mize the conditions of extubation, eg, steroid pretreatment, optimization of fluid balance, glucose control, and nutrition support. These patients could also be targeted for interventions such as aggressive bronchial hygiene protocols, early mobilization, and careful titration of analgesics. The prediction of early versus late reintubation includes different variables, suggesting that different characteristics contribute to failure and potentially via different pathways. Our prediction models suggest that residual hypoxemia on the prediction day, likely representative of underlying pulmonary disease or other comorbidities, can predispose patients to the risk of failed extubation also in a nonimminent period. In terms of accuracy of prediction, our models of early and late failure performed well. The fact that only 70% of the events were correctly classified indicates that there is unexplained residual variability. This unexplained variability could represent inherent variability, unmeasured variables, or new events not necessarily directly related to the reasons leading to the first intubation. To minimize the latter sources, we excluded reintubations performed in conjunction with surgical procedures. This study1 represents an initial step in identifying areas of opportunity to improve and minimize the occurrence of re-intubation, an event that is associated with prolonged ventilation, longer ICU stay, and increased hospital costs.2 We are currently working on a prospective cohort to better define the time course of failure. Analysis of time to failure will have to account for competing risk events such as censoring by tracheostomy or death.