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

BackgroundLong-term oxygen therapy (LTOT) is essential for patients with chronic hypoxemia, particularly due to chronic obstructive pulmonary disease (COPD). However, conventional LTOT relies on static oxygen flow rates that fail to reflect patients fluctuating physiological demands. Emerging systems leveraging artificial intelligence (AI) and automation offer real-time SpO2 monitoring and adaptive oxygen titration. This systematic review evaluates the performance, usability, and clinical readiness of such technologies in LTOT. MethodsFollowing PRISMA 2020 guidelines, we searched PubMed, Web of Science, and Scopus (2000-2024) for English-language, peer-reviewed studies describing AI or automated SpO2 systems in adults with LTOT relevance. Inclusion criteria focused on systems addressing motion artifacts, low-perfusion reliability, or skin tone bias. Extracted data included algorithm design, accuracy (e.g., MAE, RMSE), usability, and risk of bias. ResultsEight studies met the inclusion criteria: five AI-based and three automated systems. AI models achieved high SpO2 estimation accuracy (MAE as low as 0.57%) and addressed motion artifacts and demographic bias. Cabanas et al.6 notably performed skin tone-stratified validation. Automated systems like O2matic and iPOC demonstrated clinical efficacy in maintaining SpO2 targets and reducing provider workload. However, AI models were mostly in prototype or simulation stages and lacked real-world LTOT validation. Risk of bias was moderate to high, especially in participant selection and algorithm transparency. ConclusionsAI and automation offer distinct yet synergistic benefits for modernizing LTOT. Automated systems support immediate clinical use, while AI models hold promise for personalized, predictive oxygen therapy. Future work must prioritize real-world validation, equity, usability, and regulatory alignment to ensure ethical and effective integration into LTOT care.