes, and study quality. The methodologist divided the studies into 2 categories - model development and model validation - and assigned a level of evidence rating to each study. One study was rated level I evidence, 3 studies were rated level II evidence, and 19 studies were rated level III evidence. Results In some cohorts, some models would have allowed reductions in the number of infants screened for ROP without failing to identify infants requiring treatment. However, the small sample size and limited generalizability of the ROP predictive models included in this review preclude their widespread use to make all-or-none decisions about whether to screen individual infants for ROP. As an alternative, some studies proposed approaches to apply the models to reduce the number of examinations performed in low-risk infants. Conclusions Additional research is needed to optimize ROP predictive model development, validation, and application before such models can be used widely to reduce the burdensome number of ROP screening examinations.
ASJC Scopus subject areas