Systems and methods for transitioning patient care from signal based monitoring to risk based monitoring
Inventors
Baronov, Dimitar V. • Butler, Evan J. • Lock, Jesse M. • McManus, Michael F.
Assignees
Interested in licensing this patent?
MTEC can help explore whether this patent might be available for licensing for your application.
Publication Number
US-10062456-B2
Publication Date
2018-08-28
Expiration Date
Abstract
A risk-based patient monitoring system for critical care patients combines data from multiple sources to assess the current and the future risks to the patient, thereby enabling providers to review a current patient risk profile and to continuously track a clinical trajectory. A physiology observer module in the system utilizes multiple measurements to estimate Probability Density Functions (PDF) of a number of Internal State Variables (ISVs) that describe a components of the physiology relevant to the patient treatment and condition. A clinical trajectory interpreter module in the system utilizes the estimated PDFs of ISVs to identify under which probable patient states the patient can be currently categorized and assign a probability value that the patient will be in each of the identified states. The combination of patient states and their probabilities is defined as the clinical risk to the patient.
Core Innovation
The invention relates to a risk-based monitoring system for transforming measured data, including heart rate and blood oxygen, of a critical care patient into hidden internal state data that is not directly measurable with sensors. The hidden internal state data includes hidden internal state variables monitored by the system, with at least one internal state variable characterized as a physiologically relevant parameter for at least one treatment and a condition of the patient. The system receives sensor data over a series of time steps and uses that time-indexed internal state variable representation for risk determination.
A physiology observer module updates the sensor-provided data using a first and second computer process. The first process generates a conditional likelihood kernel for internal state variables at a time step, the conditional likelihood kernel comprising a set of probability density functions for distinct internal state variables. The second process uses Bayes theorem to generate posterior predicted conditional probability density functions for each internal state variable at the time step, given both the conditional likelihood kernel and a predicted conditional probability density function for the same internal state variables.
A clinical trajectory interpreter module determines, based on the generated posterior predicted conditional probability density functions, a set of possible states of a hidden internal state variable and generates a probability value representing the likelihood that the patient's oxygen delivery, which cannot be directly measured, is inadequate. A user interaction module displays graphical indicators corresponding to the states of the set of possible states, with the graphical indicators graphically identifying the probability that the hidden internal state variable is in a corresponding state across a range of time and indicating a hazard level.
Claims Coverage
The independent claim covers a risk-based monitoring system with a physiology observer that forms conditional likelihood kernels and Bayes-theorem posterior predicted conditional probability density functions for hidden internal state variables, followed by a clinical trajectory interpretation that maps those probabilistic outputs to possible hidden-state values related to inadequate oxygen delivery and displays hazard-level graphical indicators. The dependent claims further refine the probabilistic update behavior and add visualization and interaction features, including clinical trajectory display and dynamic time-point selection.
Risk-based transformation of measured sensor data into hidden internal state data
A system that transforms measured data, including heart rate and blood oxygen, of a critical care patient into hidden internal state data that is not directly measurable with sensors, where the hidden internal state is monitored by the system using internal state variables characterized by parameters physiologically relevant to at least one treatment and a condition of the patient.
Conditional likelihood kernel generation for internal state variables
A physiology observer module that generates, at a time step, a conditional likelihood kernel for the internal state variables, the conditional likelihood kernel comprising a set of probability density functions, each probability density function being for a distinct internal state variable at the time step.
Bayes theorem to generate posterior predicted conditional probability density functions
A physiology observer module that generates, using Bayes theorem, posterior predicted conditional probability density functions for each of the internal state variables at the time step given the conditional likelihood kernel for the internal state variables at the time step and a predicted conditional probability density function of each of the internal state variables for the time step.
Clinical trajectory interpretation for inadequate oxygen delivery likelihood
A clinical trajectory interpreter module that determines, based on the generated posterior predicted conditional probability density functions of the internal state variables at a time step, a set of possible states of a hidden internal state variable to generate a probability value representing the likelihood that the patient's oxygen delivery, which cannot be directly measured, is inadequate.
Graphical hazard-level indicators for hidden-state probabilities over time
A user interaction module that displays graphical indicators on the display device, each graphical indicator corresponding to one of the states of the set of possible states of the hidden internal state variable, graphically identifying the probability that the hidden internal state variable is in a corresponding state at a given point in a range of time, and configured to indicate a hazard level.
Gated conditional likelihood generation based on predetermined predicted likelihood
A physiology observer module that compares a newly received measurement tied to an internal state variable against a predetermined predicted likelihood of probable measurements and does not generate a conditional probability density function for the internal state variable when the new measurement is outside the predetermined predicted likelihood.
Time-step prediction of conditional probability density functions for t_k+1
A physiology observer module that, at a time step t_k, generates for each internal state variable m_S a predicted conditional probability density function for time t_k+1 using the posterior predicted probability density functions for that variable at time step t_k.
Integral-based computation of conditional probability density functions from a preceding time step
A system where conditional probability density functions at time step t_k+1 are computed using posterior conditional probability density functions of internal state variables from a preceding time step via an integral formula.
Clinical trajectory display derived from conditional probability density functions
A user interaction module that displays a patient clinical trajectory derived from a conditional probability density function of a hidden internal state variable across at least some time steps.
Interactive timeline controller with dynamic graphical evolution
A system further including a timeline controller that allows a user dynamically to select multiple time points over time steps t_k, where graphical indicators dynamically change to show how a hidden internal state variable evolves in response to the user-selected time point.
Across the claim set, the system’s core inventive coverage is directed to probabilistic estimation of hidden internal states from multi-time sensor inputs by constructing conditional likelihood kernels and applying Bayes theorem to obtain posterior predicted conditional probability density functions, then interpreting those probability distributions to generate probabilities for inadequate oxygen delivery and to present hazard-level graphical indicators over time. Dependent claims narrow the generation of conditional probability density functions using a predetermined predicted likelihood, extend the time-series prediction to t_k+1 using posterior predicted distributions, add an integral formula for conditional density computation, and enhance visualization via clinical trajectory display and interactive dynamic selection of time points.
Stated Advantages
Generates a probability value representing the likelihood that the patient's oxygen delivery is inadequate.
Displays hazard level based on probabilities of hidden internal state variable states.
Documented Applications
Critical care monitoring that transforms measured data (including heart rate and blood oxygen) into hidden internal state data and determines risk for inadequate oxygen delivery.
Treatment decision support involving blood transfusion risk/complications.
Modeling and risk stratification for nitric oxide treatment with standardized medical plans.
Outpatient chronic-condition monitoring for ADHD states using Vanderbilt Assessment Scale.
Dynamic Bayesian Network (DBN) modeling for DO2 estimation.
HLHS stage 1 palliation modeling using particle-filter inference.
Interested in licensing this patent?