Development and external validation of a machine learning model to predict bronchopulmonary dysplasia using dynamic...

Area of Science:

• Neonatology
• Pediatric Pulmonology
• Medical Informatics

Background:

• Bronchopulmonary dysplasia (BPD) is a significant complication in preterm infants.
• Accurate prediction of BPD is crucial for timely intervention and improved outcomes.

Purpose of the Study:

• To enhance the prediction accuracy of moderate or severe bronchopulmonary dysplasia in preterm infants.
• To evaluate the impact of incorporating postnatal dynamic factors into predictive models.

Main Methods:

• Retrospective cohort study of neonates born before 32 weeks gestation.
• Assessed static perinatal risk factors and dynamic factors (respiratory support, oxygen concentration, blood gas analysis) within the first 7 days.
• Developed and validated an integrated prediction model using development, internal, and external datasets.

Main Results:

• The integrated model combining static and dynamic factors demonstrated superior predictive performance (AUROC 0.841 in development set).
• Internal validation showed significantly improved AUROC for the integrated model (0.912 vs. 0.805).
• External validation confirmed the model's robust performance (AUROC 0.814).

Conclusions:

• Incorporating early dynamic factors, specifically respiratory support and blood gas analysis, substantially improves BPD prediction accuracy.
• The developed integrated model offers a more precise tool for identifying preterm infants at high risk for BPD.
• This enhanced prediction capability can guide clinical management and potentially reduce BPD incidence and severity.