Patient-specific glucose prediction systems and methods
Infusion devices and related medical devices, patient data management systems, and methods are provided for monitoring a physiological condition of a patient. An exemplary method of monitoring a physiological condition of a patient involves obtaining current measurement data for the physiological condition of the patient provided by a sensing arrangement, obtaining a user input indicative of one or more future events associated with the patient, and in response to the user input, determining a prediction of the physiological condition of the patient in the future based at least in part on the current measurement data and the one or more future events using one or more prediction models associated with the patient, and displaying a graphical representation of the prediction on a display device.
1 . A method of monitoring a physiological condition of a patient, the method comprising:
obtaining, from a sensing arrangement inserted at least partially subcutaneously in the patient, current measurement data for the physiological condition of the patient;
predicting one or more events likely to influence the physiological condition of the patient at one or more different times in the future based at least in part on historical event data associated with the patient;
determining a plurality of forecast values for the physiological condition of the patient associated with a plurality of different time periods in the future based at least in part on the current measurement data and the one or more events using a forecasting model associated with the patient, wherein:
the forecasting model comprises a neural network including a plurality of cells corresponding to the plurality of different time periods with cells in the plurality of cells sequentially coupled such that a first cell in the plurality of cells provides a forecast value of the plurality of forecast values to a subsequent cell in the plurality of cells; and
displaying, on a display device operatively coupled to the sensing arrangement at least partially inserted subcutaneously in the patient, the plurality of forecast values with respect to the plurality of different time periods in the future.
2 . The method of claim 1 , further comprising determining the forecasting model associated with the patient based at least in part on a relationship between historical measurement data for the physiological condition of the patient and the historical event data associated with the patient.
3 . The method of claim 2 , wherein:
each cell is configured to output an average value for the physiological condition during the respective time period in the future based at least in part on a subset of the one or more events predicted to occur within the respective time period in the future, the average value being the forecast value.
4 . The method of claim 2 , wherein determining the forecasting model comprises determining, for each respective hourly interval of a plurality of hourly intervals, a respective long short-term memory (LSTM) unit that is the cell of the plurality of cells, each LSTM unit configured to output the forecast value for the physiological condition during the respective hourly interval of the plurality of hourly intervals based at least in part on a relationship between a respective subset of the historical measurement data corresponding to the respective hourly interval and a respective subset of the historical event data corresponding to the respective hourly interval.
5 . The method of claim 4 , wherein determining the plurality of forecast values comprises, for each respective hourly interval in the future, calculating a respective hourly average forecast value associated with the respective hourly interval based at least in part on a subset of the one or more events predicted to occur within the respective hourly interval using the respective LSTM unit associated with the respective hourly interval.
6 . The method of claim 1 , further comprising obtaining, from a second sensing arrangement, contextual measurement data, wherein predicting the one or more events comprises predicting the one or more events based at least in part on the contextual measurement data.
7 . The method of claim 1 , further comprising obtaining, from a second sensing arrangement, contextual measurement data, wherein determining the plurality of forecast values comprises determining the plurality of forecast values based at least in part on the current measurement data, the one or more events, and the contextual measurement data using the forecasting model.
8 . The method of claim 1 , the current measurement data comprising sensor glucose measurement data, wherein:
predicting the one or more events comprises predicting one or more of a meal, exercise, an insulin delivery, and a dosage of medication at the one or more different times in the future; and
determining the plurality of forecast values comprises determining forecasted hourly average glucose levels for the patient associated with a plurality of hourly intervals in the future based at least in part on the sensor glucose measurement data and the one or more of the meal, the exercise, the insulin delivery, and the dosage of medication predicted at the one or more different times in the future.
9 . The method of claim 1 , wherein displaying the plurality of forecast values comprises displaying a plurality of forecasted hourly average glucose levels for the patient for a plurality of hourly intervals in the future.
10 . A non-transitory computer-readable medium having instructions stored thereon that are executable by a processing system coupled to the display device to perform the method of claim 1 .
11 . A system comprising:
a display device;
a sensing arrangement inserted at least partially subcutaneously in a patient to obtain current measurement data for a physiological condition of a patient, wherein the sensing arrangement is operatively coupled to the display device; and
a control system coupled to the display device and the sensing arrangement to determine a plurality of forecast hourly average values for the physiological condition of the patient in the future based at least in part on the current measurement data using an hourly forecasting model associated with the patient and display the plurality of forecast hourly average values on the display device, wherein the hourly forecasting model comprises a neural network including a plurality of cells corresponding to a plurality of different time periods with cells in the plurality of cells sequentially coupled such that a first cell in the plurality of cells provides a forecast value of the plurality of forecast hourly average values to a subsequent cell in the plurality of cells.
12 . The system of claim 11 , wherein the current measurement data comprises current sensor glucose measurement data and the plurality of forecast hourly average values comprise a plurality of forecast hourly average glucose levels for the patient in the future.
13 . The system of claim 12 , wherein:
the control system predicts one or more of a meal, exercise, an insulin delivery, and a dosage of medication at one or more different times in the future based at least in part on historical event log data associated with the patient; and
determining the plurality of forecast hourly average values comprises calculating the plurality of forecast hourly average glucose levels based at least in part on the current sensor glucose measurement data and the one or more of the meal, the exercise, the insulin delivery, and the dosage of medication predicted at the one or more different times in the future.
14 . The system of claim 11 , wherein the neural network comprises a recurrent neural network including a plurality of hourly long short-term memory (LSTM) units.
15 . The system of claim 14 , wherein each of the hourly LSTM units is configured to output a forecast hourly average glucose level associated with a respective hourly interval as a function of one or more events predicted to occur within the respective hourly interval and an input from one or more preceding hourly LSTM units.
16 . The system of claim 15 , wherein the function is based at least in part on a relationship between a respective subset of historical measurement data for the patient corresponding to the respective hourly interval and a respective subset of historical event data for the patient corresponding to the respective hourly interval.
17 . A method of monitoring a glucose level of a patient, the method comprising:
determining an hourly forecasting model for the patient based at least in part on a relationship between historical glucose measurement data for the patient and historical event data associated with the patient;
obtaining, from a glucose sensing arrangement inserted at least partially subcutaneously in the patient, current glucose measurement data for the patient;
predicting one or more events likely to influence the glucose level of the patient at one or more different times in the future based at least in part on the historical event data associated with the patient;
determining a plurality of hourly forecast average glucose values for the patient in the future based at least in part on the current glucose measurement data and the one or more events using the hourly forecasting model associated with the patient, wherein the hourly forecasting model comprises a neural network including a plurality of cells corresponding to a plurality of different time periods with cells in the plurality of cells sequentially coupled such that a first cell in the plurality of cells provides a forecast value of the plurality of hourly forecast average values to a subsequent cell in the plurality of cells; and
displaying, on a display device operatively coupled to the sensing arrangement at least partially inserted subcutaneously in the patient, a graphical representation of the plurality of hourly forecast average glucose values in the future.
18 . The method of claim 17 , wherein determining the hourly forecasting model comprises generating a recurrent neural network comprising the plurality of cells corresponding to the plurality of time periods by determining, for each respective cell, a function for calculating an average glucose level over the respective time period based at least in part on a relationship between a subset of the historical glucose measurement data corresponding to the respective time period and a subset of the historical event data corresponding to the respective time period.
19 . The method of claim 17 , further comprising obtaining a current operational context, wherein predicting the one or more events comprises predicting the one or more events based at least in part on a relationship between the historical event data associated with the patient and the current operational context.
20 . The method of claim 17 , further comprising obtaining, from a second sensing arrangement, a current operational context, wherein determining the plurality of hourly forecast average glucose values comprises determining the plurality of hourly forecast average glucose values based at least in part on the current glucose measurement data, the current operational context, and the one or more events using the hourly forecasting model associated with the patient.