UNPLANNED EXTUBATION - CLINICAL PREDICTORS FOR REINTUBATION

Study objective: To examine the incidence, clinical impact, and predic tors of reintubation following unplanned extubation (UE) in ICU patien ts. Design: Retrospective data collection of prospectively identified consecutive cases of UE. Setting: The adult Medical, Coronary, Surgica l, Cardiac Surgery, and Neuroscience ICUs of a University Medical Cent er. Patients: Eighty-one episodes of UE in 72 adolescent or adult (53/-19 years) ICU patients. Measurements and results: In 39 (48 percent) of 81 cases, reintubation was performed within 24 h of UE, 33 (85 per cent) within the first hour, and 31 (79 percent) as a result of respir atory distress. There were five documented complications of UE and/or reintubation but no deaths. Clinical predictors for reintubation were sought from routinely available demographic, clinical, laboratory, and respiratory factors which had been documented in the medical records for the 24 h period prior to UE. Using univariate analysis followed by stepwise logistic regression in the first 56 cases (model set), the f ollowing factors were identified as being associated with reintubation : (1) volume controlled ventilation (synchronous intermittent mandator y ventilation or assist-control ventilation) with rate more than 6/min ; (2) most recent arterial pH level prior to UE being 7.45 or more; (3 ) most recent ratio of PaO2 to fraction of inspired oxygen prior to UE less than 250 mm Hg; (4) highest heart rate in the 24 h prior to UE g reater than 120 beats per minute; (5) presence of 3 or more coexisting medical disorders (of 7 possible); (6) mental status other than alert ; and (7) indication for intubation other than ''preoperative.'' In th e model set, the presence of 4 or more of these 7 factors correctly pr edicted reintubation in 23 of 25 (92 percent positive predictive value ) and presence of 3 or fewer factors correctly predicted no reintubati on in 26 of 31 cases (83 percent negative predictive value), with outc ome of 88 percent of cases being correctly predicted. The model was te sted in the next 24 cases (validation set) in which 18 (75 percent) we re correctly predicted.