Développement d’un modèle d’apprentissage automatique pour l’identification de patients susceptibles de répondre aux manœuvres de recrutement alvéolaire

Abstract

Acute hypoxemic respiratory failure (AHRF) is a common cause of admission to pediatric intensive care units. AHRF is part of several clinical syndromes, the most severe with the highest mortality rate, is the pediatric acute respiratory distress syndrome (ARDS). The most recent guidelines from the Pediatric Acute Lung Injury Consensus Conference (PALICC) in 2023 provide clinicians with recommendations on ventilation techniques, but these are not personalized. Pulmonary recruitment maneuvers are ventilation techniques used to optimize the number of alveoli participating in gas exchange and could be a promising approach for personalizing treatment in hypoxemic patients or those suffering from ARDS. The objective of this thesis was to develop a machine learning model trained on continuous temporal data to predict which patients are likely to improve their oxygenation following recruitment maneuvers. To achieve this goal, we validated a continuous oxygenation monitoring tool and performed data extraction and processing from the high-resolution pediatric intensive care unit database at Sainte-Justine Hospital. We then developed and optimized machine learning algorithms, including logistic regression, support vector machines, neural networks, and decision trees. The best-performing model was a random forest, demonstrating an area under the ROC curve (AUROC) of 0.78 (95% CI: 0.74 – 0.83), an accuracy of 85%, a sensitivity of 73%, and a specificity of 80%. This model highlights the significant potential of computational medicine techniques in helping to personalize respiratory treatments in intensive care. Further improvements, including the integration of bedside imaging data and external validation, are needed to enhance model performance before its clinical implementation. Acute hypoxemic respiratory failure (AHRF) is a common cause of admission to pediatric intensive care units. AHRF is part of several clinical syndromes, the most severe with the highest mortality rate, is the pediatric acute respiratory distress syndrome (ARDS). The most recent guidelines from the Pediatric Acute Lung Injury Consensus Conference (PALICC) in 2023 provide clinicians with recommendations on ventilation techniques, but these are not personalized. Pulmonary recruitment maneuvers are ventilation techniques used to optimize the number of alveoli participating in gas exchange and could be a promising approach for personalizing treatment in hypoxemic patients or those suffering from ARDS. The objective of this thesis was to develop a machine learning model trained on continuous temporal data to predict which patients are likely to improve their oxygenation following recruitment maneuvers. To achieve this goal, we validated a continuous oxygenation monitoring tool and performed data extraction and processing from the high-resolution pediatric intensive care unit database at Sainte-Justine Hospital. We then developed and optimized machine learning algorithms, including logistic regression, support vector machines, neural networks, and decision trees. The best-performing model was a random forest, demonstrating an area under the ROC curve (AUROC) of 0.78 (95% CI: 0.74 – 0.83), an accuracy of 85%, a sensitivity of 73%, and a specificity of 80%. This model highlights the significant potential of computational medicine techniques in helping to personalize respiratory treatments in intensive care. Further improvements, including the integration of bedside imaging data and external validation, are needed to enhance model performance before its clinical implementation.