Predictive performance of ROX index and its variations for NIV failure.

Abstract

Objective: To determine whether the ROX index and its variations can predict the risk of intubation in ICU patients receiving NIV ventilation using large public ICU databases.

Design: Retrospective observational cohort study.

Setting: Patient data was extracted from both the AmsterdamUMCdb and the MIMIC-IV ICU databases, which contained data related to 20,109 and 50,920 unique patients.

Patients: Non-invasively mechanically ventilated.

Interventions: Retrospective review of variables.

Main variables of interest: To assess the predictive values of models for each index, the ROX and its variations mROX, ROX-HR and mROX-HR were calculated based on mean values of SpO₂, respiratory rate, FiO₂ and PaO₂ from 2-h windows within the first 12 h of NIV.

Results: 3344 patients were eligible for analysis of which 1344 were intubated, died or returned to NIV within 24 h of ending NIV. NIV failure group had higher SOFA scores and higher CRP levels at admission. There was no difference in duration of NIV therapy or 28-day mortality, but patients who failed NIV had longer length of stay. The best performing index was ROX with an AUROC of 0.626 at 10-12 h. All other indices for all other time windows were less discriminating.

Conclusions: The performance of ROX index and its variations to predict NIV failure in ICU patients across large public ICU databases was moderate at best and cannot currently be recommended for clinical decision support.