Predicting pulmonary hemorrhage in very low birth weight infants

Background: Pulmonary hemorrhage (PH) in very low birth weight (VLBW) infants is a catastrophic event with significant mortality. Predicting PH clinically is challenging, with no tools for early detection. Our objective was to identify candidate physiologic biomarkers using heart rate (HR) and pulse oximetry (SpO₂) predictive of PH events in VLBW infants. Methods: We conducted a retrospective case-control study of VLBW infants from two level IV NICUs (2016–2023). We matched infants with PH to controls using gestational age, birth weight, and postnatal age. Taking HR and SpO₂ data from 72 hours before and after PH diagnosis and age-matched times for controls, we calculated a pulse oximetry warning score (POWS) using hourly mean, skewness, kurtosis, and cross-correlation of HR and SpO₂. Results: We analyzed 48 infants, 24 PH cases, and 24 controls, with similar characteristics. Infants who developed PH exhibited a three-fold rise in POWS starting 24 hours before PH diagnosis. Control infants showed no significant change in POWS. Conclusion: POWS, a predictive model for cardiorespiratory deterioration, showed a rise in predicted risk before the clinical diagnosis of PH. This suggests that cardiorespiratory deterioration signatures precede the clinical diagnosis of PH and can be recognized with a computer algorithm. Impact: There are subtle vital sign changes that may indicate impending pulmonary hemorrhage in very low birth weight infants. Cardiorespiratory predictive monitoring could be useful for early warning of deterioration due to pulmonary hemorrhage. This study lays the foundation for the utility of predictive analytics for pulmonary hemorrhage prediction and suggests that a dedicated predictive model for pulmonary hemorrhage may be helpful.