A26-05 Differential Severity Classification and Prognostic Performance of PaO2/FIO2 Versus Spo2/fIO2 Ratios in Adults With Acute Hypoxemic Respiratory Failure

Rationale Grading severity of hypoxemia for patients with respiratory failure using SpO2/FIO2 carries promise as a continuously collected non-invasive measure. Recently SpO2/FIO has been incorporated into the new global Acute Respiratory Distress Syndrome (ARDS) criteria and Sequential Organ Failure Assessment 2 (SOFA-2) scoring. However, severity classification and prognostic performance for patients defined by SpO2/FIO2 versus PaO2/ FIO2 is unknown. Methods We conducted a retrospective study across 9 Common Longitudinal ICU data Format (CLIF) Consortium health systems representing 43 hospitals. We included adults admitted through the ED between 2021-2024 and who received >24 hours of invasive or non-invasive positive pressure ventilation (IMV or NIPPV) and/or high-flow nasal oxygen (HFNO) within 7 days of admission. We evaluated the prognostic performance of a patient’s lowest PaO2/FIO2, if available, and SpO2/FIO2 based on an SpO2<97% if not, in the first 24-hours of respiratory support. We compared the primary outcome of in-hospital death or hospice discharge by day 28 across severity categories using a pooled chi-squared test. We used multi-level aggregate logistic regression with PaO2/FIO2 or SpO2/FIO2 modeled continuously using restricted cubic splines, including hospital/site as a random effect. We compared model performance between PaO2/FIO2 or SpO2/FIO2 overall and in analyses stratified by Black vs. Not-Black race in light of previously demonstrated biased performance of pulse oximeters. Results The cohort included N = 43,440 patients with mean age of 62 years, 57% male and 62% at academic centers. 31,125 (72%) received IMV, 5,050 (12%) NIPPV, and 7265 (17%) HFNO, with 29,095 (67%) with PaO2/FIO2 and 12,383 (29%) SpO2/FIO2 defined severity. Compared with PaO2/FIO2, SpO2/FIO2 classification was associated with lower 28-day death or hospice discharge across all severity levels (Figure 1, p < 0.001). Model area under the curve’s (AUC) were similar in overall models and indicated poor discrimination as a single measure (AUC-PaO2/FIO2=0.51 [95%-CI: 0.49-0.54]: AUC-SpO2/FIO2=0.52 [95%-CI: 0.49-0.56], comparison p-value=0.68); Brier scores for these models were 0.21 and 0.16 respectively. In stratified analyses, the PaO2/FIO2 and SpO2/FIO2 models both performed better in Black versus non-Black patients (AUC P/F-Black=0.55 [95%-CI: 0.54-0.56] vs AUC P/F-NotBlack=0.54 [95%-CI: 0.53-0.55], p-value=0.04; AUC-S/F-Black=0.62 [95%-CI: 0.60-0.65) vs. AUC-S/F-NotBlack 0.57 [95%-CI: 0.56-0.59], p-value=0.001). Conclusions In this large multi-center cohort, we show that oxygenation severity defined by SpO2/FIO2 cutoffs identifies a cohort with lower risk of mortality and may have differential prognostic performance according to race. Further research is needed to refine SpO2/FIO2 cutoffs for exchangeable use compared to PaO2/FIO2. This abstract is funded by: NHLBI-K23HL169743