Yue Xu,1,* Jingjing Xue,1,* Yunfeng Deng,1,* Lili Tu,1 Yu Ding,2 Yibing Zhang,1 Xinrui Yuan,1 Kexin Xu,1 Liangmei Guo,3 Na Gao1 

1Department of Cardiology, The Second Medical Center & National Clinical Research Center for Geriatric Diseases, Chinese People’s Liberation Army (PLA) General Hospital, Beijing, 100853, People’s Republic of China; 2Department of Gastroenterology, The Second Medical Center & National Clinical Research Center for Geriatric Diseases, Chinese People’s Liberation Army (PLA) General Hospital, Beijing, 100853, People’s Republic of China; 3Health Sciences, The Second Medical Center & National Clinical Research Center for Geriatric Diseases, Chinese People’s Liberation Army (PLA) General Hospital, Beijing, 100853, People’s Republic of China

*These authors contributed equally to this work

Correspondence: Na Gao, Department of Cardiology, The Second Medical Center & National Clinical Research Center for Geriatric Diseases, Chinese People’s Liberation Army (PLA) General Hospital, Beijing, 100853, People’s Republic of China, Tel +86 18515165409, Email gaona301s@163.com Liangmei Guo, Email amy_kuo@163.com

Background: Mechanical ventilation, a key ICU life-support tech, carries risks. ML can optimize patient management, improving clinical decisions, patient outcomes, and resource use.
Objective: This review aims to summarize the current applications, challenges, and future directions of machine learning in managing mechanically ventilated patients, focusing on prediction models for extubation readiness, oxygenation management, ventilator parameter optimization, clinical prognosis, and pulmonary function assessment.
Methods: Multiple databases, including PubMed, Web of Science, CNKI and Wanfang Data were systematically searched for studies on machine learning in mechanical ventilation management. Keywords included mechanical ventilation, machine learning, weaning, etc. We reviewed recent studies on using machine learning to predict successful extubation, optimize oxygenation targets, personalize ventilator settings, forecast mechanical ventilation duration and clinical outcomes. The review also examined challenges of integrating machine learning into clinical practice, such as data integration, model interpretability, and real - time performance requirements.
Results: Machine learning models have demonstrated significant potential in predicting successful extubation, optimizing oxygenation strategies through non-invasive blood gas prediction, and dynamically adjusting ventilator parameters using reinforcement learning. These models have also shown promise in predicting mechanical ventilation duration, clinical prognosis and pulmonary function parameters. However, challenges remain, including data heterogeneity, model generalizability, workflow integration, and the need for multicenter validation.
Conclusion: Machine learning shows great potential for improving intensive care quality and efficiency in mechanically ventilated patients. However, challenges like model interpretability, real-time performance, clinical and validation remain. Future research needs to focus on these limitations via large-scale, multicenter trials, better data standardization, and improved physician training to safely and effectively integrate ML into clinical practice. Collaboration among medical, engineering, and ethical experts is also essential for advancing this promising field.

Keywords: machine learning, mechanical ventilation, weaning, prognostication models