Beyond the pulse: Racial disparities in pulse oximetry readings and the influence on intensive care unit readmission rates

Racial bias in pulse oximetry—a critical noninvasive diagnostic tool that estimates blood oxygen levels using light absorption—has been shown to systematically overestimate oxygen saturation in Black patients, potentially delaying the detection and treatment of hypoxemia. While prior work has focused narrowly on mortality within selected clinical populations or on immediate testing and treatment disparities in emergency department settings, this study investigates how pulse oximetry bias affects a broader and more representative intensive care unit (ICU) cohort, emphasizing highly consequential operational outcomes: supplemental oxygen delivery, within-hospitalization ICU readmission, and remaining hospital length of stay. Using patient-level data from the MIMIC-IV database, we apply a moderated mediation framework to trace the causal pathway from race to treatment to operational failure. We find that under peripheral oxygen saturation-only monitoring, Black patients are 5.2 percentage points less likely than White patients to receive supplemental oxygen. This undertreatment significantly increases the probability of an unplanned ICU readmission, with oxygen delivery reducing readmission risk by 4.8 percentage points, a 37% relative reduction. Furthermore, a separate operational load analysis using ordinary least squares regression indicates that such ICU bounce-backs are associated with an 86% increase in remaining hospital length of stay, pointing to a substantial downstream capacity penalty. To evaluate the mitigating effect of more accurate diagnostic information, we leverage arterial oxygen saturation (SaO 2 ) measurements as a quasi-intervention. We find that the racial disparity in oxygen therapy is substantially attenuated and becomes statistically indistinguishable from zero when SaO 2 readings are available, consistent with mitigation of the bias-treatment-outcome cascade. Our findings reveal how diagnostic inaccuracy is an upstream process defect that propagates through clinical decision-making, reinforcing structural disparities and degrading system performance. We contribute to healthcare operations management by modeling technological bias as a driver of costly clinical rework and demonstrating that targeted process changes—such as confirmatory SaO 2 testing—can improve healthcare equity and hospital efficiency.