Hypoglycemia early alarm systems based on recursive autoregressive partial least squares models

Generalization of a Deep Learning Model for Continuous Glucose Monitoring-Based Hypoglycemia Prediction: Algorithm Development and Validation Study

Background

Predicting hypoglycemia while maintaining a low false alarm rate is a challenge for the wide adoption of continuous glucose monitoring (CGM) devices in diabetes management. One small study suggested that a deep learning model based on the long short-term memory (LSTM) network had better performance in hypoglycemia prediction than traditional machine learning algorithms in European patients with type 1 diabetes. However, given that many well-recognized deep learning models perform poorly outside the training setting, it remains unclear whether the LSTM model could be generalized to different populations or patients with other diabetes subtypes.

Objective

The aim of this study was to validate LSTM hypoglycemia prediction models in more diverse populations and across a wide spectrum of patients with different subtypes of diabetes.

Methods

We assembled two large data sets of patients with type 1 and type 2 diabetes. The primary data set including CGM data from 192 Chinese patients with diabetes was used to develop the LSTM, support vector machine (SVM), and random forest (RF) models for hypoglycemia prediction with a prediction horizon of 30 minutes. Hypoglycemia was categorized into mild (glucose=54-70 mg/dL) and severe (glucose<54 mg/dL) levels. The validation data set of 427 patients of European-American ancestry in the United States was used to validate the models and examine their generalizations. The predictive performance of the models was evaluated according to the sensitivity, specificity, and area under the receiver operating characteristic curve (AUC).

Results

For the difficult-to-predict mild hypoglycemia events, the LSTM model consistently achieved AUC values greater than 97% in the primary data set, with a less than 3% AUC reduction in the validation data set, indicating that the model was robust and generalizable across populations. AUC values above 93% were also achieved when the LSTM model was applied to both type 1 and type 2 diabetes in the validation data set, further strengthening the generalizability of the model. Under different satisfactory levels of sensitivity for mild and severe hypoglycemia prediction, the LSTM model achieved higher specificity than the SVM and RF models, thereby reducing false alarms.

Conclusions

Our results demonstrate that the LSTM model is robust for hypoglycemia prediction and is generalizable across populations or diabetes subtypes. Given its additional advantage of false-alarm reduction, the LSTM model is a strong candidate to be widely implemented in future CGM devices for hypoglycemia prediction.

A novel hypoglycemia alarm framework for type 2 diabetes with high glycemic variability

In patients with type 2 diabetes (T2D), accurate prediction of hypoglycemic events is crucial for maintaining glycemic control and reducing their frequency. However, individuals with high blood glucose variability experience significant fluctuations over time, posing a challenge for early warning models that rely on static features. This article proposes a novel hypoglycemia early alarm framework based on dynamic feature selection. The framework incorporates domain knowledge and introduces multi-scale blood glucose features, including predicted values, essential for early warnings. To address the complexity of the feature matrix, a dynamic feature selection mechanism (Relief-SVM-RFE) is designed to effectively eliminate redundancy. Furthermore, the framework employs online updates for the random forest model, enhancing the learning of more relevant features. The effectiveness of the framework was evaluated using a clinical dataset. For T2D patients with a high coefficient of variation (CV), the framework achieved a sensitivity of 81.15% and specificity of 98.14%, accurately predicting most hypoglycemic events. Notably, the proposed method outperformed other existing approaches. These results indicate the feasibility of anticipating hypoglycemic events in T2D patients with high CV using this innovative framework.

A Hypoglycemia Early Alarm Method for Patients with Type 1 Diabetes Based on Multi-dimensional Sequential Pattern Mining

Hypoglycemia is a limiting factor for blood glucose management. Serious symptoms such as seizures, and coma may occur during severe hypoglycemia, and nocturnal hypoglycemia is particularly dangerous for patients with type 1 diabetes (T1D). An effective early alarm method is essential for hypoglycemia prevention but challenging, as blood glucose is affected by many factors and the hypoglycemia sequence patterns vary from person to person. In this paper, we proposed a hypoglycemia early alarm method for mining the hidden information in blood glucose based on multi-dimensional sequential pattern mining. The blood glucose, meal, and insulin time series information were used to construct a multi-dimensional database, then the UniSeq algorithm was used to extract multi-dimensional hypoglycemia sequence patterns. Hypoglycemia early alarm was realized through pattern matching with real-time blood glucose. The public OhioT1DM dataset was used for performance evaluation. The experiment results were: 75.76% Sensitivity, 75% Precision, 75.38% F1 score, and 25.17 minutes early alarm time. The result verified that multi-dimensional sequential pattern mining can extract more hidden information and demonstrate more potential significance in providing comprehensive diagnosis support for personalized treatment. Furthermore, early alarms for potential hypoglycemia can also reserve sufficient time for blood glucose management.

Prior Informed Regularization of Recursively Updated Latent-Variables-Based Models with Missing Observations

Many data-driven modeling techniques identify locally valid, linear representations of time-varying or nonlinear systems, and thus the model parameters must be adaptively updated as the operating conditions of the system vary, though the model identification typically does not consider prior knowledge. In this work, we propose a new regularized partial least squares (rPLS) algorithm that incorporates prior knowledge in the model identification and can handle missing data in the independent covariates. This latent variable (LV) based modeling technique consists of three steps. First, a LV-based model is developed on the historical time series data. In the second step, the missing observations in the new incomplete data sample are estimated. Finally, the future values of the outputs are predicted as a linear combination of estimated scores and loadings. The model is recursively updated as new data are obtained from the system. The performance of the proposed rPLS and rPLS with exogenous inputs (rPLSX) algorithms are evaluated by modeling variations in glucose concentration (GC) of people with Type 1 diabetes (T1D) in response to meals and physical activities for prediction windows up to one hour, or 12 sampling instances, into the future. The proposed rPLS family of GC prediction models are evaluated with both in-silico and clinical experiment data and compared with the performance of recursive time series and kernel-based models. The root mean squared error (RMSE) with simulated subjects in the multivariable T1D simulator where physical activity effects are incorporated in GC variations are 2.52 and 5.81 mg/dL for 30 and 60 mins ahead predictions (respectively) when information for all meals and physical activities are used, increasing to 2.70 and 6.54 mg/dL (respectively) when meals and activities occurred, but the information is with-held from the modeling algorithms. The RMSE is 10.45 and 14.48 mg/dL for clinical study with prediction horizons of 30 and 60 mins, respectively. The low RMSE values demonstrate the effectiveness of the proposed rPLS approach compared to the conventional recursive modeling algorithms.

Stacked LSTM based deep recurrent neural network with kalman smoothing for blood glucose prediction

BACKGROUND

Blood glucose (BG) management is crucial for type-1 diabetes patients resulting in the necessity of reliable artificial pancreas or insulin infusion systems. In recent years, deep learning techniques have been utilized for a more accurate BG level prediction system. However, continuous glucose monitoring (CGM) readings are susceptible to sensor errors. As a result, inaccurate CGM readings would affect BG prediction and make it unreliable, even if the most optimal machine learning model is used.

METHODS

In this work, we propose a novel approach to predicting blood glucose level with a stacked Long short-term memory (LSTM) based deep recurrent neural network (RNN) model considering sensor fault. We use the Kalman smoothing technique for the correction of the inaccurate CGM readings due to sensor error.

RESULTS

For the OhioT1DM (2018) dataset, containing eight weeks' data from six different patients, we achieve an average RMSE of 6.45 and 17.24 mg/dl for 30 min and 60 min of prediction horizon (PH), respectively.

CONCLUSIONS

To the best of our knowledge, this is the leading average prediction accuracy for the ohioT1DM dataset. Different physiological information, e.g., Kalman smoothed CGM data, carbohydrates from the meal, bolus insulin, and cumulative step counts in a fixed time interval, are crafted to represent meaningful features used as input to the model. The goal of our approach is to lower the difference between the predicted CGM values and the fingerstick blood glucose readings-the ground truth. Our results indicate that the proposed approach is feasible for more reliable BG forecasting that might improve the performance of the artificial pancreas and insulin infusion system for T1D diabetes management.

A multi-level hypoglycemia early alarm system based on sequence pattern mining

Background

Early alarm of hypoglycemia, detection of asymptomatic hypoglycemia, and effective control of blood glucose fluctuation make a great contribution to diabetic treatment. In this study, we designed a multi-level hypoglycemia early alarm system to mine potential information in Continuous Glucose Monitoring (CGM) time series and improve the overall alarm performance for different clinical situations.

Methods

Through symbolizing the historical CGM records, the Prefix Span was adopted to obtain the early alarm/non-alarm frequent sequence libraries of hypoglycemia events. The longest common subsequence was used to remove the common frequent sequence for achieving the hypoglycemia early alarm in different clinical situations. Then, the frequent sequence pattern libraries with different risk thresholds were designed as the core module of the proposed multi-level hypoglycemia early alarm system.

Results

The model was able to predict hypoglycemia events in the clinical dataset of level-I (sensitivity 85.90%, false-positive 23.86%, miss alarm rate 14.10%, average early alarm time 20.61 min), level-II (sensitivity 80.36%, false-positive 17.37%, miss alarm rate 19.63%, average early alarm time 27.66 min), and level-III (sensitivity 78.07%, false-positive 13.59%, miss alarm rate 21.93%, average early alarm time 33.80 min), respectively.

Conclusions

The proposed approach could effectively predict hypoglycemia events based on different risk thresholds to meet different prevention and treatment requirements. Moreover, the experimental results confirm the practicality and prospects of the proposed early alarm system, which reflects further significance in personalized medicine for hypoglycemia prevention.

An Artificial Neural Network-based Predictive Model to Support Optimization of Inpatient Glycemic Control

Background: Achieving glycemic control in critical care patients is of paramount importance, and has been linked to reductions in mortality, intensive care unit (ICU) length of stay, and morbidities such as infection. The myriad of illnesses and patient conditions render maintenance of glycemic control very challenging in this setting. Materials and Methods: This study involved collection of continuous glucose monitoring (CGM) data, and other associated measures, from the electronic medical records of 127 patients for the first 72 h of ICU care who upon admission to the ICU had a diagnosis of type 1 (n = 8) or type 2 diabetes (n = 97) or a glucose value >150 mg/dL (n = 22). A neural network-based model was developed to predict a complete trajectory of glucose values up to 135 min ahead of time. Model accuracy was validated using data from 15 of the 127 patients who were not included in the model training set to simulate model performance in real-world health care settings. Results: Predictive models achieved an improved accuracy and performance compared with previous models that were reported by our research team. Model error, expressed as mean absolute difference percent, was 10.6% with respect to interstitial glucose values (CGM) and 15.9% with respect to serum blood glucose values collected 135 min in the future. A Clarke Error Grid Analysis of model predictions with respect to the reference CGM and blood glucose measurements revealed that >99% of model predictions could be regarded as clinically acceptable and would not lead to inaccurate insulin therapy or treatment recommendations. Conclusion: The noted clinical acceptability of these models illustrates their potential utility within a clinical decision support system to assist health care providers in the optimization of glycemic management in critical care patients.

Hypoglycemia Prevention via Personalized Glucose-Insulin Models Identified in Free-Living Conditions

BACKGROUND

The objective of this research is to show the effectiveness of individualized hypoglycemia predictive alerts (IHPAs) based on patient-tailored glucose-insulin models (PTMs) for different subjects. Interpatient variability calls for PTMs that have been identified from data collected in free-living conditions during a one-month trial.

METHODS

A new impulse-response (IR) identification technique has been applied to free-living data in order to identify PTMs that are able to predict the future glucose trends and prevent hypoglycemia events. Impulse response has been applied to seven patients with type 1 diabetes (T1D) of the University of Amsterdam Medical Centre. Individualized hypoglycemia predictive alert has been designed for each patient thanks to the good prediction capabilities of PTMs.

RESULTS

The PTMs performance is evaluated in terms of index of fitting (FIT), coefficient of determination, and Pearson's correlation coefficient with a population FIT of 63.74%. The IHPAs are evaluated on seven patients with T1D with the aim of predicting in advance (between 45 and 10 minutes) the unavoidable hypoglycemia events; these systems show better performance in terms of sensitivity, precision, and accuracy with respect to previously published results.

CONCLUSION

The proposed work shows the successful results obtained applying the IR to an entire set of patients, participants of a one-month trial. Individualized hypoglycemia predictive alerts are evaluated in terms of hypoglycemia prevention: the use of a PTM allows to detect 84.67% of the hypoglycemia events occurred during a one-month trial on average with less than 0.4% of false alarms. The promising prediction capabilities of PTMs can be a key ingredient for new generations of individualized model predictive control for artificial pancreas.

An ARIMA Model With Adaptive Orders for Predicting Blood Glucose Concentrations and Hypoglycemia

The continuous glucose monitoring system is an effective tool, which enables the users to monitor their blood glucose (BG) levels. Based on the continuous glucose monitoring (CGM) data, we aim at predicting future BG levels so that appropriate actions can be taken in advance to prevent hyperglycemia or hypoglycemia. Due to the time-varying nonstationarity of CGM data, verified by Augmented Dickey-Fuller test and analysis of variance, an autoregressive integrated moving average (ARIMA) model with an adaptive identification algorithm of model orders is proposed in the prediction framework. Such identification algorithm adaptively determines the model orders and simultaneously estimates the corresponding parameters using Akaike Information Criterion and least square estimation. A case study is conducted with the CGM data of diabetics under daily living conditions to analyze the prediction performance of the proposed model together with the early hypoglycemic alarms. Results show that the proposed model outperforms the adaptive univariate model and ARIMA model.

Hypoglycemia Detection and Carbohydrate Suggestion in an Artificial Pancreas

Fear of hypoglycemia is a major concern for many patients with type 1 diabetes and affects patient decisions for use of an artificial pancreas system. We propose an alternative way for prevention of hypoglycemia by issuing predictive hypoglycemia alarms and encouraging patients to consume carbohydrates in a timely manner. The algorithm has been tested on 6 subjects (3 males and 3 females, age 24.2 ± 4.5 years, weight 79.2 ± 16.2 kg, height 172.7 ± 9.4 cm, HbA1C 7.3 ± 0.48%, duration of diabetes 209.2 ± 87.9 months) over 3-day closed-loop clinical experiments as part of a multivariable artificial pancreas control system. Over 6 three-day clinical experiments, there were only 5 real hypoglycemia episodes, of which only 1 hypoglycemia episode occurred due to being missed by the proposed algorithm. The average hypoglycemia alarms per day and per subject was 3. Average glucose value when the first alarms were triggered was recorded to be 117 ± 30.6 mg/dl. Average carbohydrate consumption per alarm was 14 ± 7.8 grams. Our results have shown that most low glucose concentrations can be predicted in advance and the glucose levels can be raised back to the desired levels by consuming an appropriate amount of carbohydrate. The proposed algorithm is able to prevent most hypoglycemic events by suggesting appropriate levels of carbohydrate consumption before the actual occurrence of hypoglycemia.

Impact of Type 1 Diabetes Technology on Family Members/Significant Others of People With Diabetes

BACKGROUND

The aim was to explore the impact of diabetes-related technology to ensure that such devices are used in a way that returns maximum benefit from a medical and psychological perspective.

METHOD

Spouses and caregivers of people with type 1 diabetes were invited to complete an online questionnaire about their experiences with diabetes technologies used by their family members. Participants were recruited via the Glu online community website. Questions explored impact on daily living, frequency and severity of hypoglycemia, and diabetes-related distress.

RESULTS

In all, 100 parents/caregivers and 74 partners participated in this survey. Average (mean) duration of living with a person with type 1 diabetes was 16 years (SD = 13) for partners, with duration of diabetes for children being 4.2 ± 3.2 years. Average duration of current therapy was 8.3 ± 7.3 years for adults and 3.4 ± 2.9 years for children. Of the participants, 86% partners and 82% parents/caregivers reported diabetes technology had made it easier for their family members to achieve blood glucose targets. Compared to partners, parents/caregivers reported more negative emotions (P < .001) and decreased well-being (P < .001) related to their family members type 1 diabetes. Diabetes-related distress was common, as was sleep disturbance associated with device alarms and fear of hypoglycemia. Reduced frequency and severity of hypoglycemia related to device use was reported by approximately half of participants.

CONCLUSION

There is little doubt about the medical benefit of diabetes technologies and their uptake is increasing but some downsides were reported. Barriers to uptake of technologies lie beyond the mechanics of diabetes management. Supporting users in using diabetes technology to achieve the best possible glycemic control, in the context of their own life, is crucial. Furthermore, understanding these issues with input from the type 1 diabetes community including family members and caregivers will help innovation and design of new technology.

Clinical results of an automated artificial pancreas using technosphere inhaled insulin to mimic first-phase insulin secretion

OBJECTIVE

The purpose of this study was to investigate whether or not adding a fixed preprandial dose of inhaled insulin to a fully automated closed loop artificial pancreas would improve the postprandial glucose control without adding an increased risk of hypoglycemia.

RESEARCH DESIGN AND METHODS

Nine subjects with T1DM were recruited for the study. The patients were on closed-loop control for 24 hours starting around 4:30 pm. Mixed meals (~50 g CHO) were given at 6:30 pm and 7:00 am the following day. For the treatment group each meal was preceded by the inhalation of one 10 U dose of Technosphere Insulin (TI). Subcutaneous insulin delivery was controlled by a zone model predictive control algorithm (zone-MPC). At 11:00 am, the patient exercised for 30 ± 5 minutes at 50% of predicted heart rate reserve.

RESULTS

The use of TI resulted in increasing the median percentage time in range (70-180 mg/dl, BG) during the 5-hour postprandial period by 21.6% (81.6% and 60% in the with/without TI cases, respectively, P = .06) and reducing the median postprandial glucose peak by 33 mg/dl (172 mg/dl and 205 mg/dl in the with and without TI cases, respectively, P = .004). The median percentage time in range 80-140 mg/dl during the entire study period was 67.5% as compared to percentage time in range without the use of TI of 55.2% (P = .03).

CONCLUSIONS

Adding preprandial TI (See video supplement) to an automated closed-loop AP system resulted in superior postprandial control as demonstrated by lower postprandial glucose exposure without addition hypoglycemia.

A practical and novel “standard addition” strategy to screen pharmacodynamic components in traditional Chinese medicine using Heishunpian as an example

The standard addition strategy allows accurate pharmacodynamic compounds screening and embodies the systematic nature of TCM.

An integrated multivariable artificial pancreas control system

The objective was to develop a closed-loop (CL) artificial pancreas (AP) control system that uses continuous measurements of glucose concentration and physiological variables, integrated with a hypoglycemia early alarm module to regulate glucose concentration and prevent hypoglycemia. Eleven open-loop (OL) and 9 CL experiments were performed. A multivariable adaptive artificial pancreas (MAAP) system was used for the first 6 CL experiments. An integrated multivariable adaptive artificial pancreas (IMAAP) system consisting of MAAP augmented with a hypoglycemia early alarm system was used during the last 3 CL experiments. Glucose values and physical activity information were measured and transferred to the controller every 10 minutes and insulin suggestions were entered to the pump manually. All experiments were designed to be close to real-life conditions. Severe hypoglycemic episodes were seen several times during the OL experiments. With the MAAP system, the occurrence of severe hypoglycemia was decreased significantly (P < .01). No hypoglycemia was seen with the IMAAP system. There was also a significant difference (P < .01) between OL and CL experiments with regard to percentage of glucose concentration (54% vs 58%) that remained within target range (70-180 mg/dl). Integration of an adaptive control and hypoglycemia early alarm system was able to keep glucose concentration values in target range in patients with type 1 diabetes. Postprandial hypoglycemia and exercise-induced hypoglycemia did not occur when this system was used. Physical activity information improved estimation of the blood glucose concentration and effectiveness of the control system.

Hypoglycemia Early Alarm Systems Based On Multivariable Models

Hypoglycemia is a major challenge of artificial pancreas systems and a source of concern for potential users and parents of young children with Type 1 diabetes (T1D). Early alarms to warn the potential of hypoglycemia are essential and should provide enough time to take action to avoid hypoglycemia. Many alarm systems proposed in the literature are based on interpretation of recent trends in glucose values. In the present study, subject-specific recursive linear time series models are introduced as a better alternative to capture glucose variations and predict future blood glucose concentrations. These models are then used in hypoglycemia early alarm systems that notify patients to take action to prevent hypoglycemia before it happens. The models developed and the hypoglycemia alarm system are tested retrospectively using T1D subject data. A Savitzky-Golay filter and a Kalman filter are used to reduce noise in patient data. The hypoglycemia alarm algorithm is developed by using predictions of future glucose concentrations from recursive models. The modeling algorithm enables the dynamic adaptation of models to inter-/intra-subject variation and glycemic disturbances and provides satisfactory glucose concentration prediction with relatively small error. The alarm systems demonstrate good performance in prediction of hypoglycemia and ultimately in prevention of its occurrence.