Improvement in the prediction of ventilator weaning outcomes by an artificial neural network in a medical ICU

Abstract

BACKGROUND: Twenty-five to 40% of patients pass a spontaneous breathing trial (SBT) but fail to wean from mechanical ventilation. There is no single appropriate and convenient predictor or method that can help clinicians to accurately predict weaning outcomes. This study designed an artificial neural network (ANN) model for predicting successful extubation in mechanically ventilated patients. METHODS: Ready-to-wean subjects (N 121) hospitalized in medical ICUs were recruited and randomly divided into training (n 76) and test (n 45) sets. Eight variables consisting of age, reasons for intubation, duration of mechanical ventilation, Acute Physiology and Chronic Health Evaluation II score, mean inspiratory and expiratory times, mean breathing frequency, and mean expiratory tidal volume in a 30-min SBT (pressure support ventilation of 5 cm H2O and PEEP of 5 cm H2O) were selected as the ANN input variables. The prediction performance of the ANN model was compared with the rapid shallow breathing index (RSBI), maximum inspiratory pressure, RSBI at 1 min (RSBI1) and 30 min (RSBI30) in an SBT, and absolute percentage change in RSBI from 1 to 30 min in an SBT (RSBI30) using a confusion matrix and receiver operating characteristic curves. RESULTS: The area under the receiver operating characteristic curves in the test set of the ANN model was 0.83 (95% CI 0.69 - 0.92, P 0.5 selected from the training set. CONCLUSIONS: The ANN model improved the accuracy of predicting successful extubation. By applying it clinically, clinicians can select the earliest appropriate weaning time.