A Real-Time Continuous Glucose Monitoring-Based Algorithm to Trigger Hypotreatments to Prevent/Mitigate Hypoglycemic Events

A strategy based on integer programming for optimal dosing and timing of preventive hypoglycemic treatments in type 1 diabetes management

BACKGROUND AND OBJECTIVES

One of the major problems related to type 1 diabetes (T1D) management is hypoglycemia, a condition characterized by low blood glucose levels and responsible for reduced quality of life and increased mortality. Fast-acting carbohydrates, also known as hypoglycemic treatments (HT), can counteract this event. In the literature, dosage and timing of HT are usually based on heuristic rules. In the present work, we propose an algorithm for mitigating hypoglycemia by suggesting preventive HT consumption, with dosages and timing determined by solving an optimization problem.

METHODS

By leveraging integer programming and linear inequality constraints, the algorithm can bind the amount of suggested carbohydrates to standardized quantities (i.e., those available in "off-the-shelf" HT) and the minimal distance between consecutive suggestions (to reduce the nuisance for patients).

RESULTS

The proposed method was tested in silico and compared with competitor algorithms using the UVa/Padova T1D simulator. At the cost of a slight increase of HT consumed per day, the proposed algorithm produces the lowest median and interquartile range of the time spent in hypoglycemia, with a statistically significant improvement over most competitor algorithms. Also, the average number of hypoglycemic events per day is reduced to 0 in median.

CONCLUSIONS

Thanks to its positive performances and reduced computational burden, the proposed algorithm could be a candidate tool for integration in a DSS aimed at improving T1D management.

Risk of hypoglycemia in type 1 diabetes management: An in-silico sensitivity analysis to assess and rank the quantitative impact of different behavioral factors

BACKGROUND AND OBJECTIVE

In type 1 diabetes (T1D), a quantitative evaluation of the impact on hypoglycemia of suboptimal therapeutic decision (e.g. incorrect estimation of the ingested carbohydrates, inaccurate insulin timing, etc) is unavailable. Clinical trials to measure sensitivity to patient actions would be expensive, exposed to confounding factors and risky for the participants. In this work, a T1D patient decision simulator (T1D-PDS), realistically reproducing blood glucose dynamics in a large virtual population, is used to perform extensive in-silico trials and the so-derived data employed to implement a sensitivity analysis that ranks different behavioral factors based on their impact on a clinically meaningful parameter, the time below range (TBR).

METHODS

Eleven behavioral factors impacting on hypoglycemia are considered. The T1D-PDS was used to perform multiple 2-week simulations involving 100 adults, by testing about 3500 different perturbations for nominal behavior. A local linear approximation of the function linking the TBR and the factors were computed to derive sensitivity indices (SIs), quantifying the impact of each factor on TBR variations.

RESULTS

The obtained ranking quantifies importance of factors w.r.t. the others. Factors apparently related to hypoglycemia were correctly placed on the top of the ranking, including systematic (SI=2.05%) and random (SI=1.35%) carb-counting error, hypotreatment dose (SI=-1.21%), insulin bolus time w.r.t. mealtime (SI=1.09%).

CONCLUSIONS

The obtained SIs allowed to rank behavioral factors based on their impact on TBR. The behavioral factors identified as most influential can be prioritized in patient training.

drCORRECT: An Algorithm for the Preventive Administration of Postprandial Corrective Insulin Boluses in Type 1 Diabetes Management

BACKGROUND

In type 1 diabetes therapy, precise tuning of postprandial corrective insulin boluses (CIBs) is crucial to mitigate hyperglycemia without inducing dangerous hypoglycemic events. Several heuristic formulas accounting for continuous glucose monitoring (CGM) trend have been proposed in the literature. However, these formulas suggest a lot of quantized CIB adjustments, and they lack personalization.

METHOD

drCORRECT algorithm proposed in this work employs a patient-specific time parameter and the "dynamic risk" (DR) measure to determine postprandial CIB suggestion. The expected benefits include the reduction of time in hyperglycemia, thanks to the preventive action exploited through DR. drCORRECT has been assessed retrospectively vs the literature methods proposed by Aleppo et al (AL), Bruttomesso et al (BR), and Ziegler et al (ZI) using a data set of 49 CGM daily traces recorded in free-living conditions. Retrospective evaluation of the algorithms is made possible by the use of ReplayBG, a digital twin-based tool that allows assessing alternative insulin therapies on already collected glucose data. Efficacy in terms of glucose control was measured by temporal, risk indicators, and dedicated hyperglycemic/hypoglycemic events metrics.

RESULTS

drCORRECT significantly reduces time spent in hyperglycemia when compared with AL and BR (33.52 [24.16, 39.89]% vs 39.76 [22.54, 48.15]% and 36.32 [26.91, 45.93]%, respectively); significantly reduces daily injected insulin (5.97 [3.80, 8.06] U vs 7.5 [5.21, 10.34] U), glycemia risk index (38.78 [26.58, 55.39] vs 40.78 [27.95, 70.30]), and time spent in hypoglycemia (0.00 [0.00, 1.74]% vs 0.00 [0.00, 10.23]%) when compared with ZI, resulting overall in a safer strategy.

CONCLUSIONS

The proposed drCORRECT algorithm allows preventive actions thanks to the personalized timing configuration and the introduction of the innovative DR-based CIB threshold, proving to be a valid alternative to the available heuristic literature methods.

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.

The importance of interpreting machine learning models for blood glucose prediction in diabetes: an analysis using SHAP

Machine learning has become a popular tool for learning models of complex dynamics from biomedical data. In Type 1 Diabetes (T1D) management, these models are increasingly been integrated in decision support systems (DSS) to forecast glucose levels and provide preventive therapeutic suggestions, like corrective insulin boluses (CIB), accordingly. Typically, models are chosen based on their prediction accuracy. However, since patient safety is a concern in this application, the algorithm should also be physiologically sound and its outcome should be explainable. This paper aims to discuss the importance of using tools to interpret the output of black-box models in T1D management by presenting a case-of-study on the selection of the best prediction algorithm to integrate in a DSS for CIB suggestion. By retrospectively "replaying" real patient data, we show that two long-short term memory neural networks (LSTM) (named p-LSTM and np-LSTM) with similar prediction accuracy could lead to different therapeutic decisions. An analysis with SHAP-a tool for explaining black-box models' output-unambiguously shows that only p-LSTM learnt the physiological relationship between inputs and glucose prediction, and should therefore be preferred. This is verified by showing that, when embedded in the DSS, only p-LSTM can improve patients' glycemic control.

A Markov Model of Gap Occurrence in Continuous Glucose Monitoring Data for Realistic in Silico Clinical Trials

BACKGROUND AND OBJECTIVE

Continuous glucose monitoring (CGM) sensors measure interstitial glucose concentration every 1-5 min for days or weeks. New CGM-based diabetes therapies are often tested in in silico clinical trials (ISCTs) using diabetes simulators. Accurate models of CGM sensor inaccuracies and failures could help improve the realism of ISCTs. However, the modeling of CGM failures has not yet been fully addressed in the literature. This work aims to develop a mathematical model of CGM gaps, i.e., occasional portions of missing data generated by temporary sensor errors (e.g., excessive noise or artifacts).

METHODS

Two datasets containing CGM traces collected in 167 adults and 205 children, respectively, using the Dexcom G6 sensor (Dexcom Inc., San Diego, CA) were used. Four Markov models, of increasing complexity, were designed to describe three main characteristics: number of gaps for each sensor, gap distribution in the monitoring days, and gap duration. Each model was identified on a portion of each dataset (training set). The remaining portion of each dataset (real test set) was used to evaluate model performance through a Monte Carlo simulation approach. Each model was used to generate 100 simulated test sets with the same size as the real test set. The distributions of gap characteristics on the simulated test sets were compared with those observed on the real test set, using the two-sample Kolmogorov-Smirnov test and the Jensen-Shannon divergence.

RESULTS

A six-state Markov model, having two states to describe normal sensor operation and four states to describe gap occurrence, achieved the best results. For this model, the Kolmogorov-Smirnov test found no significant differences between the distribution of simulated and real gap characteristics. Moreover, this model obtained significantly lower Jensen-Shannon divergence values than the other models.

CONCLUSIONS

A Markov model describing CGM gaps was developed and validated on two real datasets. The model describes well the number of gaps for each sensor, the gap distribution over monitoring days, and the gap durations. Such a model can be integrated into existing diabetes simulators to realistically simulate CGM gaps in ISCTs and thus enable the development of more effective and robust diabetes management strategies.

Combined Use of Glucose-Specific Model Identification and Alarm Strategy Based on Prediction-Funnel to Improve Online Forecasting of Hypoglycemic Events

BACKGROUND

Advanced decision support systems for type 1 diabetes (T1D) management often embed prediction modules, which allow T1D people to take preventive actions to avoid critical episodes like hypoglycemia. Real-time prediction of blood glucose (BG) concentration relies on a subject-specific model of glucose-insulin dynamics. Model parameter identification is usually based on the mean square error (MSE) cost function, and the model is usually used to predict BG at a single prediction horizon (PH). Finally, a hypo-alarm is raised if the predicted BG crosses a threshold. This work aims to show that real-time hypoglycemia forecasting can be improved by leveraging: a glucose-specific mean square error (gMSE) cost function in model's parameters identification, and a "prediction-funnel," that is, confidence intervals (CIs) for multiple PHs, within the hypo-alarm-raising strategy.

METHODS

Autoregressive integrated moving average with exogenous input (ARIMAX) models are selected to illustrate the proposed solution (use of gMSE and prediction-funnel) and its assessment against the conventional approach (MSE and single PH). The gMSE penalizes the model misfit in unsafe BG ranges (e.g., hypoglycemia), and the prediction-funnel allows raising an alarm by monitoring if the CIs cross a suitable threshold. The algorithms were evaluated by measuring precision (P), recall (R), F1-score (F1), false positive per day (FP/day), and time gain (TG) on a real dataset collected in 11 T1D individuals.

RESULTS

The best performance is achieved exploiting both the gMSE and the prediction-funnel: P = 65%, R = 88%, F1 = 75%, FP/day = 0.29, and mean TG = 15 minutes.

CONCLUSIONS

The combined use of a glucose-specific metric and an alarm-raising strategy based on the prediction-funnel allows achieving a more effective and reliable hypoglycemia prediction algorithm.

Generation of post-meal insulin correction boluses in type 1 diabetes simulation models for in-silico clinical trials: More realistic scenarios obtained using a decision tree approach

BACKGROUND AND OBJECTIVE

In type 1 diabetes (T1D) research, in-silico clinical trials (ISCTs) notably facilitate the design/testing of new therapies. Published simulation tools embed mathematical models of blood glucose (BG) and insulin dynamics, continuous glucose monitoring (CGM) sensors, and insulin treatments, but lack a realistic description of some aspects of patient lifestyle impacting on glucose control. Specifically, to effectively simulate insulin correction boluses, required to treat post-meal hyperglycemia (BG > 180 mg/dL), the timing of the bolus may be influenced by subjects' behavioral attitudes. In this work, we develop an easily interpretable model of the variability of correction bolus timing observed in real data, and embed it into a popular simulation tool for ISCTs.

METHODS

Using data collected in 196 adults with T1D monitored in free-living conditions, we trained a decision tree (DT) model to classify whether a correction bolus is injected in a future time window, based on predictors collected back in time, related to CGM data, previous insulin boluses and subject's characteristics. The performance was compared to that of a logistic regression classifier with LASSO regularization (LC), trained on the same dataset. After validation, the DT was embedded within a popular T1D simulation tool and an ISCT was performed to compare the simulated correction boluses against those observed in a subset of data not used for model training.

RESULTS

The DT provided better classification performance (accuracy: 0.792, sensitivity: 0.430, specificity: 0.878, precision: 0.455) than the LC and presented good interpretability. The most predictive features were related to CGM (and its temporal variations), time since the last insulin bolus, and time of the day. The correction boluses simulated by the DT, after implementation in the simulation tool, showed a good agreement with real-world data.

CONCLUSIONS

The DT developed in this work represents a simple set of rules to mimic the same timing of correction boluses observed on real data. The inclusion of the model in simulation tools allows investigators to perform ISCTs that more realistically represent the patient behavior in taking correction boluses and the post-prandial BG response. In the future, more complex models can be investigated.

A Correction Insulin Bolus Delivery Strategy for Decision Support Systems in Type 1 Diabetes

Management of type 1 diabetes (T1D) requires affected individuals to perform multiple daily actions to keep their blood glucose levels within the safe rage and avoid adverse hypo-/hyperglycemic episodes. Decision support systems (DSS) for T1D are composite tools that implement multiple software modules aiming to ease such a burden and to improve glucose control. At the University of Padova, we are developing a new DSS that currently integrate a smart insulin bolus calculator for optimal insulin dosing and a rescue carbohydrate intake advisor to tackle hypoglycemia. However, a module specifically targeting hyperglycemia, that suggests the administration of corrective insulin boluses (CIB), is still missing. For such a scope, this work aims to assess a recent literature methodology, proposed by Aleppo et al., which provides a simple strategy for dealing with hyperglycemia. The methodology is tested retrospectively on clinical data of individuals with T1D. In particular, here we leveraged a novel in silico tool that first identifies a non-linear model of glucose-insulin dynamics on data, then uses such model to simulate and compare the glucose trace obtained by "replaying" the recorded scenario and the glucose trace obtained using the CIB delivery strategy under evaluation. Results show that the CIB delivery strategy significantly reduce the percentage of time spent in hyperglycemia (-15.63%) without inducing any hypoglycemic episode, demonstrating both safety and efficacy of its use. These preliminary results suggest that the CIB delivery strategy proposed by Aleppo et al. is a promising candidate to be included in our system to counteract hyperglycemia. Future work will extensively evaluate the methodology and will compare it against other competing approaches.

A New Decision Support System for Type 1 Diabetes Management

Type 1 diabetes (T1D) is a chronic life-threatening metabolic condition which needs to be accurately and continuously managed with care by multiple daily exogenous insulin injections, frequent blood glucose concentration monitoring, ad-hoc diet, and physical activity. In the last decades, new technologies, such as continuous glucose monitoring sensors, eased the burden for T1D patients and opened new therapy perspectives by fostering the development of decision support systems (DSS). A DSS for T1D should be able to provide patients with advice aimed at improving metabolic control and reducing the number of actions related to therapy handling. Major challenges are the vast intra-/inter-subject physiological variability and the many factors that impact glucose metabolism. The present work illustrates a new DSS for T1D management. The algorithmic core includes a module for optimal, personalized, insulin dose calculation and a module that triggers the assumption of rescue carbohydrates to avoid/mitigate impending hypoglycemic events. The algorithms are integrated within a prototype communication platform that comprises a mobile app, a real-time telemonitoring interface, and a cloud server to safely store patients' data. Tests made in silico show that the use of the new algorithms lead to metabolic control indices significantly better than those obtained by the standard care for T1D. The preliminary test of the prototype platform suggests that it is robust, performant, and well-accepted by both patients and clinicians. Future work will focus on the refinement of the communication platform and the design of a clinical trial to assess the system effectiveness in real-life conditions.Clinical Relevance- The presented DSS is a promising tool to facilitate T1D daily management and improve therapy efficacy.

Linear Model Identification for Personalized Prediction and Control in Diabetes

OBJECTIVE

Type-1 diabetes (T1D) is a metabolic disease, characterized by impaired blood glucose (BG) regulation, which forces patients to multiple daily therapeutic actions, the most critical of which is exogenous insulin administration. T1D management can considerably benefit of mathematical models enabling accurate BG predictions and effective/safe automated insulin delivery. In building these models, dealing with large inter- and intra-patient variability in glucose-insulin dynamics represents a major challenge. The aim of the present work is to assess linear black-box methods, including a novel non-parametric methodology, for learning individualized models of glucose response to insulin and meal, suitable for model-based prediction and control.

METHODS

We focus on data-driven techniques for linear model-learning and compare the state-of-art parametric pipeline, exploring all its degrees of freedom (including population vs. individualized parameter identification, model class chosen among ARX/ARMAX/ARIMAX/Box-Jenkins, model order selection criteria, etc.), with a novel non-parametric approach based on Gaussian regression and stable spline kernel. By using data collected in 11 T1D individuals, we evaluate effectiveness of the different models by measuring root mean squared error (RMSE), coefficient of determination (COD), and time gain of the associated BG predictors.

RESULTS

Among the tested approaches, the non-parametric technique results in the best prediction performance: median RMSE=29.8mg/dL, and median COD=57.4%, for a prediction horizon (PH) of 60 min. With respect to the state-of-the-art parametric techniques, the non-parametric approach grants a COD improvement of about 2%, 7%, 21%, and 41% for PH = 30, 60, 90, and 120 min (paired-sample t-test p 0.001, p=0.003, p=0.03, and p=0.07 respectively).

CONCLUSION

Non-parametric linear model-learning grants statistically significant improvement with respect to the state-of-art parametric approach. The higher PH, the more pronounced the improvement.

SIGNIFICANCE

The use of a linear non-parametric model-learning approach for model-based prediction and control could bring to a more prompt, safe and effective T1D management.

Mathematical Models of Meal Amount and Timing Variability With Implementation in the Type-1 Diabetes Patient Decision Simulator

BACKGROUND

In type 1 diabetes (T1D) research, in-silico clinical trials (ISCTs) have proven effective in accelerating the development of new therapies. However, published simulators lack a realistic description of some aspects of patient lifestyle which can remarkably affect glucose control. In this paper, we develop a mathematical description of meal carbohydrates (CHO) amount and timing, with the aim to improve the meal generation module in the T1D Patient Decision Simulator (T1D-PDS) published in Vettoretti et al.

METHODS

Data of 32 T1D subjects under free-living conditions for 4874 days were used. Univariate probability density function (PDF) parametric models with different candidate shapes were fitted, individually, against sample distributions of: CHO amounts of breakfast (CHOB), lunch (CHOL), dinner (CHOD), and snack (CHOS); breakfast timing (TB); and time between breakfast-lunch (TBL) and between lunch-dinner (TLD). Furthermore, a support vector machine (SVM) classifier was developed to predict the occurrence of a snack in future fixed-length time windows. Once embedded inside the T1D-PDS, an ISCT was performed.

RESULTS

Resulting PDF models were: gamma (CHOB, CHOS), lognormal (CHOL, TB), loglogistic (CHOD), and generalized-extreme-values (TBL, TLD). The SVM showed a classification accuracy of 0.8 over the test set. The distributions of simulated meal data were not statistically different from the distributions of the real data used to develop the models (α = 0.05).

CONCLUSIONS

The models of meal amount and timing variability developed are suitable for describing real data. Their inclusion in modules that describe patient behavior in the T1D-PDS can permit investigators to perform more realistic, reliable, and insightful ISCTs.

Forecasting of Glucose Levels and Hypoglycemic Events: Head-to-Head Comparison of Linear and Nonlinear Data-Driven Algorithms Based on Continuous Glucose Monitoring Data Only

In type 1 diabetes management, the availability of algorithms capable of accurately forecasting future blood glucose (BG) concentrations and hypoglycemic episodes could enable proactive therapeutic actions, e.g., the consumption of carbohydrates to mitigate, or even avoid, an impending critical event. The only input of this kind of algorithm is often continuous glucose monitoring (CGM) sensor data, because other signals (such as injected insulin, ingested carbs, and physical activity) are frequently unavailable. Several predictive algorithms fed by CGM data only have been proposed in the literature, but they were assessed using datasets originated by different experimental protocols, making a comparison of their relative merits difficult. The aim of the present work was to perform a head-to-head comparison of thirty different linear and nonlinear predictive algorithms using the same dataset, given by 124 CGM traces collected over 10 days with the newest Dexcom G6 sensor available on the market and considering a 30-min prediction horizon. We considered the state-of-the art methods, investigating, in particular, linear black-box methods (autoregressive; autoregressive moving-average; and autoregressive integrated moving-average, ARIMA) and nonlinear machine-learning methods (support vector regression, SVR; regression random forest; feed-forward neural network, fNN; and long short-term memory neural network). For each method, the prediction accuracy and hypoglycemia detection capabilities were assessed using either population or individualized model parameters. As far as prediction accuracy is concerned, the results show that the best linear algorithm (individualized ARIMA) provides accuracy comparable to that of the best nonlinear algorithm (individualized fNN), with root mean square errors of 22.15 and 21.52 mg/dL, respectively. As far as hypoglycemia detection is concerned, the best linear algorithm (individualized ARIMA) provided precision = 64%, recall = 82%, and one false alarm/day, comparable to the best nonlinear technique (population SVR): precision = 63%, recall = 69%, and 0.5 false alarms/day. In general, the head-to-head comparison of the thirty algorithms fed by CGM data only made using a wide dataset shows that individualized linear models are more effective than population ones, while no significant advantages seem to emerge when employing nonlinear methodologies.

Advanced Diabetes Management Using Artificial Intelligence and Continuous Glucose Monitoring Sensors

Wearable continuous glucose monitoring (CGM) sensors are revolutionizing the treatment of type 1 diabetes (T1D). These sensors provide in real-time, every 1-5 min, the current blood glucose concentration and its rate-of-change, two key pieces of information for improving the determination of exogenous insulin administration and the prediction of forthcoming adverse events, such as hypo-/hyper-glycemia. The current research in diabetes technology is putting considerable effort into developing decision support systems for patient use, which automatically analyze the patient's data collected by CGM sensors and other portable devices, as well as providing personalized recommendations about therapy adjustments to patients. Due to the large amount of data collected by patients with T1D and their variety, artificial intelligence (AI) techniques are increasingly being adopted in these decision support systems. In this paper, we review the state-of-the-art methodologies using AI and CGM sensors for decision support in advanced T1D management, including techniques for personalized insulin bolus calculation, adaptive tuning of bolus calculator parameters and glucose prediction.

Development of an Error Model for a Factory-Calibrated Continuous Glucose Monitoring Sensor with 10-Day Lifetime

Factory-calibrated continuous glucose monitoring (FC-CGM) sensors are new devices used in type 1 diabetes (T1D) therapy to measure the glucose concentration almost continuously for 10-14 days without requiring any in vivo calibration. Understanding and modelling CGM errors is important when designing new tools for T1D therapy. Available literature CGM error models are not suitable to describe the FC-CGM sensor error, since their domain of validity is limited to 12-h time windows, i.e., the time between two consecutive in vivo calibrations. The aim of this paper is to develop a model of the error of FC-CGM sensors. The dataset used contains 79 FC-CGM traces collected by the Dexcom G6 sensor. The model is designed to dissect the error into its three main components: effect of plasma-interstitium kinetics, calibration error, and random measurement noise. The main novelties are the model extension to cover the entire sensor lifetime and the use of a new single-step identification procedure. The final error model, which combines a first-order linear dynamic model to describe plasma-interstitium kinetics, a second-order polynomial model to describe calibration error, and an autoregressive model to describe measurement noise, proved to be suitable to describe FC-CGM sensor errors, in particular improving the estimation of the physiological time-delay.