Construction and Validation of a Risk Prediction Model for Acute Pulmonary Hemorrhage in Very Low Birthweight Infants
Yunsu Zou, Fei Shen, Jingjing Pan, Jinxin Shen, Yan Guo, Rui Cheng, Xiaoguang Zhou, Yang YangABSTRACT
Objective
Very low birthweight infants (VLBWIs) are at high risk of pulmonary hemorrhage, yet practical and effective early prediction tools remain limited. The study aimed to develop an early predictive model for pulmonary hemorrhage in VLBWIs to support risk management.
Methods
VLBWIs admitted to Children's Hospital of Nanjing Medical University were included in the development cohort (January 1, 2019, to January 1, 2023), while those admitted between January 2, 2023, and July 31, 2025, formed the temporal validation cohort. Infants from Jiangsu Maternal and Children Hospital were used as an external validation cohort. Prediction was performed at 24 h of life, using variables collected within the first 24 h after birth. Eight machine learning algorithms were compared for model development and evaluation.
Results
The development, temporal validation, and external validation cohorts included 775, 208, and 314 VLBWIs, respectively. Seven variables were selected for the final logistic regression model, which was implemented as a web‐based nomogram. The model showed good discrimination, with area under the receiver operating characteristic curves (AUROCs) of 0.841 (95% confidence interval [CI]: 0.800–0.883) in the training set, 0.836 (95% CI: 0.766–0.906) in the internal validation set, and 0.834 (95% CI: 0.768–0.900) in the temporal validation set. Calibration was acceptable in the training and internal validation sets, with some decline in the temporal validation cohort (Brier score 0.181), but improved after exploratory recalibration. External validation yielded an AUROC of 0.778 (95% CI: 0.709–0.847) with acceptable calibration.
Conclusions
This model predicts subsequent pulmonary hemorrhage in VLBWIs within 24 h after birth. Decision curve analysis indicated favorable clinical utility.