Personalized Predictive Modeling in Type 1 Diabetes

Eleni I. Georga, Dimitrios I Fotiadis, Stelios K. Tigas

• Publisher: Academic Press
• Year: 2017
• Print ISBN: 9780128048313
• eBook ISBN: 9780128051467
• Topic: Médecine
• Subtopic: Bio-informatique

Personalized Predictive Modeling in Diabetes features state-of-the-art methodologies and algorithmic approaches which have been applied to predictive modeling of glucose concentration, ranging from simple autoregressive models of the CGM time series to multivariate nonlinear regression techniques of machine learning. Developments in the field have been analyzed with respect to: (i) feature set (univariate or multivariate), (ii) regression technique (linear or non-linear), (iii) learning mechanism (batch or sequential), (iv) development and testing procedure and (v) scaling properties. In addition, simulation models of meal-derived glucose absorption and insulin dynamics and kinetics are covered, as an integral part of glucose predictive models.This book will help engineers and clinicians to: select a regression technique which can capture both linear and non-linear dynamics in glucose metabolism in diabetes, and which exhibits good generalization performance under stationary and non-stationary conditions; ensure the scalability of the optimization algorithm (learning mechanism) with respect to the size of the dataset, provided that multiple days of patient monitoring are needed to obtain a reliable predictive model; select a features set which efficiently represents both spatial and temporal dependencies between the input variables and the glucose concentration; select simulation models of subcutaneous insulin absorption and meal absorption; identify an appropriate validation procedure, and identify realistic performance measures.- Describes fundamentals of modeling techniques as applied to glucose control- Covers model selection process and model validation- Offers computer code on a companion website to show implementation of models and algorithms- Features the latest developments in the field of diabetes predictive modeling

1

Background and Preview

Abstract

This chapter presents a comprehensive overview of the state-of-the-art methodologies and algorithmic approaches to data-driven predictive modeling of glucose concentration in type 1 diabetes. In the first section, in an attempt to familiarize the reader with the basic components of daily diabetes care, we feature the current medical technologies for monitoring and controlling blood-glucose levels as well as contemporary mobile health interventions. In this context, we specify the clinical importance of glucose prediction models and how they can be integrated into the digital ecosystem of diabetes data-driven technologies. The second part of this chapter outlines the main types of linear and nonlinear, with respect to the input, dynamic models that have been applied for the identification and short-term prediction of the underlying glucose regulation system. Finally, provided the time-varying nature of glucose-insulin dynamics, the need for adaptive learning schemes is described and the main paradigms toward this direction are presented.

Keywords

Types 1 and 2 diabetes; glucose monitoring and control; data-driven predictive models; linear and nonlinear dynamic models; adaptive models

1.1 Data-Driven Glucose Prediction Models and Clinical Impact

Diabetes is a group of metabolic disorders characterized by hyperglycaemia resulting from defects in insulin secretion, insulin action, or both [1]. Type 1 diabetes results from a cellular-mediated autoimmune destruction of the β-cells in the pancreas, leading to absolute insulin deficiency. On the other hand, type 2 diabetes is characterized by a progressive loss of insulin secretion on the background of insulin resistance [2]. According to the International Diabetes Federation, the number of people (adults 20–79 years) with diabetes worldwide is estimated to rise from 415 million in 2015 to 642 million in 2040, while type 1 diabetes is increasing by around 3% every year, particularly among children. Moreover, the long-term microvascular and macrovascular complications, associated with the chronic hyperglycaemia, render diabetes a major cause of early death in most countries.

The most vital and challenging issue for people with type 1 or advanced type 2 diabetes is the achievement and maintenance of euglycaemia overtime in a safe manner. The daily management of the disease can be seen as a feedback loop in which frequent self-monitoring of blood glucose (SMBG) is strongly associated with better glycaemic control. Intensive insulin therapy (IIT), implemented by either multiple daily injections (MDI) or continuous subcutaneous insulin infusion (CSII), could be the remedy for hyperglycaemia in type 1 diabetes, should it not increase the risk of hypoglycaemia, which is defined as a blood-glucose concentration below 70 mgdL⁻¹. The long-term benefits of IIT along with the increased frequency of hypoglycaemic events were first demonstrated by the Diabetes Control and Complications Trial [3]. Since then, despite the significant improvements in insulin analogues, hypoglycaemia has been recognized as the major barrier to the management of diabetes [4].

The reduction of the risk of hypoglycaemia is a matter of major interest in daily diabetes care. Hypoglycaemia in insulin-dependent diabetes patients is the aggregate of therapeutic hyperinsulinemia, as well as, attenuated sympathoadrenal response to falling plasma-glucose concentrations [5]. In addition, recent antecedent hypoglycaemia, prior exercise, and sleep further impair the physiological and behavioural defenses against a potential subsequent hypoglycaemia (i.e., hypoglycaemia-associated autonomic failure—HAAF) and, therefore, cause a vicious cycle of recurrent hypoglycaemia [4,6]. The awareness of these factors by individuals with diabetes may contribute to the prevention of hypoglycaemia on a daily basis. It is acknowledged that achieving and maintaining tight glycaemic control necessitates the proper consideration of extrinsic factors having a direct impact on subsequent glucose concentrations, such as nutrition, physical activity, patient’s psychological status, and the overall lifestyle [2]. In addition to these, the endogenous processes involved in the regulation of glucose homeostasis, as well as the prominent intra- and interpatient variability in response to insulin therapy [7–9], render glucose control in diabetes—a rather difficult procedure.

The technological advances in continuous glucose monitoring (CGM) and CSII have contributed to more efficient and safe therapeutic procedures, especially for insulin-treated diabetes [10]. The high temporal resolution of CGM offers the potential to gain a deeper insight into 24-h glucose dynamics in the subcutaneous space supporting a more informed and comprehensive decision-making in both clinical and self-monitoring conditions. The American Diabetes Association recommends, “CGM may be a supplemental tool to SMBG in those with hypoglycemia unawareness and/or frequent hypoglycemic episodes” [2]. The findings of numerous clinical trials have confirmed that CGM reduces HbA1c, an index of average glycaemic control over the preceding 2–3 months, in type 1 diabetes as compared to SMBG; however, they have not shown significant reductions in severe hypoglycaemia [11–13]. Modern CGM systems are able to provide customizable predictive alerts for upcoming critical events, which has been shown to improve hypoglyceamia detection [14,15]. Toward this direction, Facchinetti et al. [16] introduced the smart CGM sensor concept by adding three real-time signal processing algorithms for denoising, enhancement, and prediction to the Seven Plus CGM system (DexCom). In that way, the smoothness of the Seven Plus CGM time series improved by an average of 57%, the mean absolute relative difference (MARD) between blood-glucose measurements and CGM data reduced from 15.1% to 10.3%, and, finally, hypoglycaemic or hyperglycaemic events were predicted with an average horizon of 14 min. On the other hand, literature suggests that there are no significant differences in HbA1c or frequency of severe hypoglycaemia between CSII and MDI therapy [17,18]. Nevertheless, both CGM and CSII technologies form the basis for the development of more advanced technological solutions for controlling diabetes.

The effective integration of CGM and CSII technologies into one system, that is, sensor-augmented pump (SAP), allows improvements in glycemic control of type 1 diabetes when compared with MDI therapy or the individual components alone [1,19,20]; however, the problem of severe hypoglycaemia and, in particular, nocturnal hypoglycaemia is not solved. A far more promising approach to this problem is the automatic suspension of insulin delivery at a preset low-glucose threshold aiming at reducing basal insulinemia during the critical first few minutes of hypoglycaemia without causing rebound hyperglycaemia [21]. In particular, the so-called threshold-suspend feature available in the Medtronic Paradigm Veo pump was shown to significantly reduce the rate and the mean area under curve of nocturnal hypoglycaemic events by 31.8% and 37.5%, respectively, as compared to standard SAP therapy [22]. Moreover, the 24-h hypoglycaemic exposure was also reduced during the 3-month study, without significant changes in HbA1c levels. A following study did support that this technology has the potential to reduce the combined rate of severe and moderate hypoglycaemia in patients with type 1 diabetes [23]. Of great importance, the effectiveness of suspending the insulin pump delivery when the predicted risk of hypoglycaemia is high was demonstrated both by simulations and experiments [24–26]. It is acknowledged that SAP with threshold-suspend is the epitome of today’s diabetes technology and that is the next step for an automated closed-loop artificial pancreas, the most promising therapeutic approach to β-cell replacement.

Lately, incremental steps have also been taken toward a portable closed-loop system for overnight glucose control suitable for outpatient use [27–30]. However, further research is needed to ensure the reliable, stable, and safe operation of the whole system given the limitations of existing subcutaneous glucose sensors and wireless communications. The control algorithm is currently implemented either with proportional–integral–derivative (PID) control or model-predictive control (MPC) [31,32]. PID control as a reactive feedback mechanism cannot effectively correct sudden hypoglycaemia or hyperglycaemia. On the other hand, MPC uses mathematical models of diabetes physiology or data-driven models to predict the short-term glucose dynamics. MPC not only allows for better regulation of glucose in both steady state (e.g., overnight) and dynamic conditions (e.g., postprandial, during exercise) but also mitigates the time lags imposed by the subcutaneous route in glucose sensing and insulin absorption. We should also emphasize the utility of low-glucose predictive alerts available in both CGM and closed-loop systems [14,15,24–26], which act as a safeguard mechanism against hypoglycaemia. A modular architectural approach to closed-loop control has been proposed in Refs. [2,31] that allows the hierarchical integration of diverse components starting from functionalities assuring patient’s safety (i.e., insulin pump suspension if hypoglycaemia is anticipated) to customizable MPC of basal insulin rate in real time. Two following randomized crossover studies demonstrated the utility of that concept for designing and testing different closed-loop control configurations [33]. To this end, more proactive and sensitive to overall patient’s context predictive algorithms may result in tighter glycaemic control minimizing the risk of hypoglycaemia, while setting the appropriate circumstances for closing the loop during the day.

The ever-increasing computational power of smartphones, the rise of mobile health devices, and the improved wireless communication technologies have contributed altogether to the development of remarkable mobile tools for diabetes self-management [10]. A number of mobile diabetes applications have been developed, either for research or commercial purposes, to support self-monitoring and decision-making on a daily basis [34–40]. Contemporary paradigms offer precise self-monitoring and indirect decision support through effective tools for data tracking and data visualization. The findings of mobile diabetes interventions suggest that well-designed mobile tools with decision-support features have the potential to enhance self-management outcomes. Daily self-monitoring information can be efficiently analysed, retrospectively or in real-time, to provide patients with supportive feedback related to diabetes management. The application of advanced data analytic...

Table of contents

• Cover image
• Title page
• Table of Contents
• Copyright
• Preface
• Terminology List
• List of Abbreviations
• 1. Background and Preview
• 2. Pathophysiology and Management of Type 1 Diabetes
• 3. Methodology for Developing a Glucose Prediction Model
• 4. Physiological Models and Exogenous Input Modeling
• 5. Linear Time Series Models of Glucose Concentration
• 6. Nonlinear Models of Glucose Concentration
• 7. Prediction Models of Hypoglycemia
• 8. Adaptive Glucose Prediction Models
• 9. Existing and Potential Applications of Glucose Prediction Models
• 10. Conclusions and Future Trends
• Index