Noninvasive estimation of oxygenation index in pediatric critical care: an independent retrospective observational validation

Objective To independently validate an empirically optimized algorithm for calculating estimated Oxygenation Index (eOI) using noninvasive parameters from pediatric intensive care populations. Design Retrospective observational cohort study using an integrated patient data repository spanning over 12 years (August 2012-December 2024). Setting Single tertiary children's hospital with general pediatric ICU (PICU) and cardiothoracic ICU (CTICU). Key measures Arterial blood gas measurements were paired with coincident SpO 2 , heart rate, pulse rate, FiO 2 , and mean airway pressure measurements. The primary analyses used SpO 2 observations between 80%–100%. Using these values eOI was calculated. The primary outcome was the Bias and Limits of Agreement of the difference between measured OI and eOI. Discrimination performance of eOI for severity of hypoxemia was evaluated using receiver operating characteristic curves at OI thresholds of 4, 8, and 16. Results Analysis included 68,915 observations from 7,109 subjects (44,133 CTICU, 24,782 PICU observations). Bias was minimal in both populations: PICU 0.06 (95% CI; 0.03, 0.10) and CTICU 0.12 (95% CI; 0.09, 0.14). Limits of agreement were −5.2 to 5.4 (PICU) and −4.9 to 5.2 (CTICU). Discrimination performance was excellent, at 3 hypoxemia thresholds (AUROC; 0.91–0.98), and in the CTICU for OI ≥4 when SpO 2 >97% (AUROC; 0.83). Conclusions The new eOI algorithm provides accurate, but not precise, estimation of OI in both general pediatric and cardiothoracic ICU populations. Noninvasive OI monitoring may be shown clinically useful.