Pulse Oximetry Discrepancies and Occult Hypoxemia in ICU Patients: Predictors and Clinical Outcomes

BackgroundPulse oximeters sometimes fail to accurately reflect arterial oxygen saturation (SaO₂), particularly in darker-skinned patients resulting in undiagnosed hypoxemia, potentially delaying recognition and appropriate interventions.Research QuestionWe aimed to evaluate the prevalence and predictors of SpO₂-SaO₂ discrepancies, particularly occult hypoxemia, and to assess their association with clinical outcomes in ICU patients.Study Design and MethodsWe conducted a retrospective cohort analysis using the Blood-gas and Oximetry Linked Dataset (BOLD), analyzing critically ill patients from the eICU-CRD database (2014-2015). Patients with paired SpO₂-SaO₂ measurements within five minutes were included. We identified SpO₂-SaO₂ discrepancies as a difference of >2.99% and defined occult hypoxemia as an arterial partial pressure of oxygen (PaO₂) < 60 mm Hg or SaO₂ < 89% with an SpO₂ > 88%. The primary outcomes included ICU length of stay (LOS), Sequential Organ Failure Assessment (SOFA) score, and in-hospital mortality.ResultsAmong 36,280 ICU patients, 23.6% had SpO₂-SaO₂ discrepancies, and 4.7% had occult hypoxemia. Black patients were overrepresented in both groups, with an adjusted odds ratio (aOR) of 1.35 (95% CI: 1.25-1.47) for discrepancy and 1.22 (95% CI: 1.04-1.47) for occult hypoxemia. Higher BMI, lower pH, elevated creatinine, and higher Charlson Comorbidity Index scores were also significant predictors. Patients with discrepancies had worse clinical outcomes, including increased SOFA scores in the following 24 h (β = 0.31; p < .0001) and higher in-hospital mortality (aOR 1.15; p < .0001). Occult hypoxemia was associated with even worse outcomes, including a longer ICU LOS (IRR 1.12; p < .0001) and significantly increased mortality (aOR 1.73; p < .0001).InterpretationOne in four critically ill patient in our cohort experienced SpO₂-SaO₂ discrepancy which is associated with adverse clinical outcomes. Black race, obesity, and higher comorbidity burden were significant predictors of these discrepancies. Our findings emphasize the need for more rigorous clinician oversight in the use of this technology.