GluMarker: A Novel Predictive Modeling of Glycemic Control Through Digital Biomarkers

TL;DR

GluMarker addresses the need for comprehensive digital biomarkers in diabetes by integrating broader data sources to predict next-day glycemic control. It introduces an end-to-end pipeline with interval-based digital biomarker generation and a parallel-branch architecture that fuses continuous and discrete features via cross-attention, trained on two-day data. The approach achieves state-of-the-art $AUC$ performance on Anderson's dataset and identifies key digital biomarkers driving predictions, offering actionable insights for daily diabetes management. This work has potential clinical impact by informing personalized care and highlighting dynamic factors beyond insulin dosing that influence glycemic control.

Abstract

The escalating prevalence of diabetes globally underscores the need for diabetes management. Recent research highlights the growing focus on digital biomarkers in diabetes management, with innovations in computational frameworks and noninvasive monitoring techniques using personalized glucose metrics. However, they predominantly focus on insulin dosing and specific glucose values, or with limited attention given to overall glycemic control. This leaves a gap in expanding the scope of digital biomarkers for overall glycemic control in diabetes management. To address such a research gap, we propose GluMarker -- an end-to-end framework for modeling digital biomarkers using broader factors sources to predict glycemic control. Through the assessment and refinement of various machine learning baselines, GluMarker achieves state-of-the-art on Anderson's dataset in predicting next-day glycemic control. Moreover, our research identifies key digital biomarkers for the next day's glycemic control prediction. These identified biomarkers are instrumental in illuminating the daily factors that influence glycemic management, offering vital insights for diabetes care.

Figures (6)

• Figure 1: Architecture of GluMarker. The original data is divided into intervals to create digital biomarkers, which are then input into the model for glycemic control prediction.
• Figure 2: Visualization of data distribution. Based on the data distribution, multiple margins are generated to divide the data into several intervals, with each interval representing a digital biomarker.
• Figure 3: Visualization of three glycemic control situations. Following Equation \ref{['controlrange']}, we generate three glycemic control ranges for the prediction task.
• Figure 4: Visualization of ROC curves generated by different models. Our model produces the largest area under the ROC curve (AUC score), indicating the best classification performance and highest sensitivity, and the lowest false positive rate.
• Figure 5: Visualization of feature importance for glycemic control prediction. The impact of all features for each category ("Good", "Moderate", and "Poor") is demonstrated. For instance, the prior-day correction bolus and TAR play the most significant roles in good and poor glycemic control, respectively.