Future Blood Glucose Levels Research Articles

Using Artificial Intelligence to Forecast Continuous Glucose Monitor(CGM) readings for Type One Diabetes

Type One diabetes (T1D) is an incurable condition in which a person produces little to no insulin, leading to high glucose levels in the bloodstream. Advancements in technology have led to the development of Continuous Glucose Monitors (CGM), involving inserting a sensor under the skin that continuously measures glucose levels in the bloodstream. CGMs provide real-time data on glucose levels and patterns of fluctuation and have evolved as an essential part of diabetes management. Current CGMs cannot forecast glucose levels. In this study, we hypothesized that machine learning algorithms can be utilized to forecast glucose values using data provided by the CGM device. This could help T1D patients optimize medication timing, dietary adjustments, and physical activity to improve metabolic control and prevent T1D complications. Our study used publicly available data from a study that compared CGMs with and without routine blood glucose monitoring in adults with T1D. We used different AI models to provide 30-minute forecasts of blood glucose levels to show how model structure and parameters affect Root Mean Squared Error (RMSE), and whether the error can be reduced to 5% as per medical standards. Our research used RandomForest as a baseline algorithm and LSTM, a recurrent neural network. Our results demonstrated that the LSTM model achieved an error of under 5% in predicting future blood glucose levels. Hence, researchers can use LSTM models to enhance CGMs to aid T1D patients. The results of our study supported our hypothesis that AI can be used to predict glucose levels.

Combining wavelet transform with convolutional neural networks for hypoglycemia events prediction from CGM data

Estimating future blood glucose levels is an essential and challenging task for people with diabetes. It must be carried out based on variables such as current glucose, carbohydrate intake, physical activity, and insulin dosing. Accurate estimation is essential to maintain glucose values in a healthy range and avoid dangerous events of low glucose levels (hypoglycemia) and extremely high glucose values (hyperglycemia). Those situations maintained in time can cause not only permanent long-term damage but also short-term complications and even the death of the person. This paper proposes a new method to predict and detect hypoglycemic events over a 24-h time horizon. The technique combines applying the wavelet transform to glucose time series and deep learning convolutional neural networks. We have experimented with real data collected from 20 different people with type 1 diabetes. Our technique can also be applied to predict hyperglycemia. We incorporate a data augmentation technique consisting of a rolling windows system that improves the accuracy of the prediction. The uncertainty of the data is considered by the addition of controlled noise. The results show that the predictions obtained are accurate (higher than 88% of accuracy, sensitivity, specificity, and precision), confirming the effectiveness of the proposed method.

A 2-dimensional model framework for blood glucose prediction based on iterative learning control architecture.

The accurate, timely, and personalized prediction for future blood glucose (BG) levels is undoubtedly needed for further advancement of diabetes management technologies. Human inherent circadian rhythm and regular lifestyle resulting in similarity of daily glycemic dynamics play a positive role in the prediction of blood glucose. Inspired by the iterative learning control (ILC) method in the field of automatic control, a 2-dimensional (2-D) model framework is constructed to predict the future blood glucose levels by taking both the short-range information within a day (intra-day) and long-range information between days (inter-day) into account. In this framework, the radial basis function neural network was applied to capture nonlinear relationships in glycemic metabolism, that is, short-range temporal dependence and long-range contemporaneous dependence on previous days. We build models for each patient, and the models were tested on the in silico datasets at various prediction horizons (PHs). The learning model developed in the 2-D framework successfully increases the accuracy and reduces the delay of predictions. This modeling framework provides a new point of view for BG level prediction and contributes to the development of personalized glucose management, such as hypoglycemia warning and glycemic control.

Utilizing deep neural network for web-based blood glucose level prediction system

Machine learning algorithms can be used to forecast future blood glucose (BG) levels for diabetes patients, according to recent studies. In this study, dataset from continuous glucose monitoring (CGM) system was used as the sole input for the machine learning models. To forecast blood glucose levels 15, 30, and 45 minutes in the future, we suggested deep neural network (DNN) and tested it on 7 patients with type 1 diabetes (T1D). The suggested prediction model was evaluated against a variety of machine learning models, such as k-nearest neighbor (KNN), support vector regression (SVR), decision tree (DT), adaptive boosting (AdaBoost), random forest (RF), and eXtreme gradient boosting (XGBoost). The experimental findings demonstrated that the proposed DNN model outperformed all other models, with average root mean square errors (RMSEs) of 17.295, 25.940, and 35.146 mg/dL over prediction horizons (PHs) of 15, 30, and 45 minutes, respectively. Additionally, we have included the suggested prediction model in web-based blood glucose level prediction tools. By using this web-based system, patients may readily acquire their future blood glucose levels, allowing for the generation of preventative alarms prior to crucial hypoglycemia or hyperglycemic situations.

Patient-Reported Usefulness and Challenges in Using Hypoglycemia-Informing Features of Continuous Glucose Monitors to Manage Hypoglycemia.

The purpose of this study series, which involves a questionnaire survey and qualitative interviews, was to (a) evaluate patient-reported usefulness of continuous glucose monitor (CGM) hypoglycemia-informing features and (b) identify challenges in using these features (ie, CGM glucose numbers, trend arrows, trend graphs, and hypoglycemia alarms) during hypoglycemia in adults with type 1 diabetes (T1DM). A cross-sectional questionnaire survey study was conducted with adults who have T1DM and were using CGMs to assess the perceived usefulness of hypoglycemia-informing features. A semistructured interview study with T1DM CGM-using adults and inductive thematic analysis were subsequently performed to identify challenges in using CGM hypoglycemia-informing features to manage hypoglycemia. In the survey study (N = 252), the CGM glucose numbers, trend arrows, trend graphs, and hypoglycemia alarms were found to be very useful by 79%, 70%, 43%, and 64% of participants, respectively. Several challenges in using these features to manage hypoglycemia were identified in the qualitative study (N = 23): (1) hypoglycemia information not fully reliable,; (2) unpredictability of future blood glucose levels, (3) lack of awareness about how information can be used, and (4) disruptions associated with information. Although the majority of T1DM adults found their CGMs' hypoglycemia-informing features helpful, challenges in optimally using these features persisted. Targeted knowledge and behavioral interventions could improve CGM use to reduce hypoglycemia.

Analyzing the Performance of Transformers for the Prediction of the Blood Glucose Level Considering Imputation and Smoothing

In this paper we investigate the effect of two preprocessing techniques, data imputation and smoothing, in the prediction of blood glucose level in type 1 diabetes patients, using a novel deep learning model called Transformer. We train three models: XGBoost, a one-dimensional convolutional neural network (1D-CNN), and the Transformer model to predict future blood glucose levels for a 30-min horizon using a 60-min time series history in the OhioT1DM dataset. We also compare four methods of handling missing time series data during the model training: hourly mean, linear interpolation, cubic interpolation, and spline interpolation; and two smoothing techniques: Kalman smoothing and smoothing splines. Our experiments show that the Transformer performs better than XGBoost and 1D-CNN when only continuous glucose monitoring (CGM) is used as a predictor, and that it is very competitive against XGBoost when CGM and carbohydrate intake from the meal are used to predict blood glucose level. Overall, our results are more accurate than those appearing in the literature.

Feature Transformation for Efficient Blood Glucose Prediction in Type 1 Diabetes Mellitus Patients.

Diabetes Mellitus, a metabolic disease, causes the body to lose control over blood glucose regulation. With recent advances in self-monitoring systems, a patient can access their personalized glycemic profile and may utilize it for efficient prediction of future blood glucose levels. An efficient diabetes management system demands the accurate estimation of blood glucose levels, which, apart from using an appropriate prediction algorithm, depends on discriminative data representation. In this research work, a transformation of event-based data into discriminative continuous features is proposed. Moreover, a multi-layered long short-term memory (LSTM)-based recurrent neural network is developed for the prediction of blood glucose levels in patients with type 1 diabetes. The proposed method is used to forecast the blood glucose level on a prediction horizon of 30 and 60 min. The results are evaluated for three patients using the Ohio T1DM dataset. The proposed scheme achieves the lowest RMSE score of 14.76 mg/dL and 25.48 mg/dL for prediction horizons of 30 min and 60 min, respectively. The suggested methodology can be utilized in closed-loop systems for precise insulin delivery to type 1 patients for better glycemic control.

Multi-step ahead predictive model for blood glucose concentrations of type-1 diabetic patients

Continuous monitoring of blood glucose (BG) levels is a key aspect of diabetes management. Patients with Type-1 diabetes (T1D) require an effective tool to monitor these levels in order to make appropriate decisions regarding insulin administration and food intake to keep BG levels in target range. Effectively and accurately predicting future BG levels at multi-time steps ahead benefits a patient with diabetes by helping them decrease the risks of extremes in BG including hypo- and hyperglycemia. In this study, we present a novel multi-component deep learning model BG-Predict that predicts the BG levels in a multi-step look ahead fashion. The model is evaluated both quantitatively and qualitatively on actual blood glucose data for 97 patients. For the prediction horizon (PH) of 30 mins, the average values for root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and normalized mean squared error (NRMSE) are 23.22 pm 6.39 mg/dL, 16.77 ± 4.87 mg/dL, 12.84 pm 3.68 and 0.08 pm 0.01 respectively. When Clarke and Parkes error grid analyses were performed comparing predicted BG with actual BG, the results showed average percentage of points in Zone A of 80.17 pm 9.20 and 84.81 pm 6.11, respectively. We offer this tool as a mechanism to enhance the predictive capabilities of algorithms for patients with T1D.

Learning Carbohydrate Digestion and Insulin Absorption Curves Using Blood Glucose Level Prediction and Deep Learning Models.

Type 1 diabetes is a chronic disease caused by the inability of the pancreas to produce insulin. Patients suffering type 1 diabetes depend on the appropriate estimation of the units of insulin they have to use in order to keep blood glucose levels in range (considering the calories taken and the physical exercise carried out). In recent years, machine learning models have been developed in order to help type 1 diabetes patients with their blood glucose control. These models tend to receive the insulin units used and the carbohydrate taken as inputs and generate optimal estimations for future blood glucose levels over a prediction horizon. The body glucose kinetics is a complex user-dependent process, and learning patient-specific blood glucose patterns from insulin units and carbohydrate content is a difficult task even for deep learning-based models. This paper proposes a novel mechanism to increase the accuracy of blood glucose predictions from deep learning models based on the estimation of carbohydrate digestion and insulin absorption curves for a particular patient. This manuscript proposes a method to estimate absorption curves by using a simplified model with two parameters which are fitted to each patient by using a genetic algorithm. Using simulated data, the results show the ability of the proposed model to estimate absorption curves with mean absolute errors below 0.1 for normalized fast insulin curves having a maximum value of 1 unit.

Blood Glucose Level Forecasting on Type-1-Diabetes Subjects during Physical Activity: A Comparative Analysis of Different Learning Techniques.

Background: Type 1 Diabetes Mellitus (T1DM) is a widespread chronic disease in industrialized countries. Preventing blood glucose levels from exceeding the euglycaemic range would reduce the incidence of diabetes-related complications and improve the quality of life of subjects with T1DM. As a consequence, in the last decade, many Machine Learning algorithms aiming to forecast future blood glucose levels have been proposed. Despite the excellent performance they obtained, the prediction of abrupt changes in blood glucose values produced during physical activity (PA) is still one of the main challenges. Methods: A Jump Neural Network was developed in order to overcome the issue of predicting blood glucose values during PA. Three learning configurations were developed and tested: offline training, online training, and online training with reinforcement. All configurations were tested on six subjects suffering from T1DM that held regular PA (three aerobic and three anaerobic) and exploited Continuous Glucose Monitoring (CGM). Results: The forecasting performance was evaluated in terms of the Root-Mean-Squared-Error (RMSE), according to a paradigm of Precision Medicine. Conclusions: The online learning configurations performed better than the offline configuration in total days but not on the only CGM associated with the PA; thus, the results do not justify the increased computational burden because the improvement was not significant.

LSTMs and Deep Residual Networks for Carbohydrate and Bolus Recommendations in Type 1 Diabetes Management.

To avoid serious diabetic complications, people with type 1 diabetes must keep their blood glucose levels (BGLs) as close to normal as possible. Insulin dosages and carbohydrate consumption are important considerations in managing BGLs. Since the 1960s, models have been developed to forecast blood glucose levels based on the history of BGLs, insulin dosages, carbohydrate intake, and other physiological and lifestyle factors. Such predictions can be used to alert people of impending unsafe BGLs or to control insulin flow in an artificial pancreas. In past work, we have introduced an LSTM-based approach to blood glucose level prediction aimed at “what-if” scenarios, in which people could enter foods they might eat or insulin amounts they might take and then see the effect on future BGLs. In this work, we invert the “what-if” scenario and introduce a similar architecture based on chaining two LSTMs that can be trained to make either insulin or carbohydrate recommendations aimed at reaching a desired BG level in the future. Leveraging a recent state-of-the-art model for time series forecasting, we then derive a novel architecture for the same recommendation task, in which the two LSTM chain is used as a repeating block inside a deep residual architecture. Experimental evaluations using real patient data from the OhioT1DM dataset show that the new integrated architecture compares favorably with the previous LSTM-based approach, substantially outperforming the baselines. The promising results suggest that this novel approach could potentially be of practical use to people with type 1 diabetes for self-management of BGLs.

Blood glucose prediction model for type 1 diabetes based on artificial neural network with time-domain features

Predicting future blood glucose (BG) levels for diabetic patients will help them avoid potentially critical health issues. We demonstrate the use of machine learning models to predict future blood glucose levels given a history of blood glucose values as the single input parameter. We propose an Artificial Neural Network (ANN) model with time-domain attributes to predict blood glucose levels 15, 30, 45 and 60 min in the future. Initially, the model’s features are selected based on the previous 30 min of BG measurements before a trained model is generated for each patient. These features are combined with time-domain attributes to give additional inputs to the proposed ANN. The prediction model was tested on 12 patients with Type 1 diabetes (T1D) and the results were compared with other data-driven models including the Support Vector Regression (SVR), K-Nearest Neighbor (KNN), C4.5 Decision Tree (DT), Random Forest (RF), Adaptive Boosting (AdaBoost) and eXtreme Gradient Boosting (XGBoost) models. Our results show that the proposed BG prediction model that is based on an ANN outperformed all other models with an average Root Mean Square Error (RMSE) of 2.82, 6.31, 10.65 and 15.33 mg/dL for Prediction Horizons (PHs) of 15, 30, 45 and 60 min, respectively. Our testing showed that combining time-domain attributes into the input data resulted in enhanced performance of majority of prediction models. The implementation of proposed prediction model allows patients to obtain future blood glucose levels, so that the preventive alerts can be generated before critical hypoglycemic/ hyperglycemic events occur.

Parallel software architecture for the next generation of glucose monitoring

Diabetes is a widespread disease. Elevated blood glucose levels continuously damage multiple organs in the long-term. In the short-term, hypo- and hyperglycemic shocks are acute risks. Diabetes patients monitor their glucose level using continuous glucose monitoring systems. Based on their measured glucose level, the patient take insulin to lower their blood glucose level. With the advances in mobile computing, an increasing number of diabetes patients engage in self-built systems. They read their glucose levels from glucose-monitoring systems and calculate their insulin dosage based on the measured levels. The self-built nature of such a system raises a number of medical and software engineering concerns. Therefore, we propose a software architecture for the next generation of glucose monitoring. The proposed architecture builds on the principles of the high-level architecture. We decompose the entire glucose monitoring system to basic elements, which are either real or simulated. This opens the proposed architecture to software engineering, simulation, and fault-tolerance research. As a proof of concept, we present an illustrative configuration of the implemented software architecture that predicts future blood glucose levels 15 minutes in advance for type-1 diabetes patients. All relative errors are in the A+B zones of Clarke and Parkes error grids, with almost 95% of errors in the safest A-zones of both grids.

Blood glucose level prediction based on support vector regression using mobile platforms.

The correct treatment of diabetes is vital to a patient's health: Staying within defined blood glucose levels prevents dangerous short- and long-term effects on the body. Mobile devices informing patients about their future blood glucose levels could enable them to take counter-measures to prevent hypo or hyper periods. Previous work addressed this challenge by predicting the blood glucose levels using regression models. However, these approaches required a physiological model, representing the human body's response to insulin and glucose intake, or are not directly applicable to mobile platforms (smart phones, tablets). In this paper, we propose an algorithm for mobile platforms to predict blood glucose levels without the need for a physiological model. Using an online software simulator program, we trained a Support Vector Regression (SVR) model and exported the parameter settings to our mobile platform. The prediction accuracy of our mobile platform was evaluated with pre-recorded data of a type 1 diabetes patient. The blood glucose level was predicted with an error of 19 % compared to the true value. Considering the permitted error of commercially used devices of 15 %, our algorithm is the basis for further development of mobile prediction algorithms.

Averting iatrogenic hypoglycemia through glucose prediction in clinical practice: Progress towards a new procedure in diabetes

Background Hypoglycemia is a risk factor common to all insulin therapy. The hypothesis is that efforts to reduce or prevent this adverse side effect may fail because providers generally lack the resources to predict not only future blood glucose levels but also future risks of hypoglycemia. This lack has been remedied. A controlled study was undertaken to test the hypothesis. Methods Twenty-two insulin dependent subjects suffering more than one (1) episode/week of hypoglycemia with similar insulin regimens, similar diabetes education and similar self-management training participated in this study. For all subjects, a remote monitoring resource (registry and database) was used to capture daily SMBG and afford a return path for provider interventions and decision support. Identical telemedical methods were used which differed only for the provider either by the presence (prediction group) or by the absence (control group) of an on-screen, visual display of predicted glycemia and predicted risks of hypoglycemia. The study lasted 2 months. Results Over an average of 41 days from baseline to follow up and while using the glycemic prediction resource, providers intervened more effectively in the prediction group reducing rates of hypoglycemia nine-fold ( P < 0.0001) and insulin therapy by just −9 U/day ( P < 0.01). Mean pre-meal glycemia was not compromised. Over 61 days from baseline to final follow up but without glycemic predictions in the control group, providers’ interventions were less effective and resulted in no net changes in rates of hypoglycemia, daily insulin therapy, or mean pre-meal glycemia. Conclusions Given knowledge of future glycemia and future risks of hypoglycemia, providers in clinical practice can now avert iatrogenic hypoglycemia in less than 2 months. A shared diabetes data center furnishing remote data capture and decision support is fundamental to the implementation of this as a new clinical procedure in diabetes.

Home blood glucose prediction: clinical feasibility and validation in islet cell transplantation candidates

Diabetic subjects do home monitoring to substantiate their success (or failure) in meeting blood glucose targets set by their providers. To succeed, patients require decision support, which, until now, has not included knowledge of future blood glucose levels or of hypoglycaemia. To remedy this, we devised a glucose prediction engine. This study validates its predictions. The prediction engine is a computer program that accesses a central database in which daily records of self-monitored blood glucose data and life-style parameters are stored. New data are captured by an interactive voice response server on-line 24 h a day, 7 days a week. Study subjects included 24 patients with debilitating hypoglycaemia (unawareness), which qualified them for islet cell transplantation. Comparison of each prediction with the actually observed data was done using a Clarke Error Grid (CEG). Patients and providers were blinded as to the predictions. Prior to transplantation, a total of 31,878 blood glucose levels were reported by the study subjects. Some 31,353 blood glucose predictions were made by the engine on a total of 8,733 days-used. Of these, 79.4% were in the clinically acceptable Zones of the CEG. Of 728 observed episodes of hypoglycaemia, 384 were predicted. After transplantation, a total of 45,529 glucose measurements were reported on a total of 12,906 days-used. Some 42,316 glucose predictions were made, of which 97.5% were in the acceptable CEG Zones A and B. Successful transplantation eliminated hypoglycaemia, improved glycaemic control, lowered HbA(1)c and freed 10 of 24 patients from daily insulin therapy. It is clinically feasible to generate valid predictions of future blood glucose levels. Prediction accuracy is related to glycaemic stability. Risk of hypoglycaemia can be predicted. Such knowledge may be useful in self-management.