Abstract
AIM: To compare discrimination and calibration of prognostic models for pulmonary outcomes in very preterm (VPT) infants born < 32 weeks' gestation when including the mean airway pressure (MAP), the fraction of supplemental oxygen (FiO(2)) and the respiratory severity score (RSS) reflecting parameters of ventilation and oxygenation during the first 24 and 72 h of life. METHODS: In this retrospective single center study of 168 VPT infants, the mean airway pressure (MAP), the fraction of supplemental oxygen (FiO(2)) or RSS (considering MAP and FiO(2)) were added to a baseline model of clinical risk factors to assess the improvements for prediction of bronchopulmonary dysplasia (BPD). RESULTS: The baseline model demonstrated good calibration (slope 1.02) and discrimination (AUC 0.85) for overall BPD (BPD28), and adding any of the parameters of ventilation resulted only in slight improvement in discrimination (AUC 0.86). For moderate/severe BPD (BPD36), overprediction in the lower extremes and underprediction in the upper extremes became evident for the baseline model. While adding MAP rendered optimal specificity (81%), sensitivity (91%) was highest for FiO(2). MAP was substantially better at improving calibration for BPD36 (slope 0.98) than FiO(2) (slope 0.87). Using RSS and expanding the models to the first 72 h of life did not result in any improvements. CONCLUSION: Adding parameters of ventilation and oxygenation improves baseline models to predict the risk of BPD28 and BPD36 early after birth. Particularly, our data encourage considering MAP as potential predictor in the development of future risk models to improve the prediction accuracy and to solidify early treatment decisions intended to prevent BPD.