Purchase this article with an account.
Duoru Lin, Jingjing Chen, Zhuolin Lin, Kai Zhang, Jialing Huang, Xiaohang Wu, Zhenzhen Liu, Lisha Wang, Yizhi Liu, Weirong Chen, Haotian Lin; Prediction for risk of congenital cataracts based on non-genetic factors. Invest. Ophthalmol. Vis. Sci. 2019;60(9):6483.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Traditional preventive measures of congenital anomalies include genetic and prenatal testing and routine neonatal screening, most of which are invasive or costly. We aimed to develop a non-invasive, economical, and easy practical identification model for infants with high risk of congenital cataracts (CC), a typical congenital anomaly and one of the leading causes of childhood blindness.
This case-control study included 1129 CC patients from the national CC prevention and cure center in China and 609 healthy controls from kindergartens and communities. Data of demography, histories of family heredity and gestation, birth conditions, living environment and family situation were collected and compared. CC identification models were established by random forest (RF), naive bayes (NB), and adaptive boosting (Ada), which were tested by internal 4-fold cross validation and external validation using another prospectively collected dataset (98 CC patients and 100 healthy controls). A clinical test for “finding a needle in a haystack” was also performed.
Higher proportions of family history of CC, ≥2nd fetus, pregnant virus infection, preterm baby, eutocia, oxygen inspiration/infant incubator, complicated systemic diseases, parental smoking, low levels of parental education and household income were found among CC patients than those of healthy controls. The CC identification models showed a high discrimination in both the 4-fold cross validation (AUC= 0.904[95%CI: 0.877-0.931] in RF; 0.891[0.862-0.919] in NB; 0.869[0.838-0.900] in Ada) and external validation (AUC= 0.844±0.05), and also performed well in the clinical tests for finding a needle in a haystack (AUC=0.983±0.01 in the first round and 0.922±0.02 in the second round).
Our study show that the non-genetic factors of CC patients were markedly different from those of healthy controls. The CC identification models can accurately discriminate CC patients from healthy children, which made up for the deficiency of current prenatal testing and neonatal screening in CC prevention.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.
This PDF is available to Subscribers Only