A logistic regression model using six readily available perinatal variables provides good discrimination and calibration for predicting severe intraventricular hemorrhage in preterm infants, enabling immediate risk stratification at birth.
The incidence of severe intraventricular hemorrhage (SIVH) among preterm infants in this multicenter cohort was 6.66%.
Six perinatal predictors of SIVH were identified: gestational age, birth weight, endotracheal intubation in the delivery room, chest compressions in the delivery room, surfactant administration in the delivery room, and chorioamnionitis.
Logistic regression achieved the highest mean AUC among eight machine learning algorithms evaluated during 10-fold cross-validation in the training set.
The logistic regression model demonstrated the best discriminative ability among all eight algorithms in the independent hold-out validation set, with an AUC of 0.804.
The logistic regression model maintained robust performance in temporal validation using an independent cohort, with an AUC of 0.822.
An online interactive tool was developed to facilitate clinical application of the logistic regression model.
This research suggests that a relatively simple statistical model using only six pieces of information available at or shortly after birth can accurately predict which premature babies are at high risk for severe bleeding in the brain (severe intraventricular hemorrhage, or SIVH). The six factors used are: how early the baby was born (gestational age), birth weight, whether a breathing tube was inserted in the delivery room, whether chest compressions were needed during resuscitation, whether a lung-coating medicine called surfactant was given in the delivery room, and whether the placenta showed signs of infection (chorioamnionitis). Among more than 3,300 premature infants studied across multiple hospitals in China, about 1 in 15 developed this serious condition. The study tested eight different types of computer-based prediction models and found that a traditional logistic regression model performed best, achieving accuracy scores (AUC) of approximately 0.80–0.82 across different validation tests, including testing on data from a completely separate time period and across different geographic locations. This level of accuracy indicates good predictive ability, and the model was also well-calibrated, meaning its predicted probabilities closely matched actual outcomes. This research suggests that clinicians could use these six routinely collected pieces of information to identify premature infants at high risk for severe brain bleeding immediately at birth, without waiting for postnatal test results. An online tool was created to make this prediction easy to apply in clinical settings. Early identification of at-risk infants could potentially allow medical teams to take preventive or monitoring measures sooner, though further research would be needed to determine how such predictions should guide clinical care.
Shen F, Jin C, Wang F, Cheng R, Rong H, Yang Y, et al.. (2026). Development of a machine learning-based multicenter study for predicting severe intraventricular hemorrhage in preterm infants using perinatal variables.. Frontiers in neurology. https://doi.org/10.3389/fneur.2026.1953566