•  
  •  
 

Abstract

Purpose: This study evaluated whether temporal trajectories of physiologic and ventilator parameters in the 24 hours preceding extubation provide independent prognostic information beyond snapshot values for predicting extubation failure.


Materials and Methods: We conducted a retrospective cohort study using the MIMIC-IV database. Adults (≥ 18 years) who received at least 24 hours of invasive mechanical ventilation followed by a planned extubation were included. The primary outcome was extubation failure, defined as reintubation within 72 hours. Temporal trajectory features (slopes of linear regression over the 24-hour pre-extubation window) were derived from hourly vital signs and ventilator data in the 24-hour pre-extubation window. A snapshot-only model (last values only) was compared with a trajectory-enriched model (snapshot + trajectory features) using likelihood ratio testing, NRI and incremental AUROC analysis. Multivariable logistic regression with bootstrap internal validation (B = 500) was used.


Results: The final cohort comprised 15,778 patients; extubation failure occurred in 3,220 (20.4%). The trajectory-enriched model demonstrated an optimism-corrected AUROC of 0.745 (95% CI 0.732-0.749) vs. 0.743 (95% CI 0.731-0.746) for the snapshot-only model. The incremental AUROC of trajectory features was +0.002 (p < 0.001 by LR test). NRI was +1.4% (95% CI 0.2%-3.1%). Independent predictors included total ventilation duration (OR 1.005 per hour, 95% CI 1.005-1.006, p < 0.001), PEEP at extubation (OR 1.043 per cmH₂O, 95% CI 1.028-1.059, p < 0.001), vasopressor use in the preceding 24 hours (OR 1.340, 95% CI 1.219-1.473, p < 0.001), and SpO₂/FiO₂ slope (OR 0.971 per unit/h worsening, 95% CI 0.960-0.982, p < 0.001). RSBI and PEEP slope did not retain independent significance.


Conclusion: Temporal trajectories carry statistically significant but clinically modest independent prognostic information. SpO₂/FiO₂ slope is the only trajectory feature that retains significance after multivariable adjustment, contributing only marginally beyond snapshot values. Pre-extubation physiology alone is insufficient for accurate bedside risk stratification.

DOI

10.55729/2000-9666.1671

Share

COinS