Multivariate Residual-Based Health Index for Human Health Monitoring
Ambulatory or in-hospital monitoring of patients is provided with early warning and prioritization, enabling proactive intervention and amelioration of both costs and risks of health care. Multivariate physiological parameters are estimated by empirical model to remove normal variation. Residuals are tested using a multivariate probability density function to provide a multivariate health index for prioritizing medical effort.
1 . A system for monitoring the health of a human, comprising:
a device disposed to receive multiple physiological signals from a plurality of sensors arranged to capture physiological signals from a monitored human, having microprocessor hardware programmed to derive multiple physiological features from said signals;
a computer configured to receive monitored observations of said multiple physiological features from said device, to generate estimates of said features in the monitored observations using a model embodying behavior of said multiple physiological features under normal health conditions, and to generate residuals of said features by differencing the monitored observations with the estimates thereof; and
a computer-accessible memory storing a set of exemplary observations of residuals of said multiple physiological features characteristic of a normal health state for the monitored human, generated using said model;
said computer being specially configured to determine a likelihood that said residuals are representative of a pattern of residuals for said normal health state, using a Gaussian mixture model based on said set of exemplary observations of residuals to approximate a probability distribution for normal residual patterns, whereby said likelihood consolidates the behaviors of the individual residuals for each of the features into one overall index to summarize any deviation of the physiological health of said human from normal.
2 . A system according to claim 1 , wherein said model is a kernel regression estimator that has a library of exemplary observations of said features representative of normal health conditions, and generates said estimates of said features in the monitored observations as a weighted sum of at least some of said exemplary observations.
3 . A system according to claim 2 , wherein the monitored observation of features is used to localize said model, by determining a subset of said exemplary observations to use for generating said estimates, based on a measure of similarity between said monitored observation and said exemplary observations.
4 . A system according to claim 1 , wherein said model is a similarity-based model that has a library of exemplary observations of said features representative of normal health conditions, and generates said estimates of said features in the monitored observations as a weighted sum of at least some of said exemplary observations.
5 . A system according to claim 4 , wherein the monitored observation of features is used to localize said model, by determining a subset of said exemplary observations to use for generating said estimates, based on a measure of similarity between said monitored observation and said exemplary observations.
6 . A system according to claim 1 , wherein said computer is further specially configured to test said likelihood to render a decision whether the monitored observation of said multiple features is characteristic of said normal health state, by comparing it to a threshold.
7 . A system according to claim 6 , wherein said computer is further specially configured to test a series of said rendered decisions for persistence of like decisions regarding whether the features are characteristic of said normal health state or not.
8 . A system according to claim 1 , wherein said device is a cell phone.
9 . A system according to claim 1 , wherein said device is a hospital bedside vital signs monitor.
10 . A system according to claim 1 , wherein said computer is configured to first scale said residuals for the monitored features, using the means and standard deviations calculated from normal data residuals, and wherein said exemplary observations of residuals characteristic of a normal health state are likewise scaled.
11 . A system according to claim 1 , wherein said device receives sensor data from sensors embedded inside the monitored human in connection with an implanted cardiac device.
12 . A system according to claim 1 , wherein said device receives sensor data by wireless transmissions via extremely local radio protocol of measurements of sensors attached to the monitored human.
13 . A system according to claim 1 , wherein said device receives physiological signals comprising an electrocardiogram, a bioimpedance, and a photoplethysmogram for at least two wavelengths.
14 . A system according to claim 13 , wherein said device derives physiological features from the physiological signals, comprising a heart rate, a respiration rate, a pulse transit time, and a ratio of absorption of the at least two wavelengths
15 . A system according to claim 14 , wherein the physiological signals received by the device further comprise at least one accelerometer signal, and the device is further configured to identify what times the accelerometer signal indicates motion artifact is likely present in the derived features in order to ignore features at those times.
16 . A computerized system for monitoring the health of a human, configured by program code to perform the steps of:
receiving multivariate monitored observations of multiple vital sign features derived from physiological signals captured by at least one sensor on the monitored human;
storing a model embodying behavior of said multiple vital sign features under normal health conditions personalized to the monitored human;
generating estimates of at least some of said multiple vital sign features in the multivariate monitored observations using said model;
generating monitored residuals of the at least some of said multiple vital sign features by differencing those features of the monitored observations that are estimated with the estimates thereof;
storing a set of exemplary observations of residuals of said multiple vital sign features characteristic of a normal health state personalized to the monitored human and generated using said model; and
determining a likelihood that said monitored residuals are representative of a pattern of residuals for said normal health state, using a Gaussian mixture model based on said set of exemplary observations of residuals to approximate a probability distribution for normal residual patterns, whereby said likelihood consolidates the behaviors of the individual residuals for each of the features into one overall index to summarize any deviation of the physiological health of said human from normal.
17 . A computerized system according to claim 16 , further configured to test said likelihood to render a decision whether the monitored observation of said multiple vital sign features is characteristic of said normal health state, by comparing it to a threshold.
18 . A computerized system according to claim 17 , further configured to test a series of said rendered decisions for persistence of like decisions regarding whether the features are characteristic of said normal health state or not.
19 . A computerized system according to claim 16 , wherein determining a likelihood further comprises scaling said residuals for the monitored vital sign features, using the means and standard deviations calculated from known normal data residuals, and wherein said exemplary observations of residuals of said multiple vital sign features characteristic of a normal health state personalized to the monitored human, are likewise scaled.
20 . A computerized system according to claim 16 , wherein said model is a kernel based estimator that has a library of exemplary observations of said vital sign features representative of normal health conditions, and generates said estimates of said features in the monitored observations as a weighted sum of at least some of said exemplary observations.
21 . A computerized system according to claim 20 , wherein the monitored observation of features is used to localize said kernel-based model, by determining a subset of said exemplary observations to use for generating said estimates, based on a measure of similarity between said monitored observation and said exemplary observations.
22 . A computerized system according to claim 16 , further configured to host a web-based user interface and present a list of patients prioritized based on said likelihood for each patient that said monitored residuals are representative of a pattern of residuals for said normal health state.
23 . A computerized system according to claim 16 , wherein said multiple vital sign features comprises a heart rate, a respiration rate, a pulse transit time, and a ratio of absorption of the at least two wavelengths.