Predicting extubation failure after successful completion of a spontaneous breathing trial.

OBJECTIVE To derive a clinical prediction rule that uses bedside clinical variables to predict extubation failure (reintubation within 48 h) after a successful spontaneous breathing trial. METHODS This prospective observational cohort study was performed at the Northwestern Memorial Hospital in Chicago, Illinois, which is a large tertiary-care university hospital. Among 673 consecutive patients who received mechanical ventilation during a 15-month period, 122 were ventilated for at least 2 days and did not undergo withdrawal of support or tracheostomy. These patients were followed after extubation to identify those who were reintubated within 48 h (extubation failure). We used logistic regression analysis to identify variables that predict reintubation, and we used bootstrap resampling to internally validate the predictors and adjust for overoptimism. RESULTS Sixteen (13%) of the 122 patients required reintubation within 48 h. Three clinical variables predicted reintubation: moderate to copious endotracheal secretions (p = 0.001), Glasgow Coma Scale score < or =10 (p = 0.004), and hypercapnia (P(aCO(2)) > or = 44 mm Hg) during the spontaneous breathing trial (p = 0.001). Using logistic regression and bootstrap resampling to adjust for overfitting, we derived a clinical prediction rule that combined those 3 clinical variables (area under the receiver operating characteristic curve 0.87, 95% confidence interval 0.74-0.94). CONCLUSIONS With our clinical prediction rule that incorporates an assessment of mental status, endotracheal secretions, and pre-extubation P(aCO(2)), clinicians can predict who will fail extubation despite a successful spontaneous breathing trial.