Model-based closed-loop glucose control in type 1 diabetes: the DiaCon experience

Personalized insulin dose manipulation attack and its detection using interval-based temporal patterns and machine learning algorithms

Many patients with diabetes are currently being treated with insulin pumps and other diabetes devices which improve their quality of life and enable effective treatment of diabetes. These devices are connected wirelessly and thus, are vulnerable to cyber-attacks which have already been proven feasible. In this paper, we focus on two types of cyber-attacks on insulin pump systems: an overdose of insulin, which can cause hypoglycemia, and an underdose of insulin, which can cause hyperglycemia. Both of these attacks can result in a variety of complications and endanger a patient's life. Specifically, we propose a sophisticated and personalized insulin dose manipulation attack; this attack is based on a novel method of predicting the blood glucose (BG) level in response to insulin dose administration. To protect patients from the proposed sophisticated and malicious insulin dose manipulation attacks, we also present an automated machine learning based system for attack detection; the detection system is based on an advanced temporal pattern mining process, which is performed on the logs of real insulin pumps and continuous glucose monitors (CGMs). Our multivariate time-series data (MTSD) collection consists of 225,780 clinical logs, collected from real insulin pumps and CGMs of 47 patients with type I diabetes (13 adults and 34 children) from two different clinics at Soroka University Medical Center in Beer-Sheva, Israel over a four-year period. We enriched our data collection with additional relevant medical information related to the subjects. In the extensive experiments performed, we evaluated the proposed attack and detection system and examined whether: (1) it is possible to accurately predict BG levels in order to create malicious data that simulate a manipulation attack and the patient's body in response to it; (2) it is possible to automatically detect such attacks based on advanced machine learning (ML) methods that leverage temporal patterns; (3) the detection capabilities of the proposed detection system differ for insulin overdose and underdose attacks; and (4) the granularity of the learning model (general / adult vs. pediatric clinic / individual patient) affects the detection capabilities. Our results show that (a) it is possible to predict, with nearly 90% accuracy, BG levels using our proposed methods, and by doing so, enable malicious data creation for our detection system evaluation; (b) it is possible to accurately detect insulin manipulation attacks using temporal patterns mining using several ML methods, including Logistic Regression, Random Forest, TPF class model, TPF top k, and ANN algorithms; (c) it is easier to detect an overdose attack than an underdose attack in more than 25%, in terms of AUC scores; and (d) the adult vs. pediatric model outperformed models of other granularities in the detection of overdose attacks, while the general model outperformed the other models in the case of detecting underdose attacks; for both attacks, attack detection among children was found to be more challenging than among adults. In addition to its use in the evaluation of our detection system, the proposed BG prediction method has great importance in the medical domain where it can contribute to improved care of patients with diabetes.

Risk-based postprandial hypoglycemia forecasting using supervised learning

BACKGROUND

Predicting insulin-induced postprandial hypoglycemic events is critical for the safety of type 1 diabetes patients because an early warning of hypoglycemia facilitates correction of the insulin bolus before its administration. The postprandial hypoglycemic event counts can be lowered by reducing the size of the bolus based on a reliable prediction but at the cost of increasing the average blood glucose.

METHODS

We developed a method for predicting postprandial hypoglycemia using machine learning techniques personalized to each patient. The proposed system enables on-line therapeutic decision making for patients using a sensor augmented pump therapy. Two risk-based approaches were developed for a window of 240 min after the meal/bolus, and they were tested based on real retrospective data from 10 patients using 70 mg/dL and 54 mg/dL as thresholds according to the consensus for Level 1 and Level 2 hypoglycemia, respectively. Due to the small size of the patient cohort, we trained personalized models for each patient.

RESULTS

The median specificity and sensitivity were 79% and 71% for Level 1 hypoglycemia, respectively, and 81% and 77% for Level 2.

CONCLUSIONS

The results demonstrated that it is feasible to anticipate hypoglycemic events with a reasonable false-positive rate. The accuracy of the results and the trade-off between performance metrics allow its use in decision support systems for patients who wear insulin pumps.

Efficacy and safety of the artificial pancreas in the paediatric population with type 1 diabetes

BACKGROUND

Type 1 diabetes (DM1) is one of the most common chronic diseases in childhood and requires life-long insulin therapy and continuous health care support. An artificial pancreas (AP) or closed-loop system (CLS) have been developed with the aim of improving metabolic control without increasing the risk of hypoglycaemia in patients with DM1. As the impact of APs have been studied mainly in adults, the aim of this review is to evaluate the efficacy and safety of the AP in the paediatric population with DM1.

MAIN BODY

The real advantage of a CLS compared to last-generation sensor-augmented pumps is the gradual modulation of basal insulin infusion in response to glycaemic variations (towards both hyperglycaemia and hypoglycaemia), which has the aim of improving the proportion of time spent in the target glucose range and reducing the mean glucose level without increasing the risk of hypoglycaemia. Some recent studies demonstrated that also in children and adolescents an AP is able to reduce the frequency of hypoglycaemic events, an important limiting factor in reaching good metabolic control, particularly overnight. However, the advantages of the AP in reducing hyperglycaemia, increasing the time spent in the target glycaemic range and thus reducing glycated haemoglobin are less clear and require more clinical trials in the paediatric population, in particular in younger children.

CONCLUSIONS

Although the first results from bi-hormonal CLS are promising, long-term, head-to-head studies will have to prove their superiority over insulin-only approaches. More technological progress, the availability of more fast-acting insulin, further developments of algorithms that could improve glycaemic control after meals and physical activity are the most important challenges in reaching an optimal metabolic control with the use of the AP in children and adolescents.

Automated blood glucose control in type 1 diabetes: A review of progress and challenges

Since the 2000s, research teams worldwide have been working to develop closed-loop (CL) systems able to automatically control blood glucose (BG) levels in patients with type 1 diabetes. This emerging technology is known as artificial pancreas (AP), and its first commercial version just arrived in the market. The main objective of this paper is to present an extensive review of the clinical trials conducted since 2011, which tested various implementations of the AP for different durations under varying conditions. A comprehensive table that contains key information from the selected publications is provided, and the main challenges in AP development and the mitigation strategies used are discussed. The development timelines for different AP systems are also included, highlighting the main evolutions over the clinical trials for each system.

Effect of artificial pancreas systems on glycaemic control in patients with type 1 diabetes: a systematic review and meta-analysis of outpatient randomised controlled trials

BACKGROUND

Closed-loop artificial pancreas systems have been in development for several years, including assessment in numerous varied outpatient clinical trials. We aimed to summarise the efficacy and safety of artificial pancreas systems in outpatient settings and explore the clinical and technical factors that can affect their performance.

METHODS

We did a systematic review and meta-analysis of randomised controlled trials comparing artificial pancreas systems (insulin only or insulin plus glucagon) with conventional pump therapy (continuous subcutaneous insulin infusion [CSII] with blinded continuous glucose monitoring [CGM] or unblinded sensor-augmented pump [SAP] therapy) in adults and children with type 1 diabetes. We searched Medline, Embase, and the Cochrane Central Register of Controlled Trials for studies published from 1946, to Jan 1, 2017. We excluded studies not published in English, those involving pregnant women or participants who were in hospital, and those testing adjunct medications other than glucagon. The primary outcome was the mean difference in percentage of time blood glucose concentration remained in target range (3·9-10 mmol/L or 3·9-8 mmol/L, depending on the study), assessed by random-effects meta-analysis. This study is registered with PROSPERO, number 2015:CRD42015026854.

FINDINGS

We identified 984 reports; after exclusions, 27 comparisons from 24 studies (23 crossover and one parallel design) including a total of 585 participants (219 in adult studies, 265 in paediatric studies, and 101 in combined studies) were eligible for analysis. Five comparisons assessed dual-hormone (insulin and glucagon), two comparisons assessed both dual-hormone and single-hormone (insulin only), and 20 comparisons assessed single-hormone artificial pancreas systems. Time in target was 12·59% higher with artificial pancreas systems (95% CI 9·02-16·16; p<0·0001), from a weighted mean of 58·21% for conventional pump therapy (I²=84%). Dual-hormone artificial pancreas systems were associated with a greater improvement in time in target range compared with single-hormone systems (19·52% [95% CI 15·12-23·91] vs 11·06% [6·94 to 15·18]; p=0·006), although six of seven comparisons compared dual-hormone systems to CSII with blinded CGM, whereas 21 of 22 single-hormone comparisons had SAP as the comparator. Single-hormone studies had higher heterogeneity than dual-hormone studies (I² 79% vs 66%). Bias assessment characteristics were incompletely reported in 12 of 24 studies, no studies masked participants to the intervention assignment, and masking of outcome assessment was not done in 12 studies and was unclear in 12 studies.

INTERPRETATION

Artificial pancreas systems uniformly improved glucose control in outpatient settings, despite heterogeneous clinical and technical factors.

FUNDING

None.

A Review of the Current Challenges Associated with the Development of an Artificial Pancreas by a Double Subcutaneous Approach

INTRODUCTION

Patients with diabetes type 1 (DM1) struggle daily to achieve good glucose control. The last decade has seen a rush of research groups working towards an artificial pancreas (AP) through the application of a double subcutaneous approach, i.e., subcutaneous (SC) continuous glucose monitoring (CGM) and continuous subcutaneous insulin infusion. Few have focused on the fundamental limitations of this approach, especially regarding outcome measures beyond time in range.

METHODS

Based on insulin physiology, the limitations of CGM, SC insulin absorption, meal challenge, and physical activity in DM1 patients, we discuss the limitations of the double SC approach. Finally, we discuss safety measures and the achievements reported in some recent AP studies that have utilized the double SC approach.

RESULTS

Most studies show that a double SC AP increases the time in range compared to a sensor-augmented insulin pump and shortens the time in hypoglycemia. Despite these achievements, the proportion of time spent in hyperglycemia is still roughly 20-40%, and hypoglycemia is still present 1-4% of the time. The main factors limiting further progress are the latency of SC CGM (at least 5-10 min) and the slow pharmacokinetics of SC-delivered fast-acting insulin. The maximum blood insulin level is reached after 45 min and the maximum glucose-lowering effect is observed after 1.5-2 h, while the glucose-lowering effect lasts for at least 5 h.

CONCLUSIONS

Although using a double SC AP leads to significant improvements in glucose control, the SC approach has severe limitations that hamper further progress towards a robust AP.

A review of personalized blood glucose prediction strategies for T1DM patients

This paper presents a methodological review of models for predicting blood glucose (BG) concentration, risks and BG events. The surveyed models are classified into three categories, and they are presented in summary tables containing the most relevant data regarding the experimental setup for fitting and testing each model as well as the input signals and the performance metrics. Each category exhibits trends that are presented and discussed. This document aims to be a compact guide to determine the modeling options that are currently being exploited for personalized BG prediction.

An Adaptive Nonlinear Basal-Bolus Calculator for Patients With Type 1 Diabetes

BACKGROUND

Bolus calculators help patients with type 1 diabetes to mitigate the effect of meals on their blood glucose by administering a large amount of insulin at mealtime. Intraindividual changes in patients physiology and nonlinearity in insulin-glucose dynamics pose a challenge to the accuracy of such calculators.

METHOD

We propose a method based on a continuous-discrete unscented Kalman filter to continuously track the postprandial glucose dynamics and the insulin sensitivity. We augment the Medtronic Virtual Patient (MVP) model to simulate noise-corrupted data from a continuous glucose monitor (CGM). The basal rate is determined by calculating the steady state of the model and is adjusted once a day before breakfast. The bolus size is determined by optimizing the postprandial glucose values based on an estimate of the insulin sensitivity and states, as well as the announced meal size. Following meal announcements, the meal compartment and the meal time constant are estimated, otherwise insulin sensitivity is estimated.

RESULTS

We compare the performance of a conventional linear bolus calculator with the proposed bolus calculator. The proposed basal-bolus calculator significantly improves the time spent in glucose target ( P < .01) compared to the conventional bolus calculator.

CONCLUSION

An adaptive nonlinear basal-bolus calculator can efficiently compensate for physiological changes. Further clinical studies will be needed to validate the results.

Identification of Main Factors Explaining Glucose Dynamics During and Immediately After Moderate Exercise in Patients With Type 1 Diabetes

BACKGROUND

Physical activity is recommended for patients with type 1 diabetes (T1D). However, without proper management, it can lead to higher risk for hypoglycemia and impaired glycemic control. In this work, we identify the main factors explaining the blood glucose dynamics during exercise in T1D. We then propose a prediction model to quantify the glycemic drop induced by a mild to moderate physical activity.

METHODS

A meta-data analysis was conducted over 59 T1D patients from 4 different studies in the United States and France (37 men and 22 women; 47 adults; weight, 71.4 ± 10.6 kg; age, 42 ± 10 years; 12 adolescents: weight, 60.7 ± 12.5 kg; age, 14.0 ± 1.4 years). All participants had physical activity between 3 and 5 pm at a mild to moderate intensity for approximately 30 to 45 min. A multiple linear regression analysis was applied to the data to identify the main parameters explaining the glucose dynamics during such physical activity.

RESULTS

The blood glucose at the beginning of exercise ([Formula: see text]), the ratio of insulin on board over total daily insulin ([Formula: see text]) and the age as a categorical variable (1 for adult, 0 for adolescents) were significant factors involved in glucose evolution at exercise (all P < .05). The multiple linear regression model has an R-squared of .6.

CONCLUSIONS

The main factors explaining glucose dynamics in the presence of mild-to-moderate exercise in T1D have been identified. The clinical parameters are formally quantified using real data collected during clinical trials. The multiple linear regression model used to predict blood glucose during exercise can be applied in closed-loop control algorithms developed for artificial pancreas.

Sensitivity of the Predictive Hypoglycemia Minimizer System to the Algorithm Aggressiveness Factor

BACKGROUND

The Predictive Hypoglycemia Minimizer System ("Hypo Minimizer"), consisting of a zone model predictive controller (the "controller") and a safety supervision module (the "safety module"), aims to mitigate hypoglycemia by preemptively modulating insulin delivery based on continuous glucose monitor (CGM) measurements. The "aggressiveness factor," a pivotal variable in the system, governs the speed and magnitude of the controller's insulin dosing characteristics in response to changes in CGM levels.

METHODS

Twelve adults with type 1 diabetes were studied in closed-loop in a clinical research center for approximately 24 hours. This analysis focused primarily on the effect of the aggressiveness factor on the automated insulin-delivery characteristics of the controller, and secondarily on the glucose control results.

RESULTS

As aggressiveness increased from "conservative" to "medium" to "aggressive," the controller recommended less insulin (-3.3% vs -14.4% vs -19.5% relative to basal) with a higher frequency (5.3% vs 14.4% vs 20.3%) during the critical times when the CGM was reading 90-120 mg/dl and decreasing. Blood glucose analyses indicated that the most aggressive setting resulted in the most desirable combination of the least time spent <70 mg/dl and the most time spent 70-180 mg/dl, particularly in the overnight period. Hyperglycemia, diabetic ketoacidosis, or severe hypoglycemia did not occur with any of the aggressiveness values.

CONCLUSION

The Hypo Minimizer's controller took preemptive action to prevent hypoglycemia based on predicted changes in CGM glucose levels. The most aggressive setting was quickest to take action to reduce insulin delivery below basal and achieved the best glucose metrics.

Closed loop insulin delivery in diabetes

The primary goal of type 1 diabetes treatment is attaining near-normal glucose values. This currently remains out of reach for most people with type 1 diabetes despite intensified insulin treatment in the form of insulin analogues, educational interventions, continuous glucose monitoring, and sensor augmented insulin pump. The main remaining problem is risk of hypoglycaemia, which cannot be sufficiently reduced in all patient groups. Additionally, patients' burn-out often develops with years of tedious day-to-day diabetes management, rendering available diabetes-related technology less efficient. Over the past 40 years, several attempts have been made towards computer-programmed insulin delivery in the form of closed loop, with faster developments especially in the past decade. Automated insulin delivery has reduced human error in glycaemic control and considerably lessened the burden of routine self-management. In this chapter, data from randomized controlled trials with closed-loop insulin delivery that included type 1 diabetes population are summarized, and an evidence-based vision for possible routine utilization of closed loop is provided.

An artificial pancreas for automated blood glucose control in patients with Type 1 diabetes

Automated glucose control in patients with Type 1 diabetes is much-coveted by patients, relatives and healthcare professionals. It is the expectation that a system for automated control, also know as an artificial pancreas, will improve glucose control, reduce the risk of diabetes complications and markedly improve patient quality of life. An artificial pancreas consists of portable devices for glucose sensing and insulin delivery which are controlled by an algorithm residing on a computer. The technology is still under development and currently no artificial pancreas is commercially available. This review gives an introduction to recent progress, challenges and future prospects within the field of artificial pancreas research.

The artificial pancreas: current status and future prospects in the management of diabetes

Recent advances in insulins, insulin pumps, continuous glucose-monitoring systems, and control algorithms have resulted in an acceleration of progress in the development of artificial pancreas devices. This review discusses progress in the development of external systems that are based on subcutaneous drug delivery and subcutaneous continuous glucose monitoring. There are two major system-level approaches to achieving closed-loop control of blood glucose in diabetic individuals. The unihormonal approach uses insulin to reduce blood glucose and relies on complex safety mitigation algorithms to reduce the risk of hypoglycemia. The bihormonal approach uses both insulin to lower blood glucose and glucagon to raise blood glucose, and also relies on complex algorithms to provide for safety of the user. There are several major strategies for the design of control algorithms and supervision control for application to the artificial pancreas: proportional-integral-derivative, model predictive control, fuzzy logic, and safety supervision designs. Advances in artificial pancreas research in the first decade of this century were based on the ongoing computer revolution and miniaturization of electronic technology. The advent of modern smartphones has created the ability to utilize smartphone technology as the engineering centerpiece of an artificial pancreas. With these advances, an artificial or bionic pancreas is within reach.

Predicting Plasma Glucose From Interstitial Glucose Observations Using Bayesian Methods

One way of constructing a control algorithm for an artificial pancreas is to identify a model capable of predicting plasma glucose (PG) from interstitial glucose (IG) observations. Stochastic differential equations (SDEs) make it possible to account both for the unknown influence of the continuous glucose monitor (CGM) and for unknown physiological influences. Combined with prior knowledge about the measurement devices, this approach can be used to obtain a robust predictive model. A stochastic-differential-equation-based gray box (SDE-GB) model is formulated on the basis of an identifiable physiological model of the glucoregulatory system for type 1 diabetes mellitus (T1DM) patients. A Bayesian method is used to estimate robust parameters from clinical data. The models are then used to predict PG from IG observations from 2 separate study occasions on the same patient. First, all statistically significant diffusion terms of the model are identified using likelihood ratio tests, yielding inclusion of [Formula: see text], [Formula: see text], and [Formula: see text]. Second, estimates using maximum likelihood are obtained, but prediction capability is poor. Finally a Bayesian method is implemented. Using this method the identified models are able to predict PG using only IG observations. These predictions are assessed visually. We are also able to validate these estimates on a separate data set from the same patient. This study shows that SDE-GBs and a Bayesian method can be used to identify a reliable model for prediction of PG using IG observations obtained with a CGM. The model could eventually be used in an artificial pancreas.