Artificial intelligence-based models for predicting bronchopulmonary dysplasia in preterm infants: a systematic review

Background Bronchopulmonary dysplasia (BPD) is one of the most common chronic respiratory complications in preterm infants. Early identification of high-risk patients is essential for individualizing preventive strategies and optimizing therapeutic decisions. Artificial intelligence (AI) and machine learning (ML) models have emerged as promising tools to improve predictive performance in this population. Objective To evaluate and synthesize the predictive performance of AI-based models for BPD development in preterm infants and to assess their methodological quality. Methods A systematic review was conducted following PRISMA recommendations. We searched PubMed, Scopus and Cochrane Library up to February 2026. Studies developing or validating AI-based models for predicting BPD in preterm infants born at <32 weeks of gestation and/or with a birth weight < 1,500 g were included. Methodological quality and risk of bias were assessed using the PROBAST tool. Results Eighteen studies were included. Random Forest and XGBoost were the most frequently used algorithms. AUC values for the best-performing models ranged from 0.80 to 1.00, showing improved performance as postnatal variables were incorporated. However, the PROBAST assessment revealed a high overall risk of bias in 72.2% of the studies, primarily due to deficiencies in the analysis domain. Clinical applicability raised high concern in 38.9% of the studies due to a lack of external validation and heterogeneity in outcome definitions. Conclusions AI-based models show high discriminative capacity for BPD prediction; however, they have not yet demonstrated clear superiority over conventional statistical models in robust comparisons. Their routine clinical implementation is currently limited by poor methodological quality and scarce external validation. Systematic Review Registration https://www.crd.york.ac.uk/PROSPERO/view/CRD420261466494 , PROSPERO: CRD420261466494, registered retrospectively on August 4, 2026.