New model predicts retinopathy of prematurity based on birth characteristics

Mar 11, 2020



A model based on simple birth characteristics may provide a new, more efficient method for predicting the need for retinopathy of prematurity treatment in infants, according to researchers.

A model based on simple birth characteristics may provide a new, more efficient method for predicting the need for retinopathy of prematurity treatment in infants, according to researchers. It might also reduce screening frequency in infants at low risk.

Retinopathy of prematurity (ROP) is a sight-threatening disease common in infants with a gestational age (GA) of 31 weeks or less. It is a serious enough problem that the standard procedure in Swedish hospitals is to screen all infants with a GA of less than 31 weeks for ROP.

However, this approach is costly, since it requires repeated retinal eye examinations and each screening must be performed by a specially trained ophthalmologist. It is also often stressful for infants, and between 2008 and 2015, only a small number (5.7%) of those infants screened for ROP required treatment.

Ms Aldina Pivodic, MSc, a researcher at the University of Gothenburg in Sweden, and her colleagues wanted to see if they could avoid unnecessary screenings by using a predictive model for ROP risk based only on postnatal age, birth weight, sex, gestation age and important interactions. They conducted a study of 7,286 Swedish infants with a GA of 24 to 30 weeks, who were born between 2007 and 2017 and registered in the Swedish National Registry for Retinopathy of Prematurity. Their findings were published in JAMA Ophthalmology. The Researchers analysed the infants’ time-varying birth characteristics using Poisson regression to develop an individualised predictive model for ROP treatment, which they titled the DIGIROP-Birth prediction model.

They found that this model compared favourably to others currently in use, and had the benefit of being based purely on simple birth characteristics rather than complex longitudinal neonatal data, which is often inaccessible to ophthalmologists.

Validations
Researchers validated their model internally by screening 85 infants for ROP and comparing the results with those predicted by their model. The results of the screening were one hundred percent consistent with those of the study. The also estimated that DIGIROP-Birth could have avoided 11% of stressful early screenings in the cohort they analysed.

The researchers also validated their model against four external United States prediction models: the Children’s Hospital of Philadelphia ROP (CHOP-ROP), the Omaha ROP (OMA-ROP), the Colorado ROP (CO-ROP), and the Weight, Insulin Like Growth Factor 1, Neonatal, ROP (WINROP). The validations were performed by applying the five models (DIGIROP-birth and the four published models) to data on 1,485 American infants.

The DIGIROP-Birth model performed well when compared to the four published models in a validation using the U.S. data. Applying the same cut-off, DIGIROP-Birth and CHOP-ROP both achieved a sensitivity of 99%. Specificity was 48.9% for DIGIROP-Birth versus 44.4% for CHOP-ROP.

At a 97.8% sensitivity, DIGIROP-Birth had a specificity of 58.1% compared with 22.4% for OMA-ROP. At a sensitivity of 96.8%, DIGIROP-Birth had a specificity of 49.3%, while WINROP had a specificity of 35.8%.And at a 98.4% sensitivity, DIGIROP-Birth had a specificity of 47.9% compared with 10.5% for CO-ROP.

Further observations
In addition to developing a model based on simple birth characteristics, researchers were able to determine that some birth characteristics are more useful than others in predicting ROP treatment. Ms Pivodic and colleagues noted that, “postnatal age rather than postmenstrual age was a better predictive variable for the temporal risk of ROP treatment.”

Ms Pivodic’s team concede that their study does have some limitations. “Infants born at GA less than 24 weeks could not be included in the prediction model because of the lack of a reference algorithm for birth weight, preventing BWSDS [birth weight standard deviation scores] calculations. Given the small sample size, only a simple model could be developed for these infants, resulting in low predictive ability.”

However, they are confident that their model will help ophthalmologists in diagnosing and treating ROP in infants, writing, “The DIGIROP-Birth model is an accessible online application that appears to be generalisable and to have at least as good test statistics as other models that require longitudinal neonatal data.”

x