Can artificial intelligence predict failure of non-invasive respiratory support in the neonatal unit?

Six studies, including 3421 infants, were identified. Various AI techniques were used including deep learning models, for example multimodal deep neural networks, as well as simpler machine learning models such as logistic regression and support vector machines. AUC values ranged from 0.78 to 0.93, with most models exhibiting clinically useful performance defined as an AUC >0.8. The modal key predictive factors across the six studies were gestational age, SpO₂ and maximum FiO2. CONCLUSION: AI- generated models for predicting NIV failure as first intention in the NICU setting show potential. Deep learning models demonstrate particular promise; however, further large multicenter externally validated studies are required to assess generalizability and to aid integration into routine clinical practice. Implementation of AI models to predict NIV failure as first intention and post-extubation could lead to improved clinical decision making and personalized care.