Delayed Invasive Mechanical Ventilation and ICU Outcomes in ARDS-Risk Acute Hypoxemic Respiratory Failure: A MIMIC-IV Propensity-Weighted and Predictive Modeling Study

<title>Abstract</title> <p>Background Noninvasive respiratory support may avert intubation in selected patients with acute hypoxemic respiratory failure, but failure to recognize deterioration may postpone invasive mechanical ventilation and worsen outcome. Evidence on clinically relevant delay thresholds remains inconsistent. Objective To evaluate the association between delayed post-ICU initiation of invasive mechanical ventilation and mortality in an ARDS-risk acute hypoxemic respiratory failure cohort, and to characterize 28-day mortality prediction using conventional and machine-learning models. Methods This retrospective MIMIC-IV-derived cohort included 13,619 ICU admissions identified within an ARDS-risk disease context. Ventilation timing was classified as no invasive mechanical ventilation, pre-ICU ventilation, 0–6 hours, 6–12 hours, 12–24 hours, or more than 24 hours after ICU admission. The principal comparison was early (0–6 hours) versus delayed (&gt; 6 hours) ventilation among admissions intubated after ICU entry. Multivariable logistic regression, stabilized inverse-probability-of-treatment weighting with 1st/99th percentile truncation, and doubly adjusted models evaluated mortality. Logistic regression, random forest, and XGBoost were assessed on a stratified 70/30 split for 28-day mortality. Results The cohort comprised 5,006 admissions without invasive ventilation, 971 with pre-ICU ventilation, 6,528 with post-ICU ventilation at 0–6 hours, and 1,114 with delayed ventilation. ICU mortality was 22.5% in the 0-6-hour group and 27.4%, 25.4%, and 28.9% in the 6-12-hour, 12-24-hour, and &gt; 24-hour groups, respectively. In the IPTW/doubly adjusted analysis, delayed ventilation was associated with ICU mortality (odds ratio [OR] 1.34, 95% CI 1.14–1.59; P &lt; 0.001) and in-hospital mortality (OR 1.20, 95% CI 1.03–1.41; P = 0.020), whereas associations with 28-day mortality (OR 1.11, 95% CI 0.95–1.29) and 90-day mortality (OR 1.12, 95% CI 0.96–1.29) were not statistically significant. XGBoost showed the best test-set performance (AUC 0.748, average precision 0.566, Brier score 0.175), although sensitivity at the selected threshold was low (0.267). Conclusions Among ICU admissions that ultimately received post-ICU invasive ventilation, initiation more than 6 hours after ICU entry was associated with higher ICU and in-hospital mortality after measured-confounder adjustment. The attenuated 28- and 90-day estimates, residual imbalance, time-dependent exposure definition, and moderate predictive performance preclude causal or bedside deployment claims. Prospective, time-updated validation is needed.</p>