EquiOx: A Prospective study of pulse oximeter bias and skin pigmentation in critically-ill adults

Rationale: Pulse oximeter performance may vary by skin pigmentation, but most data are retrospective with key limitations. Objective: To quantify pulse oximeter bias [mean difference between pulse oximeter oxygen saturation (SpO2) and arterial blood functional oxygen saturation (SaO2)], and average root mean square error (ARMS) and estimate adjusted effects of skin pigment on bias or ARMS in critically-ill adults Methods: Prospective single–center study of 631 ICU patients (2022–2024) directly observed SpO2 and SaO2 pairs. Skin pigment was assessed using the subjective Monk Skin Tone Scale and objective spectrophotometry measurement (Individual Typology Angle [ITA]). Adjusted effects were estimated with targeted maximum likelihood estimation. Main Results: Among 1,760 paired measurements from 631 critically–ill adults, median SaO2 was 98% (IQR 96%, 99%) with 40 episodes of stable hypoxemia (SpO2<90%). SpO2 systematically underestimated SaO2 [median bias= −1.70 IQR (−2.84, −0.50)]. Bias was less negative in patients with darker skin (ITA <-30°) [−1.05 (−2.44, −0.10)] vs lighter skin (ITA >30°) [−2.01 (−3.34, −1.00)], and remained significantly different in adjusted analyses. ARMS was 3.87 (95% CI 3.25, 4.53) overall and 4.49 (95% CI 2.63, 7.07) in patients with darker skin. Ear probes performed worse than finger probes [bias: −2.20 vs. −1.60; ARMS: 4.90 vs. 2.70]. Conclusions: In this large ICU cohort, hypoxemia was rare, pulse oximeters systematically underestimated SaO2, performance varied by skin pigment and probe site. Bias was less negative in patients with darker skin. Pulse oximeter inaccuracies in ICU patients may be more substantial than clinicians recognize.