Adaptive Change Point Detection for Respiratory Variables

Current alarm strategies for physiological monitoring depend on predetermined thresholds without consideration for the heterogeneity between patients or intraoperative variations. To improve upon this situation, we developed an adaptive change point detection scheme to automatically notify the clinician when a change of clinical significance has occurred in the respiratory variables. We modeled end-tidal carbon dioxide, expiratory minute volume, and respiratory rate using a dynamic linear growth model, whose noise covariances are estimated by an adaptive Kalman filter based on a recursive expectation-maximization method. Change points are detected by the CUSUM testing. The comparison of the results with post-hoc expert annotations demonstrates that the algorithm can accurately detect relevant changes in the respiratory signals