Respiratory physiology after resupination following prone ventilation to predict 28-day mortality in mechanically ventilated patients: a machine learning analysis

Résumé fourni par la source

The clinical significance of resupination parameters following prone positioning remains largely unknown. This study employed machine learning to predict the survival of patients receiving mechanical ventilation (MV) by analyzing oxygenation and respiratory mechanics after resupination. Data were extracted from the COVID-Predict Dutch Data Warehouse. Patients receiving MV who underwent the supine–prone–supine sequence were selected, and the variables related to respiratory physiology within 4 h before proning and after resupination were recorded. Machine learning models were trained on the features selected using LASSO regression to predict the 28-day mortality. Patients who did not survive (157/522, 30.1%) had lower PaO2/FiO2 values, higher ventilatory ratios, increased physiological dead space, higher driving pressure, and lower static and dynamic lung compliance values at resupination. The predictive performance of the individual clinical parameters for 28-day mortality was generally modest, and FiO2, PaO2/FiO2, physiological dead space, and dynamic lung compliance were the best predictors of mortality. Overall, XGBoost showed the best recall (0.732) and maintained the highest AUC (0.719), while its F1-score (0.500) was the best for predicting mortality despite a low precision (0.380). Survival after prone positioning in patients receiving MV can be stratified by physiological responses after resupination.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.