Early deep-to-light sedation versus a sustained light-or-awake strategy in mechanically ventilated adults: a target trial emulation

Abstract Observational studies comparing sedation depth strategies in mechanically ventilated patients have relied on observed consciousness levels alone to define exposure, potentially conflating intentional drug-induced sedation with low arousal caused by residual medications, other agents, or underlying illness. We applied the target trial emulation framework with the clone-censor-weight method to compare early deep-to-light sedation (DTLS) versus a sustained light-or-awake strategy, using a joint exposure definition that requires both Richmond Agitation-Sedation Scale (RASS) depth and qualifying continuous sedative infusion. Using MIMIC-IV (2008–2022), we identified 5,086 adults receiving invasive mechanical ventilation for ≥ 48 h. Each patient was cloned to both strategies. During the primary 144-hour adherence window, inverse probability weights accounted for measured predictors of artificial censoring. The primary outcome was 28-day ventilator-free days (VFDs); successful extubation was analysed with death as a competing event. VFD outcomes were observed before artificial censoring in 1,481 sustained-light-or-awake and 622 DTLS clones. Weighted mean VFDs were 17.72 days for DTLS versus 17.76 days for sustained light-or-awake, a difference of − 0.04 days (95% CI, − 1.32–1.29). The cause-specific extubation HR was 0.97 (95% CI, 0.84–1.11). A sensitivity cohort without future minimum ventilation or ICU-duration requirements ( n = 15,842) showed fewer VFDs with DTLS (difference, − 1.03 days; 95% CI, − 1.47 to − 0.61). These findings do not establish a DTLS benefit; duration-based selection, residual confounding, and limited late-window support constrain interpretation.