Residual-based monitoring of human health
Improved human health monitoring is provided in the context of sensor measurements of typical vital signs and other biological parameters, by a system and method using an empirical model of the parameters and disposed to estimate values of the parameters in response to actual measurements. Residuals resulting from the difference between the estimates and actual measurements are analyzed for robust indications of incipient health issues. Residual analysis is both more robust and more sensitive than conventional univariate range checking on vital signs.
1. A computer-automated monitoring system for early detection of incipient health problems in a monitored patient, comprising:
a computer-accessible memory for storing a set of exemplary data observations of multiple vital signs obtained from said monitored patient comprising a personalized model of said patient:
a data feed for receiving new data observations of said multiple vital signs from said monitored patient;
a computer configured by a kernel-based modeling software module to generate an estimate of the value of at least one of said multiple vital signs, as a weighted combination of a subset of said exemplary data observations where said weights are determined from a kernel function comparison of said subset of exemplary data observations and a said received observation;
said computer being further configured by a localization software module to select said subset of exemplary data observations based on at least some of the vital signs in said received observation;
said computer being further configured by a residual analysis software module to compare said estimate to the corresponding measured value of the at least one of said multiple vital signs in said received data observation to generate a residual; and
said computer being further configured by an alerting module to trigger an alert when said generated residual exceeds a threshold, and present said alert in a web-based interface usable by a medical practitioner for monitoring said patient for incipient health problems.
2. A system according to claim 1 , wherein said kernel-based modeling software module implements a similarity-based model.
3. A system according to claim 1 , wherein said kernel-based modeling software module implements a kernel regression estimator.
4. A system according to claim 1 , wherein said kernel-based modeling software module implements a radial basis function network.
5. A computer automated monitoring system according to claim 1 , wherein said alert module generate an alert only when said residual exceeds said threshold for a preselected number of data observations in a preselected window.
6. A computer automated monitoring system according to claim 1 , wherein said computer is further configured by an adaptation module disposed to add a received data observation to said set of exemplary data observations comprising said personalized model upon selection by a clinical user of an option to adapt said model responsive to an alert.