AI Model Predicts PDA Hemodynamic Severity in Premature Infants

Researchers retrospectively analyzed electronic medical records from very low birth weight infants with patent ductus arteriosus admitted to Samsung Medical Center’s neonatal intensive care unit between 2013 and 2022. After excluding infants with congenital heart disease or missing key echocardiographic parameters, they trained machine-learning models on 85% of the data and tested them on the remaining 15%. The supplied article text reports validation across 1,033 infants and says the model used routine NICU data, with 12 hours of data required.

The Extra Trees model performed best, with an area under the receiver operating characteristic curve (AUC) of 0.829. Sensitivity was 0.767, specificity 0.766, positive predictive value 0.694, and negative predictive value 0.825. For records involving transitions between hemodynamic states, the AUC was 0.700, with sensitivity of 0.722 and specificity of 0.606 using a transition-optimized threshold.

The findings suggest potential for decision-support research, but do not establish that the model improves care or outcomes. This was a retrospective, single-center study with a held-out test set; external validation and prospective assessment are not reported in the supplied text. The model should not be treated as a substitute for echocardiographic assessment or as a basis for changing management based on these results alone.