Abstract
BACKGROUND: Postoperative pulmonary complications (PPCs) are a significant concern in cardiac surgery, affecting patient prognosis. This pilot study explored the feasibility of developing a machine learning model for preoperative PPCs risk stratification by integrating dynamic respiratory physiology from the six-minute walk test (6MWT) with clinical parameters. METHODS: A prospective study was conducted at the Department of Cardiovascular Surgery of West China Hospital, Sichuan University, from August 2021 to December 2022. We enrolled 142 consecutive patients undergoing valvular heart surgery. After quality control, 117 patients with complete synchronized respiratory monitoring during 6MWT and clinical data were included. We extracted 94 physiological features across 6MWT phases (baseline, walking, recovery) and clinical variables, developing predictive models using five machine learning algorithms evaluated through rigorous five-fold cross-validation. RESULTS: The logistic regression model demonstrated promising discriminative performance (AUC 0.86, 95% CI [0.81-0.89]) in this exploratory cohort. Preliminary physiological patterns emerged, including associations between elevated expiratory tidal volume during recovery (OR 9.70, p = 0.006) and reduced baseline minute ventilation (OR 0.15, p = 0.002) with higher PPCs risk. CONCLUSION: These pilot findings suggest that continuous physiological monitoring during 6MWT, when combined with clinical data, may provide a feasible approach for preoperative PPCs risk assessment. While requiring multi-center validation, the results highlight the potential of wearable-enabled respiratory monitoring to guide prehabilitation strategies in cardiac surgery.