Objective: To examine the association between the performance of a tracheostomy and intensive care unit and postintensive care unit mortality, controlling for treatment selection bias and confounding variables.
Design: Prospective, observational, cohort study.
Setting: Twelve French medical or surgical intensive care units.
Patients: Unselected patients requiring mechanical ventilation for > or =48 hrs enrolled between 1997 and 2004.
Interventions: None.
Measurements and main results: Two models of propensity scores for tracheostomy were built using multivariate logistic regression. After matching on these propensity scores, the association of tracheostomy with outcomes was assessed using multivariate conditional logistic regression. Results obtained with the two models were compared. Of the 2,186 patients included, 177 (8.1%) received a tracheostomy. Both models led to similar results. Tracheostomy did not improve intensive care unit survival (model 1: odds ratio, 0.94; 95% confidence interval, 0.63-1.39; p = .74; model 2: odds ratio, 1.12; 95% confidence interval, 0.75-1.67; p = .59). There was no difference whether tracheostomy was performed early (within 7 days of ventilation) or late (after 7 days of ventilation). In patients discharged free from mechanical ventilation, tracheostomy was associated with increased postintensive care unit mortality when the tracheostomy tube was left in place (model 1: odds ratio, 3.73; 95% confidence interval, 1.41-9.83; p = .008; model 2: odds ratio, 4.63; 95% confidence interval, 1.68-12.72, p = .003).
Conclusions: Tracheostomy does not seem to reduce intensive care unit mortality when performed in unselected patients but may represent a burden after intensive care unit discharge.