Between-hospital variation in liberation from invasive mechanical ventilation: Consequences of ignoring death as a...

Duration of invasive mechanical ventilation is used to benchmark intensive care units, but published metrics model duration directly and do not treat death as a competing event, so a patient who dies on day two counts as a short ventilation. We rebuilt the metric on a competing-risk footing and quantified the consequences.

Retrospective cohort of 43,880 adults starting invasive ventilation at 114 US hospitals in the eICU Collaborative Research Database. The estimand was the hospital-specific, risk- and reliability-adjusted cumulative incidence of liberation by day 7, with death as a competing event. Hospitals were profiled using overdispersion-adjusted funnel plots and compared with a conventional observed-to-expected ventilation-duration metric at matched control limits.

By day 7, 72.9% of patients were liberated, 11.4% had died, and 15.7% remained ventilated. Risk-standardised 7-day liberation ranged from 26.7% to 85.8% (median 70.7%). Case mix explained 3.7% of between-hospital variance, hospital structure and measurable processes a further 31.6% combined. The two metrics ranked hospitals only moderately alike (Spearman rho 0.79) and agreed on no outlier at matched limits (weighted kappa 0.00). The conventional metric called 26 hospitals better than expected and 11 worse; the competing-risk metric reversed that (3 and 26). Sixteen hospitals (14%) with significantly worse liberation went unflagged, and had higher risk-standardised ICU mortality (18.2% versus 13.6%, p < 0.001).

Benchmarking ventilation duration without treating death as a competing event misses the units that are failing to liberate patients, and those units have higher mortality.