AI-Driven and Automated Continuous Oxygen Saturation Monitoring and LTOT: A Systematic Review

Long-term oxygen therapy (LTOT) is a foundational intervention for individuals with chronic hypoxemia, particularly those with chronic obstructive pulmonary disease (COPD). Traditional LTOT delivery relies on static oxygen flow prescriptions, often determined during clinical assessments, which may not reflect patients' fluctuating physiological demands during everyday activities. Recent innovations in computational modeling and automated systems have introduced tools capable of continuously monitoring peripheral oxygen saturation (SpO₂) and adjusting oxygen delivery in real time. This systematic review evaluates technologies designed to optimize SpO₂ monitoring and oxygen titration in LTOT. A comprehensive literature search was conducted across five databases (PubMed, Scopus, Web of Science, MDPI, and ACM Digital Library) for peer-reviewed studies published between 2000 and 2024. Studies were included if they described automated or computational systems applied to adult LTOT populations. Data were extracted on algorithm performance, motion artifact handling, usability, validation strategies, and risk of bias. Eight studies met the eligibility criteria, including five that utilized AI-based SpO₂ estimation methods and three that implemented closed-loop oxygen control systems. AI models demonstrated strong technical accuracy (mean absolute error as low as 0.57%) and addressed key limitations such as motion artifacts and signal bias. However, most were tested under simulated or retrospective conditions. Rule-based systems such as O₂matic and iPOC showed promising results in clinical and ambulatory settings but lacked personalized adaptation. Findings suggest that AI and automation offer complementary benefits in advancing LTOT. Further work is needed to support real-world deployment, ensure equitable design, and align these innovations with regulatory and clinical practice standards.