Risk factors and day-14 prediction models for bronchopulmonary dysplasia and coexisting retinopathy of prematurity in very-low-birth-weight or very preterm infants.
4 affiliations
- 1Department of Pediatrics, The First Hospital of Hebei Medical University, Shijiazhuang, China.
- 2Newborn Department, Shijiazhuang Fourth Hospital, Shijiazhuang, China.
- 3Hebei Medical University, Shijiazhuang, China.
- 4Department of Pediatrics, Bayingolin Mongolian Autonomous Prefecture People's Hospital, Korla, China.
Abstract
Bronchopulmonary dysplasia (BPD) and retinopathy of prematurity (ROP) share developmental, oxygen-related, and inflammatory susceptibility. We examined factors associated with BPD and with ROP among infants with BPD and developed prediction models restricted to information available by postnatal day 14. This retrospective two-center study included 437 infants born before 32 weeks of gestation or weighing less than 1,500 g at birth. Conventional logistic regression evaluated mutually adjusted associations. For prediction, 15 prespecified variables available by day 14 were entered into elastic-net logistic regression. Preprocessing and hyperparameter selection were confined to the training data after an outcome-stratified 8:2 split. The classification threshold was selected from training-set out-of-fold predictions; the held-out test set was used once. Discrimination, calibration, Brier score, and repeated nested cross-validation were assessed. Among 437 infants, 172 (39.4%) had BPD; among those with BPD, 89 of 172 (51.7%) had ROP. In the association analysis, lower gestational age, longer invasive ventilation and oxygen therapy, and later first feeding were associated with BPD. Among infants with BPD, longer cumulative respiratory support, lower platelet distribution width (PDW) at 24 hours and day 14, male sex, and shorter birth length were associated with ROP. The day-14 BPD model yielded a held-out area under the receiver operating characteristic curve (AUC) of 0.789 [95% confidence interval (CI): 0.688-0.880], a Brier score of 0.199, and an accuracy of 0.739. The ROP model yielded an AUC of 0.690 (95% CI: 0.500-0.859), a Brier score of 0.235, and an accuracy of 0.629. Mean AUCs across repeated nested cross-validation were 0.846 and 0.752, respectively. Routinely available information through day 14 provided moderate internal discrimination for BPD and for ROP among infants with BPD. The models define a transparent prediction landmark and avoid use of later cumulative treatment variables, but require external and prospective validation before clinical use.
Related
- How should hydrocortisone be used for bronchopulmonary dysplasia in preterm infants?
- Bovine milk-derived osteopontin attenuates hyperoxia-induced lung injury associated with NLRP3 inflammasome regulation in neonatal mice with bronchopulmonary dysplasia
- Toward Endotype-Based Bronchopulmonary Dysplasia: Evolving Definitions and the Role of Translational Models
- Renal outcome at 2 years in neonates with nephrocalcinosis and bronchopulmonary dysplasia: a case-control study.
- Transcatheter atrial septal defect closure in infants weighing 3 to 6 kg with clinically significant bronchopulmonary dysplasia
BibTeX
@article{fan2026risk,
title = {{Risk factors and day-14 prediction models for bronchopulmonary dysplasia and coexisting retinopathy of prematurity in very-low-birth-weight or very preterm infants.}},
author = {Yuqing Fan and Haiyan Ge and Shangbin Li and Xiaonan Han and Xiaojuan Huang and Wenhui Wu and Yurun Wang and Yue Zhang and Jiashun Liu and Lu Zou and Changjun Ren and Hejia Li},
journal = {Translational pediatrics},
year = {2026},
month = sep,
doi = {10.21037/tp-2026-0653},
url = {https://europepmc.org/article/MED/42996726},
note = {CC BY-NC-ND},
}
RIS
TY - JOUR
TI - Risk factors and day-14 prediction models for bronchopulmonary dysplasia and coexisting retinopathy of prematurity in very-low-birth-weight or very preterm infants.
AU - Yuqing Fan
AU - Haiyan Ge
AU - Shangbin Li
AU - Xiaonan Han
AU - Xiaojuan Huang
AU - Wenhui Wu
AU - Yurun Wang
AU - Yue Zhang
AU - Jiashun Liu
AU - Lu Zou
AU - Changjun Ren
AU - Hejia Li
PY - 2026
DA - 2026/09/29
JO - Translational pediatrics
DO - 10.21037/tp-2026-0653
UR - https://europepmc.org/article/MED/42996726
AB - Bronchopulmonary dysplasia (BPD) and retinopathy of prematurity (ROP) share developmental, oxygen-related, and inflammatory susceptibility. We examined factors associated with BPD and with ROP among infants with BPD and developed prediction models restricted to information available by postnatal day 14. This retrospective two-center study included 437 infants born before 32 weeks of gestation or weighing less than 1,500 g at birth. Conventional logistic regression evaluated mutually adjusted associations. For prediction, 15 prespecified variables available by day 14 were entered into elastic-net logistic regression. Preprocessing and hyperparameter selection were confined to the training data after an outcome-stratified 8:2 split. The classification threshold was selected from training-set out-of-fold predictions; the held-out test set was used once. Discrimination, calibration, Brier score, and repeated nested cross-validation were assessed. Among 437 infants, 172 (39.4%) had BPD; among those with BPD, 89 of 172 (51.7%) had ROP. In the association analysis, lower gestational age, longer invasive ventilation and oxygen therapy, and later first feeding were associated with BPD. Among infants with BPD, longer cumulative respiratory support, lower platelet distribution width (PDW) at 24 hours and day 14, male sex, and shorter birth length were associated with ROP. The day-14 BPD model yielded a held-out area under the receiver operating characteristic curve (AUC) of 0.789 [95% confidence interval (CI): 0.688-0.880], a Brier score of 0.199, and an accuracy of 0.739. The ROP model yielded an AUC of 0.690 (95% CI: 0.500-0.859), a Brier score of 0.235, and an accuracy of 0.629. Mean AUCs across repeated nested cross-validation were 0.846 and 0.752, respectively. Routinely available information through day 14 provided moderate internal discrimination for BPD and for ROP among infants with BPD. The models define a transparent prediction landmark and avoid use of later cumulative treatment variables, but require external and prospective validation before clinical use.
SN - 2224-4336
SN - 2224-4344
ER -