Leveraging artificial intelligence for prediction of pulmonary hemorrhage in preterm infants

Abstract Objectives To identify clinical variables and indicators associated with pulmonary hemorrhage in preterm infants. Methods This case-control study included inborn infants <32 weeks. Data were collected in 12-h epochs from birth until hemorrhage onset or up to 72 h for controls. Machine learning used the Random Forest algorithm. Statistical analysis included T test and Mann–Whitney U test. Results Among 1133 screened infants, 35 had hemorrhage. Mean gestational age was 25.6 ± 1.6 weeks, birth weight 753 ± 224 g, and median onset of hemorrhage was 44.5 h. Affected infants more often required chest compressions and invasive ventilation. Machine learning (accuracy = 83%, AUC = 90%) identified repeated surfactant dosing and postnatal hypotension in the first 12 h of life as top predictors, along with maternal and gestational age. Mortality was higher in cases than controls (19% vs. 3%, p = 0.005). Conclusion Repeated surfactant dosing and early postnatal hypotension are key predictors for pulmonary hemorrhage in preterm infants.