Short-term prediction of future continuous glucose monitoring readings in type 1 diabetes: Development and validation of a neural network regression model

Explainable Machine-Learning Models to Predict Weekly Risk of Hyperglycemia, Hypoglycemia, and Glycemic Variability in Patients With Type 1 Diabetes Based on Continuous Glucose Monitoring

BACKGROUND AND OBJECTIVE

The aim of this study was to develop and validate explainable prediction models based on continuous glucose monitoring (CGM) and baseline data to identify a week-to-week risk of CGM key metrics (hyperglycemia, hypoglycemia, glycemic variability). By having a weekly prediction of CGM key metrics, it is possible for the patient or health care personnel to take immediate preemptive action.

METHODS

We analyzed, trained, and internally tested three prediction models (Logistic regression, XGBoost, and TabNet) using CGM data from 187 type 1 diabetes patients with long-term CGM monitoring. A binary classification approach combined with feature engineering deployed on the CGM signals was used to predict hyperglycemia, hypoglycemia, and glycemic variability based on consensus targets (time above range ≥5%, time below range ≥4%, coefficient of variation ≥36%). The models were validated in two independent cohorts with a total of 223 additional patients of varying ages.

RESULTS

A total of 46 593 weeks of CGM data were included in the analysis. For the best model (XGBoost), the area under the receiver operating characteristic curve (ROC-AUC) was 0.9 [95% confidence interval (CI) = 0.89-0.91], 0.89 [95% CI = 0.88-0.9], and 0.8 [95% CI = 0.79-0.81] for predicting hyperglycemia, hypoglycemia, and glycemic variability in the interval validation, respectively. The validation test showed good generalizability of the models with ROC-AUC of 0.88 to 0.95, 0.84 to 0.89, and 0.80 to 0.82 for predicting the glycemic outcomes.

CONCLUSION

Prediction models based on real-world CGM data can be used to predict the risk of unstable glycemic control in the forthcoming week. The models showed good performance in both internal and external validation cohorts.

Forecasting glucose values for patients with type 1 diabetes using heart rate data

BACKGROUND

Type 1 Diabetes Mellitus (T1DM) is a chronic metabolic disease affecting millions of people worldwide. T1DM requires patients to continuously monitor their blood glucose levels. Due to pancreatic dysfunctions, patients use insulin injections to correct glucose values by synthetic insulin. Continuous Glucose Monitoring (CGM) is a system which includes an algorithm allowing to measure (and in some cases to predict) glucose levels at a frequent sampling time. This enable implementing advanced devices, including automated insulin pump delivery. Nevertheless, CGM still presents some limitations, including (i) the delay (time lag) in detecting change in glucose levels compared to the traditional blood glucose measurement, and (ii) the lack of a sufficient and acceptable time to accurately predict glucose values.

METHODS

We propose a framework based on a Gated Recurrent Unit (GRU) model to forecast both short- and long-term glucose values using heart rate (HR) and interstitial glucose (IG) values. The framework acquires HR and IG data and predicts glucose values with higher precision compared to state-of-the-art models. For training and testing the proposed framework, we used the OhioT1DM Dataset, which includes physiological data such as HR and IG values collected over an 8-week observation period. Additionally, we validated our framework using two other glucose datasets to ensure its generalizability across different HR and IG sampling frequencies. The proposed framework can be used to optimize the CGM system by incorporating patient HR measurements, thereby improving the prediction of short- and long-term glucose levels and reducing risks associated with conditions like hypoglycemia.

RESULTS

Experimental tests were conducted using HR and IG data from the OhioT1DM Dataset, as well as from two additional T1DM patient datasets. We analyzed 6 patients from Ohio dataset while we validated the algorithm on 23 patients coming from two different university hospitals (6 from the University of Catanzaro medical hospital and 17 gathered from a validated study at IRCCS San Matteo Hospital in Pavia) for a total number of 29 patients. Our framework demonstrates an improvement in forecasting IG values in terms of RMSE and MAE for different choice of prediction horizons (PH). In the case of a PH of 5, 10, 20, 30, and 60 min, we reach an RMSE of 5.0, 9.38, 15.27, 20.48, and 34.16 respectively. The framework is freely available as an open-source, with an example dataset on a GitHub repository (see https://github.com/rafgia/attention_to_glycemia).

CONCLUSION

Our framework offers a promising solution for improving glucose level prediction and management in T1DM patients. By leveraging a GRU model and incorporating HR and IG values, we achieve more precise glucose level forecasting compared to state-of-the-art models. This approach not only enhances the accuracy of glucose predictions but also mitigates the risks associated with hypoglycemia.

Development and Validation of a Machine Learning Model to Predict Weekly Risk of Hypoglycemia in Patients with Type 1 Diabetes Based on Continuous Glucose Monitoring

Aim: The aim of this study was to develop and validate a prediction model based on continuous glucose monitoring (CGM) data to identify a week-to-week risk profile of excessive hypoglycemia. Methods: We analyzed, trained, and internally tested two prediction models using CGM data from 205 type 1 diabetes patients with long-term CGM monitoring. A binary classification approach (XGBoost) combined with feature engineering deployed on the CGM signals was utilized to predict excessive hypoglycemia risk defined by two targets (time below range [TBR] >4% and the upper TBR 90th percentile limit) of TBR the following week. The models were validated in two independent cohorts with a total of 253 additional patients. Results: A total of 61,470 weeks of CGM data were included in the analysis. The XGBoost models had an area under the receiver operating characteristic curve (ROC-AUC) of 0.83-0.87 (95% confidence interval; 0.83-0.88) in the test dataset. The external validation showed ROC-AUCs of 0.81-0.90. The most discriminative features included the low blood glucose index, the glycemic risk assessment diabetes equation (GRADE), hypoglycemia, the TBR, waveform length, the coefficient of variation and mean glucose during the previous week. This highlights that the pattern of hypoglycemia combined with glucose variability during the past week contains information on the risk of future hypoglycemia. Conclusion: Prediction models based on real-world CGM data can be used to predict the risk of hypoglycemia in the forthcoming week. The models showed good performance in both the internal and external validation cohorts.

Development of Machine Learning Models for the Identification of Elevated Ketone Bodies During Hyperglycemia in Patients with Type 1 Diabetes

Aims: Diabetic ketoacidosis (DKA) is a serious life-threatening condition caused by a lack of insulin, which leads to elevated plasma glucose and metabolic acidosis. Early identification of developing DKA is important to start treatment and minimize complications and risk of death. The aim of the present study is to develop and test prediction model(s) that gives an alarm about their risk of developing elevated ketone bodies during hyperglycemia. Methods: We analyzed data from 138 type 1 diabetes patients with measurements of ketone bodies and continuous glucose monitoring (CGM) data from over 30,000 days of wear time. We utilized a supervised binary classification machine learning approach to identify elevated levels of ketone bodies (≥0.6 mmol/L). Data material was randomly divided at patient level in 70%/30% (training/test) dataset. Logistic regression (LR) and random forest (RF) classifier were compared. Results: Among included patients, 913 ketone samples were eligible for modeling, including 273 event samples with ketone levels ≥0.6 mmol/L. An area under the receiver operating characteristic curve from the RF classifier was 0.836 (confidence interval [CI] 90%, 0.783-0.886) and 0.710 (CI 90%, 0.646-0.77) for the LR classifier. Conclusions: The novel approach for identifying elevated ketone levels in patients with type 1 diabetes utilized in this study indicates that CGM could be a valuable resource for the early prediction of patients at risk of developing DKA. Future studies are needed to validate the results.

Identification of individuals with diabetes who are eligible for continuous glucose monitoring forecasting

BACKGROUND AND OBJECTIVES

Predicting glucose levels in individuals with diabetes offers potential improvements in glucose control. However, not all patients exhibit predictable glucose dynamics, which may lead to ineffective treatment strategies. We sought to investigate the efficacy of a 7-day blinded screening test in identifying diabetes patients suitable for glucose forecasting.

METHODS

Participants with type 1 diabetes (T1D) were stratified into high and low initial error groups based on screening results (eligible and non-eligible). Long-term glucose predictions (30/60 min lead time) were evaluated among 334 individuals who underwent continuous glucose monitoring (CGM) over a total of 64,460,560 min.

RESULTS

A strong correlation was observed between screening accuracy and long-term mean absolute relative difference (MARD) (0.661-0.736; p < 0.001), suggesting significant predictability between screening and long-term errors. Group analysis revealed a notable reduction in predictions falling within zone D of the Clark Error Grid by a factor of three and in zone C by a factor of two.

CONCLUSIONS

The identification of eligible patients for glucose prediction through screening represents a practical and effective strategy. Implementation of this approach could lead to a decrease in adverse glucose predictions.

A personalized multitasking framework for real-time prediction of blood glucose levels in type 1 diabetes patients

Real-time prediction of blood glucose levels (BGLs) in individuals with type 1 diabetes (T1D) presents considerable challenges. Accordingly, we present a personalized multitasking framework aimed to forecast blood glucose levels in patients. The patient data was initially categorized according to gender and age and subsequently utilized as input for a modified GRU network model, creating five prediction sub-models. The model hyperparameters were optimized and tuned after introducing the decay factor and incorporating the TCN network and attention mechanism into the GRU model. This step was undertaken to improve the capability of feature extraction. The Ohio T1DM clinical dataset was used to train and evaluate the performance of the proposed model. The metrics, including Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and Clark Error Grid Analysis (EGA), were used to evaluate the performance. The results showed that the average RMSE and the MAE of the proposed model were 16.896 and 9.978 mg/dL, respectively, over the prediction horizon (PH) of 30 minutes. The average RMSE and the MAE were 28.881 and 19.347 mg/dL, respectively, over the PH of 60 min. The proposed model demonstrated excellent prediction accuracy. In addition, the EGA analysis showed that the proposed model accurately predicted 30-minute and 60-minute PH within zones A and B, demonstrating that the framework is clinically feasible. The proposed personalized multitask prediction model in this study offers robust assistance for clinical decision-making, playing a pivotal role in improving the outcomes of individuals with diabetes.

Glucose trend prediction model based on improved wavelet transform and gated recurrent unit

Glucose trend prediction based on continuous glucose monitoring (CGM) data is a crucial step in the implementation of an artificial pancreas (AP). A glucose trend prediction model with high accuracy in real-time can greatly improve the glycemic control effect of the artificial pancreas and effectively prevent the occurrence of hyperglycemia and hypoglycemia. In this paper, we propose an improved wavelet transform threshold denoising algorithm for the non-linearity and non-smoothness of the original CGM data. By quantitatively comparing the mean square error (MSE) and signal-to-noise ratio (SNR) before and after the improvement, we prove that the improved wavelet transform threshold denoising algorithm can reduce the degree of distortion after the smoothing of CGM data and improve the extraction effect of CGM data features at the same time. Based on this finding, we propose a glucose trend prediction model (IWT-GRU) based on the improved wavelet transform threshold denoising algorithm and gated recurrent unit. We compared the root mean square error (RMSE), mean absolute percentage error (MAPE), and coefficient of determination ($ {\mathrm{R}}^{2} $) of Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), Support vector regression (SVR), Gated Recurrent Unit (GRU) and IWT-GRU on the original CGM monitoring data of 80 patients for 7 consecutive days with different prediction horizon (PH). The results showed that the IWT-GRU model outperformed the other four models. At PH = 45 min, the RMSE was 0.5537 mmol/L, MAPE was 2.2147%, $ {\mathrm{R}}^{2} $ was 0.989 and the average runtime was only 37.2 seconds. Finally, we analyze the limitations of this study and provide an outlook on the future direction of blood glucose trend prediction.

Synergy between imputed genetic pathway and clinical information for predicting recurrence in early stage non-small cell lung cancer

OBJECTIVE

Lung cancer exhibits unpredictable recurrence in low-stage tumors and variable responses to different therapeutic interventions. Predicting relapse in early-stage lung cancer can facilitate precision medicine and improve patient survivability. While existing machine learning models rely on clinical data, incorporating genomic information could enhance their efficiency. This study aims to impute and integrate specific types of genomic data with clinical data to improve the accuracy of machine learning models for predicting relapse in early-stage, non-small cell lung cancer patients.

METHODS

The study utilized a publicly available TCGA lung cancer cohort and imputed genetic pathway scores into the Spanish Lung Cancer Group (SLCG) data, specifically in 1348 early-stage patients. Initially, tumor recurrence was predicted without imputed pathway scores. Subsequently, the SLCG data were augmented with pathway scores imputed from TCGA. The integrative approach aimed to enhance relapse risk prediction performance.

RESULTS

The integrative approach achieved improved relapse risk prediction with the following evaluation metrics: an area under the precision-recall curve (PR-AUC) score of 0.75, an area under the ROC (ROC-AUC) score of 0.80, an F1 score of 0.61, and a Precision of 0.80. The prediction explanation model SHAP (SHapley Additive exPlanations) was employed to explain the machine learning model's predictions.

CONCLUSION

We conclude that our explainable predictive model is a promising tool for oncologists that addresses an unmet clinical need of post-treatment patient stratification based on the relapse risk while also improving the predictive power by incorporating proxy genomic data not available for specific patients.

Recent applications of machine learning and deep learning models in the prediction, diagnosis, and management of diabetes: a comprehensive review

Diabetes as a metabolic illness can be characterized by increased amounts of blood glucose. This abnormal increase can lead to critical detriment to the other organs such as the kidneys, eyes, heart, nerves, and blood vessels. Therefore, its prediction, prognosis, and management are essential to prevent harmful effects and also recommend more useful treatments. For these goals, machine learning algorithms have found considerable attention and have been developed successfully. This review surveys the recently proposed machine learning (ML) and deep learning (DL) models for the objectives mentioned earlier. The reported results disclose that the ML and DL algorithms are promising approaches for controlling blood glucose and diabetes. However, they should be improved and employed in large datasets to affirm their applicability.

Hypoglycemia event prediction from CGM using ensemble learning

This work sought to explore the potential of using standalone continuous glucose monitor (CGM) data for the prediction of hypoglycemia utilizing a large cohort of type 1 diabetes patients during free-living. We trained and tested an algorithm for the prediction of hypoglycemia within 40 minutes on 3.7 million CGM measurements from 225 patients using ensemble learning. The algorithm was also validated using 11.5 million synthetic CGM data. The results yielded a receiver operating characteristic area under the curve (ROC AUC) of 0.988 and a precision-recall area under the curve (PR AUC) of 0.767. In an event-based analysis for predicting hypoglycemic events, the algorithm had a sensitivity of 90%, a lead-time of 17.5 minutes and a false-positive rate of 38%. In conclusion, this work demonstrates the potential of using ensemble learning to predict hypoglycemia, using only CGM data. This could help alarm patients of a future hypoglycemic event so countermeasures can be initiated.

Intelligent Control with Artificial Neural Networks for Automated Insulin Delivery Systems

Type 1 diabetes mellitus is a disease that affects millions of people around the world. Recent progress in embedded devices has allowed the development of artificial pancreas that can pump insulin subcutaneously to automatically regulate blood glucose levels in diabetic patients. In this work, a Lyapunov-based intelligent controller using artificial neural networks is proposed for application in automated insulin delivery systems. The adoption of an adaptive radial basis function network within the control scheme allows regulation of blood glucose levels without the need for a dynamic model of the system. The proposed model-free approach does not require the patient to inform when they are going to have a meal and is able to deal with inter- and intrapatient variability. To ensure safe operating conditions, the stability of the control law is rigorously addressed through a Lyapunov-like analysis. In silico analysis using virtual patients are provided to demonstrate the effectiveness of the proposed control scheme, showing its ability to maintain normoglycemia in patients with type 1 diabetes mellitus. Three different scenarios were considered: one long- and two short-term simulation studies. In the short-term analyses, 20 virtual patients were simulated for a period of 7 days, with and without prior basal therapy, while in the long-term simulation, 1 virtual patient was assessed over 63 days. The results show that the proposed approach was able to guarantee a time in the range above 95% for the target glycemia in all scenarios studied, which is in fact well above the desirable 70%. Even in the long-term analysis, the intelligent control scheme was able to keep blood glucose metrics within clinical care standards: mean blood glucose of 119.59 mg/dL with standard deviation of 32.02 mg/dL and coefficient of variation of 26.78%, all below the respective reference values.

End-to-end design of wearable sensors

Wearable devices provide an alternative pathway to clinical diagnostics by exploiting various physical, chemical and biological sensors to mine physiological (biophysical and/or biochemical) information in real time (preferably, continuously) and in a non-invasive or minimally invasive manner. These sensors can be worn in the form of glasses, jewellery, face masks, wristwatches, fitness bands, tattoo-like devices, bandages or other patches, and textiles. Wearables such as smartwatches have already proved their capability for the early detection and monitoring of the progression and treatment of various diseases, such as COVID-19 and Parkinson disease, through biophysical signals. Next-generation wearable sensors that enable the multimodal and/or multiplexed measurement of physical parameters and biochemical markers in real time and continuously could be a transformative technology for diagnostics, allowing for high-resolution and time-resolved historical recording of the health status of an individual. In this Review, we examine the building blocks of such wearable sensors, including the substrate materials, sensing mechanisms, power modules and decision-making units, by reflecting on the recent developments in the materials, engineering and data science of these components. Finally, we synthesize current trends in the field to provide predictions for the future trajectory of wearable sensors.

Optimization and Evaluation of an Intelligent Short-Term Blood Glucose Prediction Model Based on Noninvasive Monitoring and Deep Learning Techniques

Continuous noninvasive blood glucose monitoring and estimation management by using photoplethysmography (PPG) technology always have a series of problems, such as substantial time variability, inaccuracy, and complex nonlinearity. This paper proposes a blood glucose (BG) prediction model for more precise prediction based on BG series decomposition by complete aggregation empirical mode decomposition based on adaptive white noise (CEEMDAN) and the gated recurrent unit (GRU) that is optimized by improved bacterial foraging optimization (IBFO). Hierarchical clustering technology recombines the decomposed BG series according to their sample entropy and the correlations with the original BG trends. Dynamic BG trends are regressed separately for each recombined BG series by the GRU model to realize the more precise estimations, which are optimized by IBFO for its structure and superparameters. Through experiments, the optimized and basic LSTM, RNN, and support vector regression (SVR) are compared to evaluate the performance of the proposed model. The experimental results indicate that the root mean square error (RMSE) and mean absolute percentage error (MAPE) of the 15-min IBFO-GRU prediction is improved on average by about 13.1% and 18.4%, respectively, compared with those of the RNN and LSTM optimized by IBFO. Meanwhile, the proposed model improved the Clarke error grid results by about 2.6% and 5.0% compared with those of the IBFO-LSTM and IBFO-RNN in 30-min prediction and by 4.1% and 6.6% in 15-min ahead forecast, respectively. The evaluation outcomes of our proposed CEEMDAN-IBFO-GRU model have high accuracy and adaptability and can effectively provide early intervention control of the occurrence of hyperglycemic complications.

Penalty weighted glucose prediction models could lead to better clinically usage

BACKGROUND AND OBJECTIVE

Numerous attempts to predict glucose value from continuous glucose monitors (CGM) have been published. However, there is a lack of proper analysis and modeling of penalty for errors in different glycemic ranges. The aim of this study was to investigate the potential for developing glucose prediction models with focus on the clinical aspects.

METHODS

We developed and compared six different models to test which approach were best suited for predicting glucose levels at different lead times between 10 and 60 min. The models were: last observation carried forward, linear extrapolation, ensemble methods using LSBoost and bagging, neural networks, one without error-weights and one with error-weights. The modeling and test were based on 225 type 1 diabetes patients with 315,000 h of CGM data.

RESULTS

Results show that it is possible to predict glucose levels based on CGM with a reasonable accuracy and precision with a 30-min prediction lead time. A comparison of different methods shows that there are improvements on performance gained from using more advanced machine learning algorithms (MARD 10.26-10.79 @ 30-min lead time) compared to a simple modeling (MARD 10.75-12.97 @ 30-min lead time). Moreover, the proposed use of error weights could lead to better clinical performance of these models, which is an important factor for real usage. E.g., the percentages in the C-zone of the consensus error grid without error-weights (0.57-0.68%) vs including error-weights (0.28%).

CONCLUSIONS

The results point toward that using error weighting in the training of the models could lead to better clinical performance.