US-20260123895-A1 - BAYESIAN FRAMEWORK FOR PERSONALIZED MODEL IDENTIFICATION AND PREDICTION OF FUTURE BLOOD GLUCOSE IN TYPE 1 DIABETES USING EASILY ACCESSIBLE PATIENT DATA

US20260123895A1US 20260123895 A1US20260123895 A1US 20260123895A1US-20260123895-A1

Abstract

A method of predicting future blood glucose concentrations of an individual patient includes: selecting an individualized nonlinear physiological model of glucose-insulin dynamics, the selected model having a plurality of model parameters whose values are to be determined; estimating values for each of the model parameters in the plurality of model parameters, a first subset of the model parameters having values estimated from a priori population data and a second subset of the model parameters having values personalized for the individual patient by applying a parameter estimation technique to a priori information and data for the individual patient to obtain a posteriori information; and; applying a nonlinear prediction technique to the selected model using the estimated values for each of the model parameters to obtain a predicted blood glucose concentration of the individual patient at a future time.

Inventors

Giacomo CAPPON
Andrea Facchinetti
Giovanni Sparacino
Simone Del Favero

Assignees

DEXCOM, INC.

Dates

Publication Date: 20260507
Application Date: 20260105

Claims (20)

1 . A method of predicting future blood glucose concentrations of an individual patient having a continuous glucose monitoring (CGM) sensor, the method comprising: selecting, by one or more processors, an individualized physiological model having a plurality of model parameters; estimating, by the one or more processors, population-based values of a first subset of the plurality of model parameters from a population data, wherein the estimated population-based values of the first subset relate to at least one of insulin absorption or glucose absorption; estimating, by the one or more processors, personalized values of a second subset of the model parameters by applying a parameter estimation technique to each parameter of the second subset and data for the individual patient, wherein the data for the individual patient comprises insulin infusion information for the individual patient and carbohydrate intake information for the individual patient; selecting, by the one or more processors, a probability function based on past glucose measurements of the individual patient from the CGM sensor, the estimated population-based values of the first subset of the plurality of model parameters, and the estimated personalized values of the second subset of the plurality of model parameters; applying, by the one or more processors, a particle filter to the selected individualized physiological model using the probability function to obtain a predicted blood glucose concentration of the individual patient at a future time; generating, by the one or more processors, a proactive alert based on the predicted blood glucose concentration indicating upcoming dysglycemia; and causing, by the one or more processors, the proactive alert to be output on a device associated with the individual patient.
2 . The method of claim 1 , wherein the proactive alert comprises a recommended treatment to mitigate the upcoming dysglycemia indicated by the predicted blood glucose concentration.
3 . The method of claim 1 , wherein the device associated with the individual patient comprises the one or more processors.
4 . The method of claim 1 , wherein the causing comprises communicating, by the one or more processors, the proactive alert to the device associated with the individual patient.
5 . The method of claim 1 , wherein the parameter estimation technique is a Markov Chain Monte Carlo (MCMC) Bayesian estimator.
6 . The method of claim 1 , wherein the parameter estimation technique is selected from the group consisting of a maximum a posteriori technique, a maximum likelihood technique and a prediction error minimization technique.
7 . The method of claim 1 , wherein the selected individualized physiological model includes a subcutaneous insulin absorption sub-model, an oral glucose absorption sub-model and a glucose-insulin kinetics model.
8 . The method of claim 7 , wherein the subcutaneous insulin absorption sub-model includes a first model parameter specifying a value of insulin distribution and a second model parameter specifying a delay in an appearance of insulin in a first compartment, the first and second model parameters being in the first subset of the plurality of model parameters having the estimated population-based values.
9 . The method of claim 8 , wherein the subcutaneous insulin absorption sub-model includes a third model parameter specifying a diffusion rate constant from a first to a second compartment and a fourth model parameter specifying a rate constant of subcutaneous insulin absorption from the second compartment to plasma, the third and fourth model parameters being in the second subset of the plurality of model parameters having the estimated personalized values.
10 . The method of claim 1 , wherein the selected individualized physiological model is a maximal physiological model in which selected ones of the plurality of model parameters are neglected.
11 . The method of claim 1 , further comprising augmenting the selected individualized physiological model with a residual error model to describe residual modeling errors.
12 . The method of claim 11 , wherein the residual modeling errors are modeled as an Auto regressive Integrated Moving Average (ARIMA) of order (5,5,1).
13 . The method of claim 12 , wherein the residual error model has model parameters estimated using a predicted error method (PEM).
14 . The method of claim 1 , wherein the predicted blood glucose concentration is predicted at a plurality of different times.
15 . A computer-program product comprising a non-transitory computer-usable medium having computer-readable program code embodied therein, the computer-readable program code adapted to be executed to implement a method of predicting future blood glucose concentrations of an individual patient having a continuous glucose monitoring (CGM) sensor, the method comprising: selecting an individualized physiological model having a plurality of model parameters; estimating population-based values of a first subset of the plurality of model parameters from a population data, wherein the estimated population-based values of the first subset relate to at least one of insulin absorption or glucose absorption; estimating personalized values of a second subset of the model parameters by applying a parameter estimation technique to each parameter of the second subset and data for the individual patient, wherein the data for the individual patient comprises insulin infusion information for the individual patient and carbohydrate intake information for the individual patient; selecting a probability function based on past glucose measurements of the individual patient from the CGM sensor, the estimated population-based values of the first subset of the plurality of model parameters, and the estimated personalized values of the second subset of the plurality of model parameters; applying a particle filter to the selected individualized physiological model using the probability function to obtain a predicted blood glucose concentration of the individual patient at a future time; generating a proactive alert based on the predicted blood glucose concentration indicating upcoming dysglycemia; and causing the proactive alert to be output on a device associated with the individual patient.
16 . The computer-program product of claim 15 , wherein the parameter estimation technique is a Markov Chain Monte Carlo (MCMC) Bayesian estimator.
17 . The computer-program product of claim 16 , wherein the parameter estimation technique is selected from the group consisting of a maximum a posteriori technique, a maximum likelihood technique and a prediction error minimization technique.
18 . The computer-program product of claim 16 , wherein the selected individualized physiological model includes a subcutaneous insulin absorption sub-model, an oral glucose absorption sub-model and a glucose-insulin kinetics model.
19 . The computer-program product of claim 18 , wherein the subcutaneous insulin absorption sub-model includes a first model parameter specifying a value of insulin distribution and a second model parameter specifying a delay in an appearance of insulin in a first compartment, the first and second model parameters being in the first subset of the plurality of model parameters having the estimated population-based values.
20 . The computer-program product of claim 19 , wherein the subcutaneous insulin absorption sub-model includes a third model parameter specifying a diffusion rate constant from a first to a second compartment and a fourth model parameter specifying a rate constant of subcutaneous insulin absorption from the second compartment to plasma, the third and fourth model parameters being in the second subset of the plurality of model parameters having the estimated personalized values.

Description

INCORPORATION BY REFERENCE TO RELATED APPLICATION Any and all priority claims identified in the Application Data Sheet, or any correction thereto, are hereby incorporated by reference under 37 CFR 1.57. This application is a continuation of U.S. patent application Ser. No. 17/454,693, filed Nov. 12, 2021, which claims the benefit of U.S. Provisional Patent Application No. 63/141,345, filed Jan. 25, 2021. This application relates to U.S. application Ser. Nos. 17/112,870, 17/112,828, and 17/112,814, all filed Dec. 4, 2020. The aforementioned applications are incorporated by reference herein in their entirety, and are hereby expressly made a part of this specification. BACKGROUND Type 1 diabetes (TID) is a chronical autoimmune disease caused by the progressive destruction of beta cells in the pancreas, which leads to the inability of producing endogenous insulin by the organism. As a result, blood glucose (BG) concentration tends to exceed the hyperglycemic threshold (BG>180 mg/dl), a situation that, if frequent and prolonged, could lead to several serious cardiovascular long-term complications, as well as nephropathy and neuropathy. To reduce BG levels, administration of exogenous insulin several times a day is necessary. Unfortunately, excessive exogenous insulin dosing could lead patients to hypoglycemia, i.e., BG<70 mg/dl, which is dangerous even in the short-term since it could cause fainting, light-headiness, coma and even death. Effective T1D treatment relies on BG frequent monitoring, made through either the classic fingerstick device or more modern minimally invasive continuous glucose monitoring (CGM) sensors, and is far from being trivial. Indeed, T1D management represents, from a patient perspective, a life-long learning process to understand how several everyday factors (e.g., illness, diet, and physical activity) affect BG levels and how interventions (e.g., rescue carbohydrate intake and, of course, insulin administration) can be used to keep BG in the safe range. In this context, many efforts have been made by the research community to provide new tools able to help patients with TID. Among them, CGM-based algorithms able to predict future BG concentration in real-time have the potential to significantly improve TID therapy efficacy by enabling proactive therapeutic decisions based on the expected future glucose levels, rather than the current one. SUMMARY In a first aspect, a method is provided of predicting future blood glucose concentrations of an individual patient, comprising: selecting an individualized nonlinear physiological model of glucose-insulin dynamics, the selected model having a plurality of model parameters whose values are to be determined; estimating values for each of the model parameters in the plurality of model parameters, a first subset of the model parameters having values estimated from a priori population data and a second subset of the model parameters having values personalized for the individual patient by applying a parameter estimation technique to a priori information and data for the individual patient to obtain a posteriori information; and; applying a nonlinear prediction technique to the selected model using the estimated values for each of the model parameters to obtain a predicted blood glucose concentration of the individual patient at a future time. In an embodiment of the first aspect, the parameter estimation technique is a Markov Chain Monte Carlo (MCMC) Bayesian estimator. In an embodiment of the first aspect, the parameter estimation technique is selected from the group consisting of a maximum a posteriori technique, a maximum likelihood technique and a prediction error minimization technique. In an embodiment of the first aspect, the nonlinear prediction technique is a particle filter. In an embodiment of the first aspect, the nonlinear prediction technique is a particle filter. In an embodiment of the first aspect, the nonlinear prediction technique is selected from the group consisting of an extended Kalman filter technique and an unscented Kalman filter technique. In an embodiment of the first aspect, the nonlinear prediction technique is selected from the group consisting of an extended Kalman filter technique and an unscented Kalman filter technique. In an embodiment of the first aspect, the selected individualized nonlinear physiological model includes a subcutaneous insulin absorption sub-model, an oral glucose absorption sub-model and a glucose-insulin kinetics model. In an embodiment of the first aspect, the selected individualized nonlinear physiological model uses past blood glucose concentration levels, carbohydrate intake information and exogenous insulin data of the individual patient as inputs. In an embodiment of the first aspect, the subcutaneous insulin absorption sub-model includes a first model parameter specifying a value of insulin distribution and a second model parameter specifying a delay in an appearance of insulin in a fi