Description and purpose
Non-invasive mechanical ventilation (NIV) is widely used in acute hypoxemic respiratory failure (AHRF), but failure is common and delayed intubation worsens outcomes. Current clinical scores have limited accuracy for early prediction. Emerging evidence indicates complex interacting pathophysiological mechanisms. This project integrates multimodal physiological, biomarker, and imaging data using artificial intelligence to enable early identification of patients at high risk of NIV failure.
Purpose
The objectives of this project were to integrate advanced monitoring techniques (respiratory mechanics, non-invasive imaging, and biomarkers) to develop an AI-based early predictive model of NIV failure. The primary objective was NIV failure, defined as endotracheal intubation. Secondary objectives included ventilator-free days, survival outcomes, and the analysis of pathophysiological mechanisms underlying NIV failure.
Expected results
The expected results were the development of an artificial intelligence–based predictive tool for early identification of patients at high risk of NIV failure during the initial ICU stay. The project also provided a risk stratification model for future clinical studies and generated new insights into the main pathophysiological mechanisms underlying NIV failure.
Achieved results
Within the project, relevant results were achieved through the application of advanced analytical techniques (e.g. unsupervised clustering) to identify distinct pathophysiological phenotypes in patients with respiratory failure. The identified clusters exhibited differentiated profiles of respiratory mechanics, ventilation distribution, and clinical response. Systematic comparison of lung injury biomarkers revealed specific associations between phenotypes, patient’s specificity and damage, and clinical outcomes, supporting a personalized medicine approach.
