Background. Monitoring cardiovascular hemodynamics in the modern clinical setting is a major challenge. Increasing amounts of physiologic data must be analyzed and interpreted in the context of the individual patients pathology and inherent biologic variability. Certain data-driven analytical methods are currently being explored for smart monitoring of data streams from patients as a first tier automated detection system for clinical deterioration. As a prelude to human clinical trials, an empirical multivariate machine learning method called Similarity-Based Modeling (SBM), was tested in an In Silico experiment using data generated with the aid of a detailed computer simulator of human physiology (Quantitative Circulatory Physiology or QCP) which contains complex control systems with realistic integrated feedback loops. Methods. SBM is a kernel-based, multivariate machine learning method that that uses monitored clinical information to generate an empirical model of a patients physiologic state. This platform allows for the use of predictive analytic techniques to identify early changes in a patients condition that are indicative of a state of deterioration or instability. The integrity of the technique was tested through an In Silico experiment using QCP in which the output of computer simulations of a slowly evolving cardiac tamponade resulted in progressive state of cardiovascular decompensation. Simulator outputs for the variables under consideration were generated at a 2-min data rate (0.083Hz) with the tamponade introduced at a point 420 minutes into the simulation sequence. The functionality of the SBM predictive analytics methodology to identify clinical deterioration was compared to the thresholds used by conventional monitoring methods. Results. The SBM modeling method was found to closely track the normal physiologic variation as simulated by QCP. With the slow development of the tamponade, the SBM model are seen to disagree while the simulated biosignals in the early stages of physiologic deterioration and while the variables are still within normal ranges. Thus, the SBM system was found to identify pathophysiologic conditions in a timeframe that would not have been detected in a usual clinical monitoring scenario. Conclusion. In this study the functionality of a multivariate machine learning predictive methodology that that incorporates commonly monitored clinical information was tested using a computer model of human physiology. SBM and predictive analytics were able to differentiate a state of decompensation while the monitored variables were still within normal clinical ranges. This finding suggests that the SBM could provide for early identification of a clinical deterioration using predictive analytic techniques.
Keywords: predictive analytics, hemodynamic, monitoring.