Prediction of noninvasive ventilation failure: a narrative review.

Abstract

With the increasing clinical application of noninvasive ventilation (NIV) and the accumulation of evidence-based medical data, NIV has become an important therapeutic intervention for various types of respiratory failure. However, a subset of patients still experience NIV failure. Compared with those with successful NIV, patients who develop NIV failure tend to have a poorer prognosis and an elevated risk of mortality. Therefore, the accurate identification and early prediction of NIV failure risk are crucial for guiding clinical decision-making and optimizing treatment strategies. This article reviews recent advances in research on the risk prediction of NIV failure, covering single-variable prediction approaches, simple variable combination-based prediction, weight-based multivariable prediction models, and emerging artificial intelligence algorithm-driven prediction models. Despite the progress made in NIV failure prediction, existing models still have limitations, such as insufficient generalizability, complex calculation of some multivariable scoring systems, and poor interpretability of artificial intelligence (AI) models. The core contribution of this review is to systematically sort out the advantages and limitations of various prediction methods, integrate prediction indicators and models suitable for different patient populations, and provide practical guidance for clinical practice and direction for future research.