Hybrid Approach Integrating LSTM and Statistical Process Control for Anomaly Detection in SpO2 Monitoring Under Various Altitude Environments

Detecting anomalies in blood oxygen saturation (SpO2) levels presents significant challenges due to the complex interaction of physiological and environmental factors, particularly in high-altitude environments, where the complexity of SpO2 fluctuations and individual heterogeneity make it difficult to distinguish true anomalies. Traditional monitoring methods, such as Statistical Process Control (SPC) using control charts, are widely used to detect systematic changes, but they often fail to identify true SpO2 anomalies that are not caused by brief equipment malfunctions or the influence of other physiological indicators. To address this limitation, this article proposes an innovative approach that integrates SPC with a Long Short-Term Memory (LSTM) model to enhance anomaly detection in SpO2 levels. The LSTM model predicts future SpO2 values based on multivariate physiological time-series data, generating residuals between the predicted and actual values, and one of control charts is then used to detect SpO2 deviations that cannot be explained by other physiological indicators. The model delivers high-accuracy multi-step predictions, achieving a Root Mean Squared Error (RMSE) of 1.13% for both the 3-step and 5-step predictions. By combining population-level (Shewhart-based X-bar) and individual-level (EWMA of residuals) control charts, this dual-chart system improves the precision of SpO2 anomaly detection across different altitude levels and activity states. This hybrid approach offers valuable insight into SpO2 alarm mechanism and provides a more robust framework for real-time SpO2 monitoring and oxygen supply management in high-risk environments.