Validation of a portable, deployable system for continuous vital sign monitoring using a multiparametric wearable sensor and personalised analytics in an Ebola treatment centre.

BACKGROUND: The recent Ebola epidemic in West Africa strained existing healthcare systems well beyond their capacities due to the extreme volume and severity of illness of the patients. The implementation of innovative digital technologies within available care centres could potentially improve patient care as well as healthcare worker safety and effectiveness. METHODS: We developed a Modular Wireless Patient Monitoring System (MWPMS) and conducted a proof of concept study in an Ebola treatment centre (ETC) in Makeni, Sierra Leone. The system was built around a wireless, multiparametric 'band-aid' patch sensor for continuous vital sign monitoring and transmission, plus sophisticated data analytics. Results were used to develop personalised analytics to support automated alerting of early changes in patient status. RESULTS: During the 3-week study period, all eligible patients (n=26) admitted to the ETC were enrolled in the study, generating a total of 1838 hours of continuous vital sign data (mean of 67.8 hours/patient), including heart rate, heart rate variability, activity, respiratory rate, pulse transit time (inversely related to blood pressure), uncalibrated skin temperature and posture. All patients tolerated the patch sensor without problems. Manually determined and automated vital signs were well correlated. Algorithm-generated Multivariate Change Index, pulse transit time and arrhythmia burden demonstrated encouraging preliminary findings of important physiological changes, as did ECG waveform changes. CONCLUSIONS: In this proof of concept study, we were able to demonstrate that a portable, deployable system for continuous vital sign monitoring via a wireless, wearable sensor supported by a sophisticated, personalised analytics platform can provide high-acuity monitoring with a continuous, objective measure of physiological status of all patients that is achievable in virtually any healthcare setting, anywhere in the world.

Functionality of empirical model-based predictive analytics for the early detection of hemodynamic instabilty.

UNLABELLED: 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 patient’s 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 patient’s physiologic state. This platform allows for the use of predictive analytic techniques to identify early changes in a patient’s 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.