Pulse Oximeter Bias: A Billion-Dollar Blind Spot in Cardiac AI

The promise of artificial intelligence in healthcare is often painted with broad strokes of efficiency and improved outcomes. Yet, a critical examination of documented failures reveals a more nuanced, and at times alarming, reality. One such failure, deeply rooted in historical oversight but amplified by the increasing reliance on algorithmic decision-making, is the persistent racial bias in pulse oximetry, a technology whose flaws have been known for decades. This isn’t merely an academic concern; it directly impacts patient safety, leading to potentially dangerous misdiagnoses and undertreatment, particularly for individuals with darker skin tones.

The Enduring Problem of Pulse Oximeter Bias

For over thirty years, the medical community has been aware that pulse oximeters, ubiquitous devices used to measure blood oxygen saturation, exhibit significant racial bias. This bias manifests as an overestimation of oxygen levels in individuals with darker skin pigmentation, leading to what is often termed “occult hypoxemia”, dangerously low oxygen levels that go undetected by the device. The clinical implications are profound: delayed interventions, inappropriate treatment escalation, and, in severe cases, preventable morbidity and mortality. Research has consistently highlighted this disparity. Studies, including those published in the New England Journal of Medicine (NEJM) and referenced by Johns Hopkins University, have shown that pulse oximeters can have a 50-860% greater error rate for individuals with dark skin compared to those with lighter skin tones [DP04]. This is not a new revelation; the underlying physics of how pulse oximeters use light absorption to estimate oxygen saturation are inherently sensitive to melanin levels. The red and infrared light used by these devices is absorbed differently by melanin, leading to inaccurate readings. The implications of this bias are far-reaching, affecting patient care across various settings, from emergency rooms to intensive care units and even home monitoring. Clinicians relying on these readings may inadvertently undertreat hypoxemia in patients of color, assuming adequate oxygenation when, in reality, it is critically low. This systemic failure underscores a fundamental challenge in medical technology: the need for rigorous, equitable validation across diverse populations.

Regulatory Oversight and Unaddressed Failures

The continued prevalence of this known bias has historically raised serious questions about the regulatory pathways for medical devices. Multiple device manufacturers have produced and continue to market pulse oximeters that exhibit this racial bias. The FDA’s Center for Devices and Radiological Health (CDRH) is the primary regulatory body responsible for ensuring the safety and effectiveness of medical devices in the United States. Devices typically achieve market clearance through pathways such as the FDA 510(k) Pathway, which requires demonstrating substantial equivalence to a predicate device. However, the pulse oximeter issue demonstrates that even FDA-cleared devices can have systematic racial bias. Recognizing this, the FDA has recently taken steps to address these concerns, including issuing draft guidance in January 2025 aimed at improving the accuracy and performance of pulse oximeters across skin tones. This draft guidance recommends expanded testing requirements, diverse testing populations with a specific focus on including individuals with a wide range of skin tones, and increased sample sizes for studies. This proactive move by the FDA seeks to close a critical gap in traditional regulatory frameworks, where initial clearance may not have adequately addressed or mandated ongoing evaluation for algorithmic bias across diverse populations. Experts like Ziad Obermeyer, a physician and researcher, have extensively studied and commented on algorithmic bias in healthcare, including in devices like pulse oximeters. His work, alongside insights from figures like Eric Topol, emphasizes the urgent need for a paradigm shift in how medical technologies are developed, tested, and regulated to prevent such disparities. The challenge is further compounded by the increasing integration of AI into medical devices, often categorized as Software as a Medical Device (SaMD). While pulse oximeters themselves may not always be classified as SaMD, the data they generate is frequently fed into AI-powered clinical decision support systems. If the foundational data from a biased pulse oximeter is flawed, any subsequent AI analysis or recommendation built upon it will inherit and potentially amplify that bias, leading to cascading errors in patient care. This underscores the need for robust validation protocols that specifically address fairness and equity in AI/ML medical devices.

The Path Forward: Responsible AI and Equitable Validation

The persistent issue of pulse oximeter racial bias serves as a stark reminder that technological advancement without equitable validation is not progress. The FDA CDRH has a critical role to play in evolving its regulatory approach to explicitly address algorithmic bias, not just for novel AI systems but for established devices as well. The recent draft guidance for pulse oximeters is an example of this evolving approach. The FDA SaMD Framework offers a potential avenue for this, emphasizing a total product lifecycle approach that includes real-world performance monitoring and a focus on transparency and interpretability. FDA SaMD Framework guidance For responsible AI, the development and deployment process must embed fairness from inception. This means:

• Diverse Training Data: Ensuring that AI models are trained on datasets representative of the entire patient population, not just a dominant demographic.
• Bias Detection and Mitigation: Proactively testing for and mitigating biases at every stage of development, including rigorous subgroup analysis.
• Transparency and Explainability: Making the decision-making processes of AI systems understandable, especially when outcomes differ across demographic groups.
• Continuous Monitoring: Implementing robust post-market surveillance systems to detect and address algorithmic drift and emerging biases in real-world use. NEJM article on AI bias in healthcare

The work of institutions like Johns Hopkins University in highlighting these disparities is invaluable, pushing for greater accountability and more equitable healthcare technologies. The medical community, including clinicians and patient safety advocates, must demand higher standards from device manufacturers and regulatory bodies.

A Call for Systemic Change

The ongoing saga of pulse oximeter racial bias is more than a historical footnote; it is a live incident of algorithmic bias with tangible, negative consequences for patient safety. It demonstrates that even widely accepted and FDA-cleared medical devices can harbor systemic inequalities that go unaddressed for decades. This case should serve as a powerful cautionary tale for the development and deployment of all medical AI. For Patient Safety Advocates, Clinicians, and FDA/Regulatory Officers, the message is clear: vigilance against algorithmic bias must be paramount. The trust placed in medical technology by patients and providers alike demands that devices, whether traditional or AI-powered, are rigorously tested and continuously monitored for equitable performance across all populations. Without a fundamental shift towards proactive bias detection and mitigation, we risk perpetuating and even amplifying health disparities through the very technologies designed to improve care. The time for a comprehensive, equitable approach to medical device validation is long overdue. Johns Hopkins study on pulse oximeter accuracy