ReplayBG: A Digital Twin-Based Methodology to Identify a Personalized Model From Type 1 Diabetes Data and Simulate Glucose Concentrations to Assess Alternative Therapies.

OBJECTIVE: Design and assessment of new therapies for type 1 diabetes (T1D) management can be greatly facilitated by in silico simulations. The ReplayBG simulation methodology here proposed allows "replaying" the scenario behind data already collected by simulating the glucose concentration obtained in response to alternative insulin/carbohydrate therapies and evaluate their efficacy leveraging the concept of digital twin. METHODS: ReplayBG is based on two steps. First, a personalized model of glucose-insulin dynamics is identified using insulin, carbohydrate, and continuous glucose monitoring (CGM) data. Then, this model is used to simulate the glucose concentration that would have been obtained by "replaying" the same portion of data using a different therapy. The validity of the methodology was evaluated on 100 virtual subjects using the UVa/Padova T1D Simulator (T1DS). In particular, the glucose concentration traces simulated by ReplayBG are compared with those provided by T1DS in five different scenarios of insulin and carbohydrate treatment modifications. Furthermore, we compared ReplayBG with a state-of-the-art methodology for the scope. Finally, two case studies using real data are also presented. RESULTS: ReplayBG simulates with high accuracy the effect of the considered insulin and carbohydrate treatment alterations, performing significantly better than state-of-art method in almost all considered situations. CONCLUSION: ReplayBG proved to be a reliable and robust tool to retrospectively explore the effect of new treatments for T1D on the glucose dynamics. It is freely available as open source software at https://github.com/gcappon/replay-bg. SIGNIFICANCE: ReplayBG offers a new approach to preliminary evaluate new therapies for T1D management before clinical trials.

Individualized Models for Glucose Prediction in Type 1 Diabetes: Comparing Black-Box Approaches to a Physiological White-Box One.

OBJECTIVE: Accurate blood glucose (BG) prediction are key in next-generation tools for type 1 diabetes (T1D) management, such as improved decision support systems and advanced closed-loop control. Glucose prediction algorithms commonly rely on black-box models. Large physiological models, successfully adopted for simulation, were little explored for glucose prediction, mostly because their parameters are hard to individualize. In this work, we develop a BG prediction algorithm based on a personalized physiological model inspired by the UVA/Padova T1D Simulator. Then we compare white-box and advanced black-box personalized prediction techniques. METHODS: A personalized nonlinear physiological model is identified from patient data through a Bayesian approach based on Markov Chain Monte Carlo technique. The individualized model was integrated within a particle filter (PF) to predict future BG concentrations. The black-box methodologies considered are non-parametric models estimated via gaussian regression (NP), three deep learning methods: long-short-term-memory (LSTM), gated recurrent unit (GRU), temporal convolutional networks (TCN), and a recursive autoregressive with exogenous input model (rARX). BG forecasting performances are assessed for several prediction horizons (PH) on 12 individuals with T1D, monitored in free-living conditions under open-loop therapy for 10 weeks. RESULTS: NP models provide the most effective BG predictions by achieving a root mean square error (RMSE), RMSE = 18.99 mg/dL, RMSE = 25.72 mg/dL and RMSE = 31.60 mg/dL, significantly outperforming: LSTM, GRU (for PH = 30 minutes), TCN, rARX, and the proposed physiological model for PH=30, 45 and 60 minutes. CONCLUSIONS: Black-box strategies remain preferable for glucose prediction even when compared to a white-box model with sound physiological structure and individualized parameters.

An Adaptive Model Predictive Controller to Address the Biovariability in Blood Clotting Response During Therapy With Warfarin.

OBJECTIVE: Effective dosing of anticoagulants aims to prevent blood clot formation while avoiding hemorrhages. This complex task is challenged by several disturbing factors and drug-effect uncertainties, requesting frequent monitoring and adjustment. Biovariability in drug absorption and action further complicates titration and calls for individualized strategies. In this paper, we propose an adaptive closed-loop control algorithm to assist in warfarin therapy management. METHODS: The controller was designed and tested in silico using an established pharmacometrics model of warfarin, which accounts for inter-subject variability. The control algorithm is an adaptive Model Predictive Control (a-MPC) that leverages a simplified patient model, whose parameters are updated with a Bayesian strategy. Performance was quantitatively evaluated in simulations performed on a population of virtual subjects against an algorithm reproducing medical guidelines (MG) and an MPC controller available in the literature (l-MPC). RESULTS: The proposed a-MPC significantly (p 0.05) lowers rising time (2.8 vs. 4.4 and 11.2 days) and time out of range (3.3 vs. 7.2 and 12.9 days) with respect to both MG and l-MPC, respectively. Adaptivity grants a significantly (p 0.05) lower number of subjects reaching unsafe INR values compared to when this feature is not present (8.9% vs.15% of subjects presenting an overshoot outside the target range and 0.08% vs. 0.28% of subjects reaching dangerous INR values). CONCLUSION: The a-MPC algorithm improve warfarin therapy compared to the benchmark therapies. SIGNIFICANCE: This in-silico validation proves effectiveness of the a-MPC algorithm for anticoagulant administration, paving the way for clinical testing.

Data Gap Modeling in Continuous Glucose Monitoring Sensor Data.

Continuous glucose monitoring (CGM) sensors are minimally-invasive sensors used in diabetes therapy to monitor interstitial glucose concentration. The measurements are collected almost continuously (e.g. every 5 min) and permit the detection of dangerous hypo/hyperglycemic episodes. Modeling the various error components affecting CGM sensors is very important (e.g., to generate realistic scenarios for developing and testing CGM-based applications in type 1 diabetes simulators). In this work we focus on data gaps, which are portions of missing data due to a disconnection or a temporary sensor error. A dataset of 167 adults monitored with the Dexcom (San Diego, CA) G6 sensor is considered. After the evaluation of some statistics (the number of gaps for each sensor, the gap distribution over the monitoring days and the data gap durations), we develop a two-state Markov model to describe such statistics about data gap occurrence. Statistics about data gaps are compared between real data and simulated data generated by the model with a Monte Carlo simulation. Results show that the model describes quite accurately the occurrence and the duration of data gaps observed in real data.

Machine-Learning Based Model to Improve Insulin Bolus Calculation in Type 1 Diabetes Therapy.

OBJECTIVE: This paper aims at proposing a new machine-learning based model to improve the calculation of mealtime insulin boluses (MIB) in type 1 diabetes (T1D) therapy using continuous glucose monitoring (CGM) data. Indeed, MIB is still often computed through the standard formula (SF), which does not account for glucose rate-of-change ( ∆G), causing critical hypo/hyperglycemic episodes. METHODS: Four candidate models for MIB calculation, based on multiple linear regression (MLR) and least absolute shrinkage and selection operator (LASSO) are developed. The proposed models are assessed in silico, using the UVa/Padova T1D simulator, in different mealtime scenarios and compared to the SF and three ∆G-accounting variants proposed in the literature. An assessment on real data, by retrospectively analyzing 218 glycemic traces, is also performed. RESULTS: All four tested models performed better than the existing techniques. LASSO regression with extended feature-set including quadratic terms (LASSO Q) produced the best results. In silico, LASSO Q reduced the error in estimating the optimal bolus to only 0.86 U (1.45 U of SF and 1.36-1.44 U of literature methods), as well as hypoglycemia incidence (from 44.41% of SF and 44.60-45.01% of literature methods, to 35.93%). Results are confirmed by the retrospective application to real data. CONCLUSION: New models to improve MIB calculation accounting for CGM- ∆G and easy-to-measure features can be developed within a machine learning framework. Particularly, in this paper, a new LASSO Q model was developed, which ensures better glycemic control than SF and other literature methods. SIGNIFICANCE: MIB dosage with the proposed LASSO Q model can potentially reduce the risk of adverse events in T1D therapy.

Model-Based Detection and Classification of Insulin Pump Faults and Missed Meal Announcements in Artificial Pancreas Systems for Type 1 Diabetes Therapy.

OBJECTIVE: The artificial pancreas (AP) is an innovative closed-loop system for type 1 diabetes therapy, in which insulin is infused by portable pumps and insulin dosage is modulated by a control algorithm on the basis of the measurements collected by continuous glucose monitoring (CGM) sensors. AP systems safety and effectiveness could be affected by several technological and user-related issues, among which insulin pump faults and missed meal announcements. This work proposes an algorithm to detect in real-time these two types of failure. METHODS: The algorithm works as follows. First, a personalized autoregressive moving-average model with exogenous inputs is identified using historical data of the patient. Second, the algorithm is used in real time to predict future CGM values. Then, alarms are triggered when the difference between predicted vs measured CGM values is higher than opportunely set thresholds. In addition, by using two different set of parameters, the algorithm is able to distinguish the two types of failures. The algorithm was developed and assessed in silico using the latest version of the FDA-approved Padova/UVa T1D simulator. RESULTS: The algorithm showed a sensitivity of  âˆ¼ 81.3% on average when detecting insulin pump faults with  âˆ¼ 0.15 false positives per day on average. Missed meal announcements were detected with a sensitivity of  âˆ¼ 86.8% and 0.15 FP/day. CONCLUSION: The presented method is able to detect insulin pump faults and missed meal announcements in silico, correctly distinguishing one from another. SIGNIFICANCE: The method increases the safety of AP systems by providing prompt alarms to the diabetic subject and effectively discriminating pump malfunctioning from user errors.

Modeling the error of factory-calibrated continuous glucose monitoring sensors: application to Dexcom G6 sensor data.

Minimally-invasive continuous glucose monitoring (CGM) sensors are used in diabetes therapy to monitor interstitial glucose (IG) concentration almost continuously (e.g. every 5 min) and detect/predict dangerous hypo/hyperglycemic episodes. When compared with frequent blood glucose (BG) concentration references, CGM measurements are unavoidably affected by error. Models of the CGM error can be important in several applications, e.g. for testing in simulation the safety and effectiveness of CGM-based artificial pancreas algorithms. In this work, we model the error of the Dexcom G6, a CGM sensor that recently entered the market and does not require in vivo calibrations. The dataset includes CGM and BG data collected in 11 subjects wearing two Dexcom G6 sensors in parallel. The model is derived applying a methodology to dissect and model 3 main CGM error components: BG-to-IG kinetics, calibration error and measurement noise. An aspect of novelty of the method is its capability of handling factory-calibrated CGM sensor data. Results of model identification show that the time-variability of sensor calibration error during the sensor lifetime (10 days) can be well represented by a regression model with time-variant parameters described by 2nd-order polynomials in time.

A Bayesian Framework to Identify Type 1 Diabetes Physiological Models Using Easily Accessible Patient Data.

Mathematical physiological models of type 1 diabetes (T1D) glucose-insulin dynamics have been of great help in designing and preliminary assessing new algorithm for glucose control. Derivation of models at the individual level is however difficult because of identifiability issues. Recently, fitting these models against data of real patients with T1D has been made possible by both the use of Bayesian estimation techniques and the availability of individual datasets including plasma glucose and insulin concentration samples gathered in clinical protocols. The aim of this work is to make a step further and develop a methodology able to estimate the parameters of T1D physiological models using easily accessible data only, i.e. continuous glucose monitoring (CGM) sensor, carbohydrate intakes (CHO), and exogenous insulin infusion (I) data. The methodology is tested on synthetic data of 100 patients generated by a composite model of glucose-insulin dynamics. To solve identifiability problems, a Bayesian approach numerically implemented by Markov Chain Monte Carlo (MCMC) has been used to obtain point estimates and confidence intervals of model unknown parameters exploiting a priori knowledge available from the literature. Results show goodness of model fit and acceptable precision of parameter estimates. The methodology is also successful in reconstructing of "non-accessible" glucose-insulin fluxes, i.e. glucose rate of appearance and plasma insulin. These preliminary results encourage further development of this framework and its assessment in more challenging setups.

Retrofitting Real-Life Dexcom G5 Data.

BACKGROUND: We proposed in 2014 a retrofitting algorithm to retrospectively increase the accuracy of continuous glucose monitoring (CGM) data by using some blood glucose (BG) measurements. The method proved effective on Dexcom SEVEN Plus when about 10 highly accurate YSI measurements/session were available. In this study, we test the method on Dexcom G5 sensor in a more realistic setup, where only five capillary BG measurements (self-monitoring blood glucose [SMBG]) per 12 h-session are available. Furthermore, we investigate how accuracy is affected by the number of BG measurements. METHOD: The algorithm was tested in 51 adults and 46 adolescents studied for 7 days with Dexcom G5. Each patient also underwent an ∼12-h hospital admission where frequent SMBG and YSI measurements were collected. First, five SMBGs per 12-h session were used to retrofit the CGM. Then, we varied the number of SMBGs provided to the method from 2 to 10 per 12-h session. RESULT: Retrofitted CGM traces with five SMBGs per 12-h session have lower mean absolute difference than original CGM, reduced from 16.2 to 10.7 mg/dL (P < 0.001) in adults and from 17.6 to 11.5 mg/dL (P < 0.001) in adolescents, and mean absolute relative difference is reduced from 9.0% to 6.4% (P < 0.001) in adults and from 10.3% to 6.8% (P < 0.001) in adolescents. Reducing the number of BG measurements reduces improvement in the accuracy from >30% with 10 SMBGs per 12-h session to <16% with 2 SMBGs/day. CONCLUSION: The retrofitting method retrospectively improves the accuracy of CGM data, even if applied to one of the most accurate CGM sensors currently available on the market.