Background: Tracheostomy is common in intensive care unit patients, but the appropriate timing is controversial.
Objective: To determine whether earlier tracheostomy is associated with greater long-term survival.
Design: Retrospective cohort analysis.
Setting: Acute care hospitals in Ontario, Canada (n = 114).
Patients: All mechanically ventilated intensive care unit patients who received tracheostomy between April 1, 1992 and March 31, 2004, excluding extreme cases (< 2 or > or = 28 days) and children (< 18 yrs).
Measurements: For crude analyses, tracheostomy timing was classified as early (< or = 10 days) vs. late (> 10 days) with mortality measured at multiple follow-up intervals. Proportional hazards analyses considered tracheostomy as a time-dependent variable to adjust for measurable confounders and possible survivor treatment bias. We used stratification, propensity score, and instrumental variable analyses to adjust for patient differences.
Results: A total of 10,927 patients received tracheostomy during the study, of which one-third (n = 3758) received early and two-thirds late (n = 7169). Patients receiving early tracheostomy had lower unadjusted 90-day (34.8% vs. 36.9%; p = 0.032), 1 yr (46.5% vs. 49.8%; p = 0.001), and study mortality (63.9% vs. 67.2%; p < 0.001) than patients receiving late tracheostomy. Multivariable analyses treating tracheostomy as a time-dependent variable showed that each additional delay of 1 day was associated with increased mortality (hazard ratio 1.008, 95% confidence interval 1.004-1.012), equivalent to an increase in 90-day mortality from 36.2% to 37.6% per week of delay (relative risk increase 3.9%; number needed to treat, 71 patients to save one life per week delay).
Limitations: This analysis provides guidance regarding timing but not patient selection for tracheostomy.
Conclusions: Physicians performing early tracheostomy should not anticipate a large potential survival benefit. Future research should concentrate on identifying which patients will receive the most benefit.