Landmark-specific associations between tracheostomy during prespecified allocation windows and 90-day mortality in mechanically ventilated adults: a multicenter observational study

Evidence on tracheostomy timing during prolonged invasive mechanical ventilation remains inconclusive. We conducted a multicenter retrospective cohort study using electronic health records and administrative claims from four tertiary hospitals. Sequentially nested landmark cohorts were defined at Days 7, 10, and 14 after intubation among patients alive, mechanically ventilated, and without prior tracheostomy. At each landmark, tracheostomy during a prespecified 48-hour window was compared with no tracheostomy during that window using stabilized inverse probability weighting and Cox models. Among 6710 adults, the cohorts included 3196, 1872, and 808 patients; 534, 375, and 203 underwent tracheostomy during the window. Adjusted hazard ratios were 0.99 (95% CI 0.80–1.23), 0.94 (0.75–1.18), and 0.61 (0.44–0.84), respectively; risk differences were + 1.4 (95% CI − 4.3 to + 6.7), + 2.1 (− 5.2 to + 8.4), and − 13.3 (− 22.0 to − 4.6) percentage points. The 72-h Day 14 analysis was similar (hazard ratio, 0.64; 95% CI 0.48–0.86; risk difference, − 11.2% points; 95% CI − 19.7 to − 3.8). Cross-landmark heterogeneity was observed in the 48-hour analysis ( P = 0.039) but not in the 72-hour analysis ( P = 0.091). These conditional associations neither identify an optimal timing nor establish a causal mortality benefit; the Day 14 finding remains hypothesis-generating.