Blood Glucose Prediction Research Articles

Seasonal Local Models for Glucose Prediction in Type 1 Diabetes.

Linear empirical dynamic models have been widely used for blood glucose prediction and risks prevention in people with type 1 diabetes. More accurate blood glucose prediction models with longer prediction horizon (PH) are desirable to enable warnings to patients about imminent blood glucose changes with enough time to take corrective actions. In this study, a blood glucose prediction method is developed by integrating the predictions of a set of seasonal local models (each of them corresponding to different glucose profiles observed along historical data). In the modeling step, the number of sets and their corresponding glucose profiles characteristics are obtained by clustering techniques (Fuzzy C-Means). Then, Box-Jenkins methodology is used to identify a seasonal model for each set. Finally, blood glucose predictions of local models are integrated using different techniques. The proposed method is tested by using 18 60-h closed-loop experiments (including different exercise types and artificial pancreas strategies) and achieving mean absolute percentage error (MAPE) of 2.94%, 3.89%, 5.41%, 6.29% and 8.66% for 15-, 30-, 45-, 60-, and 90-min PHs, respectively.

CausalBG: Causal Recurrent Neural Network for the Blood Glucose Inference With IoT Platform

Predicting blood glucose concentration facilitates timely preventive measures against health risks induced by abnormal glucose events. Advances in IoT devices, such as continuous blood glucose monitors (CGMs) have made it convenient for measurements of blood glucose in real time. However, accurate and personalized blood glucose concentration prediction is still challenging. Previous inference models yield low-inference accuracy due to the ineffective feature extraction and the limited, imbalanced personal training data. The underlying causal correlations among the blood glucose series are scarcely captured by these models. In this article, we propose CausalBG, a causal recurrent neural network (CausalRNN) deployed on an IoT platform with smartphones and CGM for the accurate and efficient individual blood glucose concentration prediction. CausalBG automatically captures the underlying causal relationships embedded in the blood glucose features through CausalRNN, and efficiently shares the limited personal data among users for the sufficient training via the multitask framework. Evaluations and case studies on 112 users demonstrate that CausalBG significantly outperforms the conventional predictive models on the blood glucose dynamics inference.

Investigation of goodness of model data fit using PLSR and PCR regression models to determine informative wavelength band in NIR region for non-invasive blood glucose prediction

Diabetes mellitus is a metabolic disorder characterized by the deficiency in insulin secretion or insulin action on target tissues. Diabetes being diagnosed, requires frequent monitoring. Current measurement techniques are invasive in nature, which is painful. Non-invasive measurement methods for the prediction of blood glucose is a big challenge. Hence this paper focuses on a comparative study to analyze the performance of the two regression models: partial least square regression (PLSR) and principal component regression (PCR) models to arrive at the best regression model for the prediction of blood glucose non-invasively based on NIR diffuse reflectance spectroscopy. The regression model developed involves spectral data collected from 32 subjects from a diabetic center using diffuse reflectance spectrometer (DRS). The spectroscopic data were acquired from the subjects both invasively using clinical method and non-invasively using DRS setup in the NIR spectral range whose wavelength ranges from 750 to 1040 nm. Multivariate analysis using PLSR is found to be a better regression model in terms of the fitted response and estimated mean square prediction error. It is observed that the mean square error value is 0.04 mg/dl for the tenth component in the PLSR model. The developed regression model also proves to be a useful tool in identifying the informative wavelength bands for blood glucose measurement.

Data-driven modeling and prediction of blood glucose dynamics: Machine learning applications in type 1 diabetes.

Diabetes mellitus (DM) is a metabolic disorder that causes abnormal blood glucose (BG) regulation that might result in short and long-term health complications and even death if not properly managed. Currently, there is no cure for diabetes. However, self-management of the disease, especially keeping BG in the recommended range, is central to the treatment. This includes actively tracking BG levels and managing physical activity, diet, and insulin intake. The recent advancements in diabetes technologies and self-management applications have made it easier for patients to have more access to relevant data. In this regard, the development of an artificial pancreas (a closed-loop system), personalized decision systems, and BG event alarms are becoming more apparent than ever. Techniques such as predicting BG (modeling of a personalized profile), and modeling BG dynamics are central to the development of these diabetes management technologies. The increased availability of sufficient patient historical data has paved the way for the introduction of machine learning and its application for intelligent and improved systems for diabetes management. The capability of machine learning to solve complex tasks with dynamic environment and knowledge has contributed to its success in diabetes research. Recently, machine learning and data mining have become popular, with their expanding application in diabetes research and within BG prediction services in particular. Despite the increasing and expanding popularity of machine learning applications in BG prediction services, updated reviews that map and materialize the current trends in modeling options and strategies are lacking within the context of BG prediction (modeling of personalized profile) in type 1 diabetes. The objective of this review is to develop a compact guide regarding modeling options and strategies of machine learning and a hybrid system focusing on the prediction of BG dynamics in type 1 diabetes. The review covers machine learning approaches pertinent to the controller of an artificial pancreas (closed-loop systems), modeling of personalized profiles, personalized decision support systems, and BG alarm event applications. Generally, the review will identify, assess, analyze, and discuss the current trends of machine learning applications within these contexts. A rigorous literature review was conducted between August 2017 and February 2018 through various online databases, including Google Scholar, PubMed, ScienceDirect, and others. Additionally, peer-reviewed journals and articles were considered. Relevant studies were first identified by reviewing the title, keywords, and abstracts as preliminary filters with our selection criteria, and then we reviewed the full texts of the articles that were found relevant. Information from the selected literature was extracted based on predefined categories, which were based on previous research and further elaborated through brainstorming among the authors. The initial search was done by analyzing the title, abstract, and keywords. A total of 624 papers were retrieved from DBLP Computer Science (25), Diabetes Technology and Therapeutics (31), Google Scholar (193), IEEE (267), Journal of Diabetes Science and Technology (31), PubMed/Medline (27), and ScienceDirect (50). After removing duplicates from the list, 417 records remained. Then, we independently assessed and screened the articles based on the inclusion and exclusion criteria, which eliminated another 204 papers, leaving 213 relevant papers. After a full-text assessment, 55 articles were left, which were critically analyzed. The inter-rater agreement was measured using a Cohen Kappa test, and disagreements were resolved through discussion. Due to the complexity of BG dynamics, it remains difficult to achieve a universal model that produces an accurate prediction in every circumstance (i.e., hypo/eu/hyperglycemia events). Recently, machine learning techniques have received wider attention and increased popularity in diabetes research in general and BG prediction in particular, coupled with the ever-growing availability of a self-collected health data. The state-of-the-art demonstrates that various machine learning techniques have been tested to predict BG, such as recurrent neural networks, feed-forward neural networks, support vector machines, self-organizing maps, the Gaussian process, genetic algorithm and programs, deep neural networks, and others, using various group of input parameters and training algorithms. The main limitation of the current approaches is the lack of a well-defined approach to estimate carbohydrate intake, which is mainly done manually by individual users and is prone to an error that can severely affect the predictive performance. Moreover, a universal approach has not been established to estimate and quantify the approximate effect of physical activities, stress, and infections on the BG level. No researchers have assessed model predictive performance during stress and infection incidences in a free-living condition, which should be considered in future studies. Furthermore, a little has been done regarding model portability that can capture inter- and intra-variability among patients. It seems that the effect of time lags between the CGM readings and the actual BG levels is not well covered. However, in general, we foresee that these developments might foster the advancement of next-generation BG prediction algorithms, which will make a great contribution in the effort to develop the long-awaited, so-called artificial pancreas (a closed-loop system).

LSTMs and Neural Attention Models for Blood Glucose Prediction: Comparative Experiments on Real and Synthetic Data.

We have shown in previous work that LSTM networks are effective at predicting blood glucose levels in patients with type I diabetes, outperforming human experts and an SVR model trained with features computed by manually engineered physiological models. In this paper we present the results of a much larger set of experiments on real and synthetic datasets in what-if, agnostic, and inertial scenarios. Experiments on a more recent real-patient dataset, which we are releasing to the research community, demonstrate that LSTMs are robust to noise and can easily incorporate additional features, such as skin temperature, heart rate and skin conductance, without any change in the architecture. A neural attention module that we designed specifically for time series prediction improves prediction performance on synthetic data; however, the improvements do not transfer to real data. Conversely, using time of day as an additional input feature consistently improves the LSTM performance on real data but not on synthetic data. These and other differences show that behavior on synthetic data cannot be assumed to always transfer to real data, highlighting the importance of evaluating physiological models on data from real patients.

Minimizing postprandial hypoglycemia in Type 1 diabetes patients using multiple insulin injections and capillary blood glucose self-monitoring with machine learning techniques

Background: Diabetic patients treated with intensive insulin therapies require a tight glycemic control and may benefit from advanced tools to predict blood glucose (BG) concentration levels and hypo/hyperglycemia events. Prediction systems using machine learning techniques have mainly focused on applications for sensor augmented pump (SAP) therapy. In contrast, insulin bolus calculators that rely on BG prediction for multiple daily insulin (MDI) injections for patients under self-monitoring blood glucose (SMBG) are scarce because of insufficient data sources and limited prediction capability of forecasting models.Methods: We trained individualized models that can predict postprandial hypoglycemia via different machine learning algorithms using retrospective data from 10 real patients. In addition, we designed and tested a hypoglycemia reduction strategy for a similar in silico population. The system generates a bolus reduction suggestion as the scaled weighted sum of the predictions. We evaluated the general and postprandial glycemic outcomes of the in silico population to assess the systems capability of avoiding hypoglycemias.Results: The median [IQR] sensitivity and specificity for hypoglycemia cases where the BG level was below 70 mg/dL were 0.49 [0.2–0.5] and 0.74 [0.7–0.9], respectively. For hypoglycemia cases where the BG level was below 54 mg/dL, the median [IQR] sensitivity and specificity were 0.51 [0.4–0.6] and 0.74 [0.7–0.8], respectively.Conclusions: The results indicated a decrease of 37% in the median number of postprandial hypoglycemias median decrease of 44% for hypoglycemias of 70 mg/dL and 54 mg/dL, respectively. This dramatic reduction makes this method a good candidate to be integrated into any Decision Support System for diabetes management.

49-LB: Reported Utility of Automated Blood Glucose Forecasts

Background: A growing number of people use apps to track diabetes data and identify retrospective patterns that can inform management behavior. In September 2018, One Drop launched Automated Decision Support, which predicts blood glucose 8 hours in the future for non-insulin-dependent users with type 2 diabetes (T2D). Predictions help inform diabetes decision-making prospectively, but only if users find them useful. Objective: To understand how predictive insights facilitate behavior change in people with T2D, One Drop collected feedback from users and trained a machine learning model to predict when forecasts would be perceived as useful. This model was used to select timing and content of future forecasts and accompanying support messages. We then measured changes in feedback to assess forecast utility. Methodology: An initial sample of 23,876 forecasts were sent via in-app notifications to 4679 users with T2D. Forecasts consisted of BG trend, duration, and level (“rising, but not too high, over the next 3 hours”) and, when appropriate, a support message relevant to forecast and user history. Forecast delivery was random, triggered with 50% probability when information was logged, no more than once/day/user. Forecasts could be rated “Useful” or “Not Useful.” A machine learning model, trained on the initial sample, predicted the probability of each type of feedback. A second sample of 28838 forecasts were sent to 5506 users, with probability of usefulness determined by the model. Results: In the 1st sample, 42.8% of forecasts received feedback from 69.6% of users; 87.1% was “Useful.” In the 2nd sample, 63.7% of forecasts received feedback from 67.1% of users; 92.4% was “Useful.” Discussion: A new machine learning model tailored forecast delivery, reducing the “Not Useful” rate by 41.1% (from 12.9% of feedback to 7.6%). Correlations between useful forecasts and behavior change will be explored in future work. Disclosure D.R. Goldner: Employee; Self; One Drop. Stock/Shareholder; Self; One Drop. M. Heyman: Consultant; Self; Lexicon Pharmaceuticals, Inc., Lilly Diabetes, Tandem Diabetes Care. Employee; Self; One Drop. A. Hirsch: Employee; Self; Informed Data Systems Inc. Research Support; Self; Fitbit, Inc., MannKind Corporation. Stock/Shareholder; Self; Informed Data Systems Inc. C.Y. Osborn: Employee; Self; One Drop. O. Lubling: Employee; Self; One Drop. B. Huddleston: Employee; Self; One Drop. R.L. Vlaha: Employee; Self; One Drop. K. Prenger: Employee; Self; One Drop. R. Murthy: Employee; Self; One Drop. J. Starke: Employee; Self; One Drop. J. Briceno: Employee; Self; One Drop. D. Ledvina: Employee; Self; One Drop. J. Dachis: Stock/Shareholder; Self; Informed Data Systems Inc.

922-P: Diabits—An AI-Powered Smartphone Application for Blood Glucose Monitoring and Predictions

This paper presents DiaBits, an iPhone application that allows patients with diabetes to efficiently monitor and manage their blood glucose levels in real time. Using current CGM data, smartphone sensor data, and user inputs, DiaBits takes advantage of modern machine learning techniques to predict future blood glucose fluctuations up to 60 minutes in advance. These predictions, which are over 95% clinically acceptable according to standard Error Grid Analysis methods, give the patients an opportunity to take corrective action in advance in order to keep their blood glucose values within normal range. Additionally, the presence of predictive alarms and statistical tools for analysis of past data makes it easier for the users to manage their condition and have better blood glucose control. The paper discusses some of the details of the predictive approach, provides accuracy results for 1.63 million predictions made for free-living users of the application, gives a detailed overview of the available features and the interface of DiaBits, and describes some challenges related to using machine learning for real-time predictions of blood glucose behavior. Disclosure A. Hayeri: Other Relationship; Self; Bio Conscious Technologies INC.

A Coupled Insulin and Meal Effect Neuro-Fuzzy Model for The Prediction of Blood Glucose Level in Type 1 Diabetes Mellitus Patients.

Abstract Diabetes Mellitus is a metabolic disorder that affects the ability of the human body to properly utilize and regulate glucose. It is pervasive world-wide yet tenuous and costly to manage. Diabetes Mellitus is also difficult to model because it is nonlinear, dynamic and laden with mostly patient specific uncertainties. A neuro-fuzzy model for the prediction of blood glucose level in Type 1 diabetic patients using coupled insulin and meal effects is developed. This study establishes that the necessary and sufficient conditions to predict blood glucose level in a Type 1 diabetes mellitus patient are: knowledge of the patient’s insulin effects and meal effects under diverse metabolic scenarios and the transparent coupling of the insulin and meal effects. The neuro-fuzzy models were trained with data collected from a single Type 1 diabetic patient covering a period of two months. Clarke’s Error Grid Analysis (CEGA) of the model shows that 87.5% of the predictions fall into region A, while the remaining 12.5% of the predictions fall into region B within a four (4) hour prediction window. The model reveals significant variation in insulin and glucose responses as the Body Mass Index (BMI) of the patient changes.

944-P: Blood Glucose Prediction for Type 2 Diabetic Patients Using Machine Learning

The blood glucose prediction algorithms were widely studied for the patients with type 1 diabetes. We analyzed the data of 16 type 2 diabetic patients underwent continuous glucose monitoring (CGM) through deep-learning models. Feed Forward Neural Network (FFNN) and Long Short-Term Memory (LSTM) were used to model and predict the blood glucose. Blood glucose levels after 5, 15, 30, and 45 min were predicted for each patient. To evaluate the predictive power of model, the Root Mean Square Error (RMSE) was calculated by comparing the actual blood glucose and predictive blood glucose. The value of RMSE by FFNN was lower than that by LSTM. The predictive success rates for the glycemic excursion at 5 minutes by FFNN, LSTM were 96% and 95%, the predictive power of two models gradually decreased with time (Figure 1). These results suggest that the well trained deep-learning model would be possible to predict the harmful glycemic events among type 2 diabetic patient. Disclosure D. Kim: None. H. Lee: None. Y. Kim: None. S. Kim: None. S. Lee: None. S. Chun: None.

A personalized diet and exercise recommender system for type 1 diabetes self-management: An in silico study

Management of diet and exercise levels needs to be personalized for patients with Type 1 Diabetes (T1D) to reduce the number of hypoglycemia events and to achieve a good glycemic control. This study developed a model-based Recommender system that could provide an (optimal) personalized intervention on diet and exercise for T1D patients, which could be potentially implemented as a mobile application (app) for self-management of T1D in the future work. At each intervention time, the Recommender makes prediction of blood glucose based on a patient-specific model of glucose dynamics, and then provides optimal interventions, which could be a meal/snack size or a target heart rate during exercise, by minimizing a risk function with constraints under a future time horizon. Simulations were conducted to evaluate the Recommender through 30 virtual subjects generated from a modified UVa/Padova simulator with an added exercise-glucose subsystem. The performance of the Recommender was compared to two self-management schemes: the Starter scheme and the Skilled scheme, where the Skilled represents an off-line optimal scheme providing a lower bound on the risk index. Compared to the Starter, the Recommender reduced the mean Low Blood Glucose Index by 84% and reduced the Blood Glucose Risk Index by 49% (P < 0.05), and it had comparable performance as the Skilled. The Recommender also reduced the number of hypoglycemia events during and post-exercise compared to the Starter and the Skilled.

The Smart Healthcare Initiative: Blood Glucose Prediction with Wireless Sensor Network

The Smart Healthcare Initiative: Blood Glucose Prediction with Wireless Sensor Network

A Noninvasive Accurate Measurement of Blood Glucose Levels with Raman Spectroscopy of Blood in Microvessels.

Raman spectra of human skin obtained by laser excitation have been used to non-invasively detect blood glucose. In previous reports, however, Raman spectra thus obtained were mainly derived from the epidermis and interstitial fluid as a result of the shallow penetration depth of lasers in skin. The physiological process by which glucose in microvessels penetrates into the interstitial fluid introduces a time delay, which inevitably introduces errors in transcutaneous measurements of blood glucose. We focused the laser directly on the microvessels in the superficial layer of the human nailfold, and acquired Raman spectra with multiple characteristic peaks of blood, which indicated that the spectra obtained predominantly originated from blood. Incorporating a multivariate approach combining principal component analysis (PCA) and back propagation artificial neural network (BP-ANN), we performed noninvasive blood glucose measurements on 12 randomly selected volunteers, respectively. The mean prediction performance of the 12 volunteers was obtained as an RMSEP of 0.45 mmol/L and R2 of 0.95. It was no time lag between the predicted blood glucose and the actual blood glucose in the oral glucose tolerance test (OGTT). We also applied the procedure to data from all 12 volunteers regarded as one set, and the total predicted performance was obtained with an RMSEP of 0.27 mmol/L and an R2 of 0.98, which is better than that of the individual model for each volunteer. This suggested that anatomical differences between volunteer fingernails do not reduce the prediction accuracy and 100% of the predicted glucose concentrations fall within Region A and B of the Clarke error grid, allowing acceptable predictions in a clinically relevant range. The Raman spectroscopy detection of blood glucose from microvessels is of great significance of non-invasive blood glucose detection of Raman spectroscopy. This innovative method may also facilitate non-invasive detection of other blood components.

Testing of a Short-Term Blood Glucose Prediction Algorithm Using the DirecNet Database

A short-term blood glucose prediction algorithm was validated using the DirecNet clinical database. Noise at 0, 10, 15, 20, and 25% levels was added to blood glucose tracks to assess the stability of the algorithm. Computer modeling showed that the average prediction error was 2.0, 3.0, 6.6, 7.4, and 13.7%, respectively.

A precise non-invasive blood glucose measurement system using NIR spectroscopy and Huber’s regression model

Diabetes is one of the prominent diseases around the world. Presently, invasive techniques need a finger prick blood sample . A repetitively painful procedure that produces the chance of infection. To resolve this issue, non-invasive measurement approach is proposed. In this paper, an efficient NIR wave based optical detection system is proposed with optimized post-processing regression model. After real-time data analysis, it has been found that the coefficient of determination ( $$R^{2}$$ ) is improved with the value of 0.9084 using proposed regression model. Mean absolute derivative is also increased with 3.87 mg/dl corresponding to predicted blood glucose concentration. Mean absolute relative difference has exceeded to 3.25%, and average error is improved with 3.77% using proposed regression model. Average accuaracy has been analyzed 94–95% for predicted blood glucose concentration.

The effect of psychological distress on self-care intention and behaviour in young adults with type 1 diabetes.

An extended theory of planned behaviour including psychological distress was used to predict blood glucose monitoring in young adults with type 1 diabetes. Participants completed two surveys, a week apart (n = 167). Attitudes and perceived behavioural control were significantly associated with intention but subjective norm was not. Intention predicted blood glucose monitoring at T2. Higher scores in psychological distress significantly predicted poorer blood glucose monitoring at T2 but did not moderate the intention-behaviour relationship. Findings demonstrate some concerns with the theory of planned behaviour to predict self-care behaviour, yet highlight the importance of addressing psychological distress in diabetes self-management.

Blood Glucose Level Prediction for Diabetics Based on Nutrition and Insulin Administration Logs Using Personalized Mathematical Models.

According to recent surveys, the current ways of diabetics trying to estimate their insulin need based on experience and conjecture are sometimes inefficient in practice. This paper proposes a prediction algorithm and presents the validation of the model in outpatient care. The algorithm consists of two state-of-the-art models that calculate nutrition absorption and glycaemia including insulin evolution. The combined model is extended with personalized parameter training including genetic algorithm and Nelder–Mead method, and a more realistic, diurnal parameter profile as a representation of the natural biorhythm. This method implemented in a user-friendly application can help diabetics calculate their insulin need. The tests were performed on a data set including a clinical trial involving more than 20 diabetic patients. We experienced 55% improvement in the results due to model training compared to the tests based on literature parameters. In the best case, 92.5% of the predicted blood glucose level values were in the range of clinically acceptable errors, which means around 2.8 mmol/l root mean square error. The results of the validation based on outpatient data are promising compared to others found in the literature. Handling other important factors such as physical activity and stress remains a challenge for future research.

Long-acting insulin management for blood glucose prediction models

Long-acting insulin management for blood glucose prediction models

Long-acting insulin management for blood glucose prediction models.

Currently available state-of-the-art mathematical models for blood glucose level prediction have been developed for intensive care departments, and thus they do not support long-acting or ‘basal’ insulins applied typically once a day. Our goal was to adjust the current models to support basal insulin, and thus provide a short term blood glucose prediction service applicable in outpatient care, in the form of a smartphone application. We propose a method that simulates the absorption of basal insulin as a series of smaller insulin doses according to four alternative ‘dosing profiles’, instead of using a single big dose of bolus insulin. The corrected model was tested on 18 data sets originating from a clinical trial in which 16 insulin dependent patients (7 female and 9 male) used a continuous glucose monitor device to record their blood glucose levels for six days, while their meals were recorded. The prediction errors of the corrected model were compared to the errors of the original model with the usual statistical methods and the error grid analysis. We also evaluated the night periods separately from the day-time. The proposed model correction was found to reduce the error of the prediction with respect to all investigated evaluation criteria by 0.59-1.02 mmol/l, moving the average absolute error close to the error range on the measurement devices. This reduction could bring online, continuous blood glucose prediction services closer to mass deployment in lifestyle support applications for diabetics.

考虑多种因素的近红外光谱血糖预测模型对比

考虑多种因素的近红外光谱血糖预测模型对比