Predictive monitoring of the glucose-insulin endocrine metabolic regulatory system
There is a need for more effective and efficient predictive data analysis, such as more effective and efficient data analysis solutions for performing predictive monitoring of the glucose-insulin endocrine metabolic regulatory system. Certain embodiments utilize systems, methods, and computer program products that perform predictive data analysis by utilizing at least one of glucose surge excursion detections, steady-state glucose-insulin machine learning models, and parameter space refinement machine learning models.
1. A computer-implemented method comprising:
receiving, by one or more processors and from a continuous glucose monitoring (CGM) device, CGM data for an end-user;
determining, by the one or more processors, an excursion start time of the CGM data based at least in part on a plurality of excursion initiation probabilities for a plurality of temporal units associated with the CGM data;
determining, by the one or more processors, an excursion end time based at least in part on a plurality of excursion termination probabilities for a subset of the plurality of temporal units that occur after a predefined time interval following the excursion start time, wherein each excursion termination probability of the plurality of excursion termination probabilities is adjusted by an adjustment factor configured to:
(a) reward a latency of a corresponding temporal unit in an ordering of the subset of the plurality of temporal units, and
(b) penalize a precedence of the corresponding temporal unit in the ordering of the subset of the plurality of temporal units;
identifying, by the one or more processors, a time interval within the CGM data that corresponds to a glucose surge excursion based at least in part on the excursion start time and the excursion end time, wherein the time interval excludes the excursion start time and the excursion end time;
determining, by the one or more processors and based at least in part on a machine learning model that is trained using ground-truth data corresponding to the glucose surge excursion, one or more glucose-insulin predictions based at least in part on the glucose surge excursion; and
based at least in part on the one or more glucose-insulin predictions, performing an automated exogenous insulin injection into the end-user with an automated insulin delivery computing entity.
2. The computer-implemented method of claim 1 , wherein:
one or more of the plurality of excursion initiation probabilities are determined based at least in part on one or more of (i) a first comparison between a neighboring CGM moving average and a neighboring CGM moving average threshold, (ii) a first comparison between a CGM first derivative approximation and a CGM first derivative approximation threshold, or (iii) a first comparison between a CGM z-score and a CGM z-score threshold;
one or more of the plurality of excursion termination probabilities are determined based at least in part on one or more of (i) a second comparison between the neighboring CGM moving average and the neighboring CGM moving average threshold, (ii) a second comparison between the CGM first derivative approximation and the CGM first derivative approximation threshold, or (iii) a second comparison between the CGM z-score and the CGM z-score threshold; and
the neighboring CGM moving average threshold is determined based at least in part on a CGM measurement for a temporal unit of the plurality of temporal units and one or more CGM measurements for one or more temporal units of the plurality of temporal units preceding the temporal unit.
3. The computer-implemented method of claim 1 , wherein one or more of the plurality of excursion initiation probabilities are determined based at least in part on one or more user-supplied meal session initiation indicators.
4. The computer-implemented method of claim 1 , wherein one or more of the plurality of excursion termination probabilities are determined based at least in part on one or more user-supplied meal session termination indicators.
5. The computer-implemented method of claim 1 , wherein:
the plurality of excursion initiation probabilities is determined by providing CGM measurements to the machine learning model,
the machine learning model is configured to process the CGM measurements to generate the plurality of excursion initiation probabilities, and
the machine learning model is trained using the ground-truth data, wherein the ground-truth data is determined based at least in part on one or more user-supplied meal session initiation indicators.
6. The computer-implemented method of claim 1 , wherein:
the plurality of excursion termination probabilities is determined by providing CGM measurements associated with the subset of the plurality of temporal units to the machine learning model,
the machine learning model is configured to process the CGM measurements to generate the plurality of excursion termination probabilities, and
the machine learning model is trained using the ground-truth data, wherein the ground-truth data is determined based at least in part on one or more user-supplied meal session termination indicators.
7. The computer-implemented method of claim 1 , wherein the one or more glucose-insulin predictions comprise an insulin sensitivity prediction.
8. The computer-implemented method of claim 2 , wherein the neighboring CGM moving average threshold is 2.5 milligrams per deciliter per minute, the CGM first derivative approximation threshold is 2.5 milligrams per deciliter per minute, the CGM z-score threshold is 1.0, or any combination thereof.
9. The computer-implemented method of claim 1 , wherein the machine learning model comprises a steady-state glucose-insulin prediction machine learning model, a glucose-biased glucose-insulin predication machine learning model, a hybrid glucose-insulin predication machine learning model, an excursion termination detection machine learning model, a parameter space refinement machine learning model, or any combination thereof.
10. The computer-implemented method of claim 1 , wherein the machine learning model is configured using one or more quasi-global parameter optimization operations.
11. The computer-implemented method of claim 1 , wherein determining the excursion end time comprises:
determining one or more potential excursion end times based at least in part on the plurality of excursion initiation probabilities;
determining an excursion probability for a potential excursion end time of the one or more potential excursion end times, wherein the excursion probability is based at least in part on a temporal deviation between the excursion start time and the potential excursion end time; and
determining the excursion end time based at least in part on the excursion probability.
12. A system comprising one or more memories, one or more processors communicatively coupled to the one or more memories, and an automated insulin delivery computing entity, the one or more processors configured to:
receive, from a continuous glucose monitoring (CGM) device, CGM data for an end-user;
determine an excursion start time of the CGM data based at least in part on a plurality of excursion initiation probabilities for a plurality of temporal units associated with the CGM data;
determine an excursion end time based at least in part on a plurality of excursion termination probabilities for a subset of the plurality of temporal units that occur after a predefined time interval following the excursion start time, wherein each excursion termination probability of the plurality of excursion termination probabilities is adjusted by an adjustment factor configured to:
(a) reward a latency of a corresponding temporal unit in an ordering of the subset of the plurality of temporal units, and
(b) penalize a precedence of the corresponding temporal unit in the ordering of the subset of the plurality of temporal units;
identify a time interval within the CGM data that corresponds to a glucose surge excursion based at least in part on the excursion start time and the excursion end time, wherein the time interval excludes the excursion start time and the excursion end time;
determine, based at least in part on a machine learning model that is trained using ground-truth data corresponding to the glucose surge excursion, one or more glucose-insulin predictions based at least in part on the glucose surge excursion; and
based at least in part on the one or more glucose-insulin predictions, perform an automated exogenous insulin injection into the end-user with the automated insulin delivery computing entity.
13. The system of claim 12 , wherein one or more of the plurality of excursion initiation probabilities are determined based at least in part on one or more user-supplied meal session initiation indicators.
14. The system of claim 12 , wherein one or more of the plurality of excursion termination probabilities are determined based at least in part on one or more user-supplied meal session termination indicators.
15. The system of claim 12 , wherein:
the plurality of excursion initiation probabilities is determined by providing CGM measurements to the machine learning model,
the machine learning model is configured to process the CGM measurements to generate the plurality of excursion initiation probabilities, and
the machine learning model is trained using the ground-truth data, wherein the ground-truth data is determined based at least in part on one or more user-supplied meal session initiation indicators.