Early Prediction of Type 2 Diabetes Using Deep Learning on Continuous Glucose Monitoring (Cgm) Data - A Systematic Review

Abstract

Background: Type 2 diabetes mellitus (T2DM) is a major global health burden characterized by chronic hyperglycemia resulting from insulin resistance and β-cell dysfunction. Despite well-established diagnostic criteria, early detection of individuals at risk remains challenging. Continuous glucose monitoring (CGM) provides real-time, high-frequency glucose data, offering a dynamic view of glucose fluctuations that may precede overt diabetes. Recent advances in deep learning (DL) have created opportunities to analyze these complex time-series data to identify early glycemic irregularities predictive of future diabetes onset.

Objective: This systematic review aims to synthesize and critically evaluate existing evidence on the use of deep learning algorithms applied to CGM data for early prediction of type 2 diabetes. It explores the types of models developed, their predictive performance, validation strategies, and methodological quality.
Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines, an electronic search will be conducted in PubMed, Embase, Scopus, IEEE Xplore, and Web of Science from inception to October 2025. Eligible studies will include original research using deep learning methods (e.g., recurrent neural networks, convolutional neural networks, transformers) to predict T2DM onset or progression using CGM data. Two reviewers will independently perform screening, data extraction, and risk of bias assessment using the PROBAST tool. Extracted data will include study characteristics, model architectures, input representations, and performance metrics such as area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and calibration. Narrative synthesis will be provided, and meta-analysis will be performed if sufficient homogeneity exists.

Results: The review will summarize key findings regarding the performance of deep learning models in predicting early T2DM, highlight methodological limitations, and identify trends in model interpretability and validation.

Conclusion: This systematic review will provide comprehensive insights into the current landscape of deep learning-based CGM analysis for early diabetes prediction. It will identify existing gaps, propose methodological improvements, and outline future directions for translating these models into clinical practice.