A26-06 Granular Longitudinal Data Enhance the Prognostic Validity of SpO2:fiO2 Ratio in a Modern Cohort of Patients With Acute Hypoxic Respiratory Failure Requiring Invasive Mechanical Ventilation

Rationale Oxygenation is a crucial domain of respiratory physiology. Advances in clinical engineering allow for frequent, automated recording of the peripheral oxygen saturation (SpO2) to fraction of inspired oxygen (FiO2) ratios (SFR). However, the impact of these longitudinal data on the prognostic validity of SFR is unknown, and performance compared to the partial pressure of oxygen (PaO2) to FiO2 ratio (PFR) is unclear. We aimed to assess the impact of absolute SFR and SFR trajectory on 30-day mortality during the first 24 hours of mechanical ventilation (MV) in a large cohort of acute hypoxic respiratory failure (AHRF) containing granular electronic health record (EHR) data. Methods We identified encounters involving new initiation of MV and hypoxia (minimum SFR < 315) across our six-center health system between 2020-2024. We calculated SFR using SpO2 measurements < 97. Oxygenation exposures were operationalized as follows: median SFR/24 hours of MV, median PFR/24 hours of MV, median SFR at baseline (0-6 hours), median SFR over 3 time points (6-12, 12-18, and 18-24 hours), change in SFR between baseline and 18-24 hours, and SFR trajectory. We selected four trajectories a priori: major improvement, minor improvement, stable, and worsening. We assessed the relationship between exposures and 30-day mortality using logistic regression adjusted for age and sex. We assessed model performance using area under the receiver operating characteristic (AUROC) and likelihood ratio (LR). Results Of 12,638 encounters, a median number of 38 SFR measurements were recorded (IQR 32,47), compared with 1 PFR measurement (IQR 0,5). Median SFR/24 hours was predictive of 30-day mortality (p < 0.001) and change in SFR added prognostic information independent of baseline SFR (p for LR < 0.001). Over 90% of the cohort had complete data for trajectory modeling; 18% were classified as major improvement, 33% as minor improvement, 30% as stable, and 8% as worsening. Trajectory membership was predictive of 30-day mortality even after adjustment for baseline SFR (p < 0.001, standardized risk 0.24, 0.29, 0.36, and 0.45 respectively) (Figure 1). In the subgroup with available PFR (n = 7,319), SFR had comparable performance to PFR (AUROC 0.68 vs. 0.64). Conclusion Longitudinal data from a modern EHR enhance the prognostic validity of SFR and exhibit superior discrimination for mortality when compared to PFR. Absolute SFR and early SFR trajectory may inform prognostication in clinical practice, allow for noninvasive calculation of risk scores in critical illness, and facilitate enrichment strategies for clinical trials of AHRF therapeutics. This abstract is funded by: NIH HL155159 (Reilly), HL161196 (Meyer)