Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Tailoring Adverse Event Prediction in Type 1 Diabetes with Patient-Specific Deep Learning Models

Giorgia Rigamonti, Mirko Paolo Barbato, Davide Marelli, Paolo Napoletano

TL;DR

This work addresses the challenge of predicting blood glucose in Type 1 Diabetes by moving beyond CGM-only models to patient-specific, multimodal deep learning. It introduces a two-stage Bi-GRU framework: a population-level model trained with Leave-One-Subject-Out Cross-Validation, followed by subject-specific fine-tuning on limited data, using inputs that include CGM, insulin dosing, and carbohydrate intake. The approach outperforms CGM-only baselines and demonstrates robust personalization with limited data, aided by a shrinkage loss to emphasize rare adverse events. The findings highlight the practical potential for adaptive, personalized glucose forecasting in wearable and mobile health platforms to improve timely interventions and diabetes management.

Abstract

Effective management of Type 1 Diabetes requires continuous glucose monitoring and precise insulin adjustments to prevent hyperglycemia and hypoglycemia. With the growing adoption of wearable glucose monitors and mobile health applications, accurate blood glucose prediction is essential for enhancing automated insulin delivery and decision-support systems. This paper presents a deep learning-based approach for personalized blood glucose prediction, leveraging patient-specific data to improve prediction accuracy and responsiveness in real-world scenarios. Unlike traditional generalized models, our method accounts for individual variability, enabling more effective subject-specific predictions. We compare Leave-One-Subject-Out Cross-Validation with a fine-tuning strategy to evaluate their ability to model patient-specific dynamics. Results show that personalized models significantly improve the prediction of adverse events, enabling more precise and timely interventions in real-world scenarios. To assess the impact of patient-specific data, we conduct experiments comparing a multimodal, patient-specific approach against traditional CGM-only methods. Additionally, we perform an ablation study to investigate model performance with progressively smaller training sets, identifying the minimum data required for effective personalization-an essential consideration for real-world applications where extensive data collection is often challenging. Our findings underscore the potential of adaptive, personalized glucose prediction models for advancing next-generation diabetes management, particularly in wearable and mobile health platforms, enhancing consumer-oriented diabetes care solutions.

Tailoring Adverse Event Prediction in Type 1 Diabetes with Patient-Specific Deep Learning Models

TL;DR

This work addresses the challenge of predicting blood glucose in Type 1 Diabetes by moving beyond CGM-only models to patient-specific, multimodal deep learning. It introduces a two-stage Bi-GRU framework: a population-level model trained with Leave-One-Subject-Out Cross-Validation, followed by subject-specific fine-tuning on limited data, using inputs that include CGM, insulin dosing, and carbohydrate intake. The approach outperforms CGM-only baselines and demonstrates robust personalization with limited data, aided by a shrinkage loss to emphasize rare adverse events. The findings highlight the practical potential for adaptive, personalized glucose forecasting in wearable and mobile health platforms to improve timely interventions and diabetes management.

Abstract

Effective management of Type 1 Diabetes requires continuous glucose monitoring and precise insulin adjustments to prevent hyperglycemia and hypoglycemia. With the growing adoption of wearable glucose monitors and mobile health applications, accurate blood glucose prediction is essential for enhancing automated insulin delivery and decision-support systems. This paper presents a deep learning-based approach for personalized blood glucose prediction, leveraging patient-specific data to improve prediction accuracy and responsiveness in real-world scenarios. Unlike traditional generalized models, our method accounts for individual variability, enabling more effective subject-specific predictions. We compare Leave-One-Subject-Out Cross-Validation with a fine-tuning strategy to evaluate their ability to model patient-specific dynamics. Results show that personalized models significantly improve the prediction of adverse events, enabling more precise and timely interventions in real-world scenarios. To assess the impact of patient-specific data, we conduct experiments comparing a multimodal, patient-specific approach against traditional CGM-only methods. Additionally, we perform an ablation study to investigate model performance with progressively smaller training sets, identifying the minimum data required for effective personalization-an essential consideration for real-world applications where extensive data collection is often challenging. Our findings underscore the potential of adaptive, personalized glucose prediction models for advancing next-generation diabetes management, particularly in wearable and mobile health platforms, enhancing consumer-oriented diabetes care solutions.
Paper Structure (18 sections, 6 equations, 3 figures, 2 tables)

This paper contains 18 sections, 6 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Method
  4. Patient Identification via Multimodal Data
  5. Patient-Independent vs Patient-Specific Models
  6. Experiments
  7. Materials
  8. Data Balancing and Splitting
  9. Preprocessing
  10. Experimental Configuration
  11. Comparative Analysis and Ablation Study
  12. Training Configuration
  13. Analytical performance assessment criteria
  14. Results and Discussions
  15. Patient Identification Results
  16. ...and 3 more sections

Figures (3)

  • Figure 1: Bi-GRU architecture combining patient-independent learning with patient-specific fine-tuning to produce personalized BGC predictions.
  • Figure 2: Confusion matrices for the patient identification task using (a) the OhioT1DM and (b) the DiaTrend datasets. Darker shades indicate higher classification accuracy.
  • Figure 3: Ablation study on patient-specific models with decreasing training data. (a--d) 30-min prediction horizon; (e--h) 60-min prediction horizon. Light orange and light blue dashed lines represent patient-independent models trained on OhioT1DM and DiaTrend, respectively. Solid orange and blue lines correspond to patient-specific models trained on the same datasets.