Published Genetic Variants in Retinopathy of Prematurity: Random Forest Analysis Suggests a Negligible Contribution to Risk and Severity

Abstract

Purpose: Our recent investigations suggested association between severe retinopathy of prematurity (ROP) and some genetic polymorphisms contributing to angiogenesis. While these findings may help to identify specific elements in ROP pathogenesis, the predictive value of these genetic variants at birth is unknown. We applied a high-dimensional nonparametric method called random forest technique (RFT) to evaluate the predictive value of genetic polymorphisms in ROP at birth. Methods: We used published genetic (i.e., VEGF T⁻⁴⁶⁰C, G^{+ 405}C, and C⁻²⁵⁷⁸A; IGF-I receptor G^{+ 3174}A, angiopoietin II G⁻³⁵C; estrogen receptor PvuII Pp; and endothelial NO-synthase 27-bp b/a and T⁻⁷⁸⁶C) and birth data of 134 preterm infants without and 103 preterm infants with ROP requiring laser or cryotherapy. We used RFT to determine the relative importance scores (IS) of each clinical parameter at birth and genetic polymorphisms in the prediction of ROP. The accuracy of ROP prediction at birth was calculated when birth data, genotype data, and birth data PLUS genotype data were taken into account. Results: The most important predictors of ROP were prematurity, low birth weight, intrauterine retardation, and Apgar scores with IS values between 7.46 and 13.20. IS values of genotype data were much lower in the range between 0.86 and 4.19. When birth data solely, genotype data solely, and birth data plus genotype data together were used for prediction, the accuracy of prediction was 0.653, 0.636, and 0.674, respectively. Conclusions: The tested genetic polymorphisms (including those published as significant risk factors of ROP) are not good predictors of ROP at birth.

KEYWORDS:

• polymorphism
• prediction
• random forest
• ROP