FT-GAT: Graph neural network for predicting spontaneous breathing trial success in patients with mechanical ventilation.

Using Artificial Intelligence to Predict Mechanical Ventilation Weaning Success in Patients with Respiratory Failure, Including Those with Acute Respiratory Distress Syndrome

“…Once patients meet these clinical criteria, the decision as to the readiness for weaning is typically guided either by the physician’s clinical expertise and experience or a systematic assessment of respiratory weaning criteria based on established protocols [ 5 , 22 , 23 ]. Various respiratory parameters with diverse sensitivity and specificity, such as Airway Occlusion Pressure (P0.1), Maximal Inspiratory Pressure (MIP), the rapid shallow breathing index (RSBI), CROP (Dynamic Compliance, Respiratory Rate, Oxygenation, and MIP), respiratory rate, and vital capacity have all been proposed as predictors of weaning success or failure [ 5 , 7 ].…”

“…The inclusion of P0.1 in conjunction with the RSBI appears to enhance its specificity [ 7 ]. However, it is important to note that none of these parameters alone is deemed sufficient for an accurate prediction of successful weaning at the individual patient level [ 23 ].…”

“…Successful completion of the SBT, as evidenced by successful extubation, is achieved in some patients. However, approximately 25–40% of patients fail the first SBT, and 10–25% experience extubation failure, which necessitates re-intubation [ 9 , 22 , 23 ].…”

“…Delays in evaluating the readiness of the patient for weaning are a common cause of late weaning. As a consequence, patients with prolonged ventilation might experience airway trauma, dysphagia, delirium following extubation, drug dependencies, ventilator-associated pneumonia, diaphragm and muscle wasting, other forms of increased morbidity, and even higher mortality rates [ 2 , 23 ]. On the other hand, a withdrawal of MV that is too rapid may trigger respiratory collapse, impeding the patient’s recovery and exposing them to the associated risks of ventilator-associated pneumonia or other injuries to the lungs induced by the ventilator [ 3 ].…”

“…Kuo [ 52 ] developed an ANN for predicting MV weaning based on data from 121 ICU patients that has an AUROC of 0.83 and that was found to be superior to traditional weaning assessment tools such as SBT. Another NN model, designed by Kim [ 23 ], used a novel DL model called FT-GAT in order to predict a successful SBT and, eventually, extubation. The AUROC of this model was 0.8, with a similar AUROC being found upon temporal validation.…”