Forecasting clinical events from short physiologic timeseries
Systems, methods and computer-readable media are provided for monitoring patients and quantitatively predicting whether an event such as a significant change in health status meriting intervention, is likely to occur within a future time interval subsequent to computing the prediction. Medical data for a patient is collected from one or more different inputs and used to determine time series data. From this, a forecasted numerical value is computed for one or more physiologic parameters associated with the patient, which may be used to further monitor the patient and facilitate decision making about a need for intensified monitoring or intervention to prevent or manage physiologic or hemodynamic deterioration. An evolutionary algorithm, such as particle swarm optimization and/or differential evolution, may be used to determine the most probable value of the physiologic parameter(s) at one or more future times.
1. A non-transitory computer-readable media having computer-executable instructions embodied thereon that when executed, facilitate a method for periodically monitoring at least one patient, the method comprising:
collecting a set of physiological data of a patient, wherein the set of physiological data comprises a set of data points collected at a plurality of data measurement times by one or more physiologic sensors, and wherein each data point of the set of data points includes a corresponding date-time stamp;
generating a time series using the set of data points of the set of physiological data of the patient collected from the one or more physiologic sensors;
identifying a portion of the time series that is useable for computing one or more forecasted values associated with variables for the set of physiological data at a future time based on the generated time series;
determining, using one or more evolutionary algorithms, whether the one or more forecasted values are outside a control limit range associated with variables for the set of physiological data, the one or more evolutionary algorithms including at least one of particle swarm optimization, differential evolution, or Hankel matrix, the one or more evolutionary algorithms are to solve for a most probable forecasted value for the one or more forecasted values,
wherein the particle swarm optimization generates predicted physiologic parameters for the set of physiological data by generating a plurality swarms of the time series, each swarm exhibiting a tendency toward a global outcome for minimizing an objective function; and
issuing an alert on a user interface, wherein the alert indicates that the at least one patient is forecasted to deteriorate or improve.
2. The non-transitory media of claim 1 , wherein the one or more physiologic sensors include at least one sensor positioned on an appendage of the patient configured to collect heart rate, blood pressure, oxygen saturation, central venous pressure, or muscle activity.
3. The non-transitory media of claim 1 , wherein the one or more physiologic sensors include at least one optical sensor configured to detect movement, kinematic modeling, or velocity measurements.
4. The non-transitory media of claim 1 , wherein the one or more physiologic sensors include at least one sensor that is ingestible, sub-dermal, or affixed to the skin of the patient.
5. The non-transitory media of claim 1 , wherein the method further comprises receiving the set of physiological data from an electronic health record (EHR) computer system.
6. The non-transitory media of claim 1 , wherein the set of physiological data comprises serial measurements captured by each of the one or more physiologic sensors.
7. The non-transitory media of claim 6 , wherein the time series is represented as the Hankel matrix of the serial measurements.
8. The non-transitory media of claim 1 , wherein the set of physiological data includes physiological data captured at irregular intervals by the one or more physiologic sensors.
9. The non-transitory computer-readable media according to claim 1 , wherein the one or more evolutionary algorithms identify a skeleton algebraic sequence that characterizes the portion of the time series.
10. A computer-implemented method for assessing a patient comprising:
collecting, by at least one processor, a set of physiological data of the patient, wherein the physiological data comprises a set of data points collected at a plurality of data measurement times by one or more physiologic sensors, and wherein each data point of the set of data points includes a corresponding date-time stamp;
generating, by the at least one processor, a time series using the set of data points of the set of physiological data of the patient collected from the one or more physiologic sensors;
identifying, by the at least one processor, a portion of the time series that is useable for computing one or more forecasted values associated with variables for the set of physiological data at a future time based on the generated time series;
determining, by the at least one processor, using one or more evolutionary algorithms, whether the one or more forecasted values are outside a control limit range associated with variables for the set of physiological data, the one or more evolutionary algorithms including at least one of a particle swarm optimization, differential evolution, or Hankel matrix, the one or more evolutionary algorithms are to solve for a most probable forecasted value for the one or more forecasted values,
wherein the particle swarm optimization generates predicted physiologic parameters for the set of physiological data by generating a plurality swarms of the time series, each swarm exhibiting a tendency toward a global outcome for minimizing an objective function; and
issuing, by the at least one processor, an alert on a user interface, wherein the alert indicates that the patient is forecasted to deteriorate or improve.
11. The computer-implemented method of claim 10 , wherein when the patient is forecasted to deteriorate, the computer-implemented method further comprising generating instructions to modify a treatment program for the patient, wherein modifying the treatment program comprises administration of a systemically, regionally or locally applied pharmaceutical.
12. The computer-implemented method of claim 10 , wherein the forecasted values indicate one or more of shock, cardiac decompensation, respiratory insufficiency, hypovolemia, progression of diabetes, congestive heart failure, infection or sepsis, dehydration, hemorrhage, hypotension, exposure to chemical or biological agents, or inflammatory response.
13. The computer-implemented method of claim 10 , wherein the one or more physiologic sensors include at least one sensor positioned on an appendage of the patient configured to collect heart rate, blood pressure, oxygen saturation, central venous pressure, or muscle activity.
14. The computer-implemented method of claim 10 , wherein the one or more physiologic sensors include at least one optical sensor configured to detect movement, kinematic modeling, or velocity measurements.
15. The computer-implemented method of claim 10 , the one or more evolutionary algorithms solve for the most probable forecasted value at one or more future time points.
16. The computer-implemented method of claim 10 , wherein the set of physiological data includes physiological data captured at irregular intervals by the one or more physiologic sensors.
17. A patient monitoring system for periodically monitoring at least one patient using an evolutionary algorithm to prevent physiologic or hemodynamic deterioration comprising:
a processor; and
computer storage memory having computer-executable instructions stored thereon which, when executed by the processor, implement a method of monitoring a patient, the method comprising:
collecting a set of physiological data of the patient, wherein the physiological data comprises a set of data points collected at a plurality of data measurement times by one or more physiologic sensors, and wherein each data point of the set of data points includes a corresponding date-time stamp;
generating a time series using the set of data points of the set of physiological data;
identifying a portion of the time series that is useable for computing one or more forecasted values associated with variables for the set of physiological data at a future time based on the generated time series;
determining, using one or more evolutionary algorithms, whether the one or more forecasted values are outside a control limit range associated with variables for the set of physiological data, the one or more evolutionary algorithms including at least one of a particle swarm optimization, differential evolution, or Hankel matrix, the one or more evolutionary algorithms are to solve for a most probable forecasted value for the one or more forecasted values,
wherein the particle swarm optimization generates predicted physiologic parameters for the set of physiological data by generating a plurality swarms of the time series, each swarm exhibiting a tendency toward a global outcome for minimizing an objective function; and
issuing an alert on a user interface, wherein the alert indicates that the at least one patient is forecasted to deteriorate or improve.
18. The patient monitoring system of claim 17 , wherein the system further comprises one or more sensors configured to collect physiological information for the patient.