Retinopathy of prematurity: A comprehensive risk analysis for prevention and prediction of disease

• Published in: PloS one Vol. 12; no. 2; p. e0171467
• Main Authors: Owen, Leah A, Morrison, Margaux A, Hoffman, Robert O, Yoder, Bradley A, DeAngelis, Margaret M
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 14-02-2017; Public Library of Science (PLoS)
• Subjects: Age; Analysis of Variance; Biology and Life Sciences; Birth Weight; Blindness; Classification; Criteria; Diagnosis; Ecological risk assessment; Feasibility studies; Gestational Age; Health aspects; Health screening; Humans; Infant, Newborn; Infant, Premature; Infants; Intensive care; Intensive Care Units, Neonatal; Logistic Models; Lung diseases; Magnesium; Mathematical models; Medical examination; Medical screening; Medicine; Medicine and Health Sciences; Mitigation; Morbidity; Multiple births; Neonatal Screening - methods; Neonates; Newborn babies; Pediatrics; People and Places; Physical Sciences; Population (statistical); Prediction models; Premature infants; Prevention; Prognosis; Prophylaxis; Regression analysis; Research and Analysis Methods; Retina; Retinopathy; Retinopathy of prematurity; Retinopathy of Prematurity - diagnosis; Retinopathy of Prematurity - prevention & control; Retrospective Studies; Risk analysis; Risk Assessment - methods; Risk Assessment - statistics & numerical data; Risk Factors; ROC Curve; Statistical analysis; Statistical models; Studies; Surgery; Variables; United States > US; Utah
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0171467

Retinopathy of prematurity (ROP) is a blinding morbidity of preterm infants. Our current screening criteria have remained unchanged since their inception and lack the ability to identify those at greatest risk.
We sought to comprehensively analyze numerous proposed maternal, infant, and environmental ROP risk variables in a robustly phenotyped population using logistic regression to determine the most predictive model for ROP development and severity. We further sought to determine the statistical interaction between significant ROP risk variables, which has not previously been done in the field of ROP. We hypothesize that our comprehensive analysis will allow for better identification of risk variables that independently correlate with ROP disease. Going forward, this may allow for improved infant risk stratification along a time continuum from prenatal to postnatal development, making prevention more feasible.
We performed a retrospective cohort analysis of preterm infants referred for ROP screening in one neonatal intensive care unit from 2010-2015. The primary outcome measure was presence of ROP. Secondary outcome measures were ROP requiring treatment and severe ROP not clearly meeting current treatment criteria. Univariate, stepwise regression and statistical interaction analyses of 57 proposed ROP risk variables was performed to identify variables which were significantly associated with each outcome measure.
We identified 457 infants meeting our inclusion criteria. Within this cohort, numerous factors showed a significant individual association with our ROP outcome measures; however, stepwise regression analysis found the most predictive model for overall ROP risk included estimated gestational age, birth weight, the need for any surgery, and maternal magnesium prophylaxis. The corresponding Area Under the Curve (AUC) for this model was 0.8641, while the traditional model of gestational age and birth weight predicted ROP disease less well with an AUC of 0.8489. Development of severe ROP was best predicted by estimated gestational age (week), the need for any surgery and increased probability of death or moderate-severe BPD at 7 days. Finally, the model most predictive for type 1 ROP included estimated gestational age (week) and the presence of severe chronic lung disease. No significant statistical interaction was found between variables.
Our work is unique as we report comprehensive analysis of the greatest number of proposed ROP risk variables to date in a robustly phenotyped population. We describe novel risk models for our ROP outcome measures and demonstrate independence of these variables using statistical modeling not previously applied to ROP. This may better allow for individual infant risk stratification and importantly mitigation of future risk.