Decision making of tracheostomy and extubation outcomes in mechanically ventilated patients evaluated by logistic regression and decision tree analyses
Background: Most studies determining the predictors of extubation outcomes in patients with mechanical ventilation have not included high-risk populations who avoid extubation and undergo tracheostomy.
Objective: To evaluate predictors of extubation intolerance by analyzing patients regardless of whether extubation was attempted or not.
Design: Retrospective cohort study.
Settings: Mixed intensive care unit (ICU) of Kumamoto University Hospital.
Patients and participants: Medical data of 288 consecutive mechanically ventilated adults were collected. Initial outcomes of endotracheal tube treatment were classified as 1) successful extubation, 2) extubation failure, and 3) tracheostomy without attempting extubation. Clinical variables responsible for those outcomes were determined by logistic regression and decision tree analyses. We defined combined outcome of extubation failure and tracheostomy as extubation intolerance in the present study.
Results: Of 288 patients, 17 failed extubation and 37 opted for tracheostomy without extubation. Logistic regression analysis revealed that the significant predictors of extubation failure were weak cough strength, poor consciousness, and excessive airway secretion. The propensity score of extubation failure calculated by logistic regression analysis in the tracheostomy group was as high as that of extubation failure group. A decision tree to predict the outcomes was described by branching with consciousness, style of ICU admission, and volume of airway secretion.
Conclusions: The principle predictors of extubation intolerance were related to instability of airway patency, and the decision making of tracheostomy was shown to be appropriate. These statistical methods could reduce the selection bias of study subjects.