Resident Physician New York Medical College/St Mary and St Clare Program, New Jersey
Disclosure information not submitted.
Introduction: The ROX index, (SpO₂/FiO₂)/RR, is a noninvasive tool to monitor patients with acute hypoxemic respiratory failure (AHRF), particularly those on noninvasive ventilation. While static ROX has been evaluated for predicting intubation, the use of dynamic ROX (ΔROX) to guide steroid therapy or identify clinical deterioration remains underexplored.
Methods: We performed a simulation-based machine learning study using data from 300 virtual patients with AHRF. Patients with ADHF, ESRD, chronic hypercapnia, tracheostomy, or DNI status were excluded. Simulated data included demographics, comorbidities & ROX index values at T0, 6h, 12h, 24h, and 48h. ΔROX was calculated at each time point to reflect temporal changes.
Patients were categorized into steroid escalation, de-escalation, or maintenance groups. The primary outcome was steroid therapy decision & the secondary outcome was the need for intubation. Random forest models were trained using ΔROX & clinical variables. Performance was evaluated using train-test splits with AUC, accuracy, precision, recall, F1-score & feature importance.
To validate findings clinically, we designed a retrospective study of inpatients with AHRF (2022–25). Inclusion criteria: age ≥18, systemic steroids initiated within 24h of admission, and PaO₂/FiO₂ < 300 within 6h of deterioration. Exclusion criteria: comfort care, hypercapnic respiratory failure, heart failure-related pulmonary edema, or non-respiratory steroid use. ΔROX will be calculated at T0, 6h, 12h, 24h, and 48h using charted SpO₂, FiO₂, and RR. Planned analyses include repeated-measures ANOVA, logistic regression & ROC curve analysis.
Results: In the simulated cohort, the ΔROX-based model predicted steroid management with an AUC of 0.99 and an accuracy of 96.7%. For intubation prediction, the model achieved an AUC of 0.91 & an accuracy of 82.2%. Patients with improving ΔROX were more likely to undergo steroid de-escalation, had lower intubation rates & shorter stays.
Conclusions: ΔROX are strongly associated with both steroid decisions & intubation risk in AHRF. This simulation-based model suggests ΔROX may serve as a valuable tool for clinical decision-making. A real-world EMR-based validation study is in progress to confirm these findings and explore integration of ΔROX-based alerts into clinical workflows.
