Uncertainty-aware Blood Glucose Prediction from Continuous Glucose Monitoring Data

TL;DR

Using the HUPA-UCM diabetes dataset for validation, it is found that Transformer-based models equipped with evidential output heads provide the most effective uncertainty-aware framework, achieving consistently higher predictive accuracies and better-calibrated uncertainty estimates whose magnitudes significantly correlate with prediction errors.

Abstract

In this work, we investigate uncertainty-aware neural network models for blood glucose prediction and adverse glycemic event identification in Type 1 diabetes. We consider three families of sequence models based on LSTM, GRU, and Transformer architectures, with uncertainty quantification enabled by either Monte Carlo dropout or through evidential output layers compatible with Deep Evidential Regression. Using the HUPA-UCM diabetes dataset for validation, we find that Transformer-based models equipped with evidential output heads provide the most effective uncertainty-aware framework, achieving consistently higher predictive accuracies and better-calibrated uncertainty estimates whose magnitudes significantly correlate with prediction errors. We further evaluate the clinical risk of each model using the recently proposed Diabetes Technology Society error grid, with risk categories defined by international expert consensus. Our results demonstrate the value of integrating principled uncertainty quantification into real-time machine-learning-based blood glucose prediction systems.

Figures (9)

• Figure 1: Plot of empirical coverage probabilities (ECP) vs nominal coverage probabilities (NCP) for three types of models trained on heart rate-included inputs and defined by a 30 min prediction horizon. MCE denotes the mean calibration error which is the average of the absolute value of deviation between ECP and NCP.
• Figure 2: The rolling forecast plot for TEM model (a) and Transformer-based model with Monte Carlo dropout (b) trained on basal insulin-included inputs with 1 hour horizon for Patient with ID '3' in the HUPA dataset HUPA, with hypoglycemia (<70 mg/dL) and hyperglycemia (>180 mg/dL) thresholds being represented by red, horizontal dashed lines. Dashed ellipses locate examples of regions where model would fail to detect hypoglycemia events unless uncertainty estimates are used.
• Figure 3: Single long horizon prediction windows for a couple of points near hypoglycemia thresholds in Fig. \ref{['fig:sample_plot_evidential']}, with the vertical lines in each diagram marking the start of the prediction horizon. In both prediction windows, the predicted glucose is deviating away from the threshold even when the actual measured glucose has crossed over, but the $1\sigma$ uncertainty interval encompasses the hypoglycemia region.
• Figure 4: Single long horizon prediction windows for a couple of points near hyperglycemia thresholds in Fig. \ref{['fig:sample_plot_evidential']}. In (a), the predicted glucose did not exceed the threshold even when the actual measured glucose has crossed over, but the $1\sigma$ uncertainty interval encompasses the hyperglycemia region. In (b), the predicted glucose level hovers above the threshold even when the actual measured value has dipped below it, but the size of the $1\sigma$ uncertainty interval appropriately captures the deviations.
• Figure 5: DTS error grids for two patients with marked difference in their risk zone distributions. The various pairs of solid lines defined away from the dashed diagonal mark the boundaries of the five risk zones.