"It's like a glimpse into the future": Exploring the Role of Blood Glucose Prediction Technologies for Type 1 Diabetes Self-Management

Literature Title

“It’s like a glimpse into the future”: Exploring the Role of Blood Glucose Prediction Technologies for Type 1 Diabetes Self-Management

Literature Information

• Subject Area: Health information technology and human-computer interaction, with a focus on self-management support for Type 1 Diabetes (T1D).
• Keywords: Health, Type 1 Diabetes, Artificial Intelligence, Qualitative Research, Mobile Health

Research Background and Problem

• Problem or Challenge:

  • Self-management of Type 1 Diabetes (T1D) requires complex decision-making, such as predicting blood glucose fluctuations, which imposes significant stress on many patients.
  • Although Artificial Intelligence (AI)-supported Blood Glucose Prediction (BGP) technologies hold promise, their integration into daily decision-making is challenging due to prediction uncertainty and interpretability issues.

• Significance:

  • High-quality diabetes management can reduce the risks of acute and long-term health complications, such as hypoglycemia-induced fainting and chronic cardiovascular or kidney diseases.
  • Providing effective tools to alleviate the management burden is crucial for improving the quality of life of T1D patients.

• Research Motivation and Related Work:

  • Existing T1D technologies often rely on users' independent interpretation of data but lack support for future blood glucose prediction.
  • While there is extensive research on technologies for Type 2 Diabetes (T2D), there is limited exploration of AI applications in T1D, particularly regarding how patients interact with BGP and its impact on self-management.

Proposed Solution

• Methods and Proposed Solution:

  • Development of the MOON-T1D application, a mobile app providing simulated blood glucose predictions (using simulated rather than real user data).
  • Investigated the experiences of 15 T1D patients using MOON-T1D through the Experience Sampling Method (ESM) and semi-structured interviews.

• Innovative Aspects:

  • Unlike existing BGP studies, which rarely integrate user experience, this research focuses on user needs, emotional responses, and changes in daily management practices.
  • Designed a research method grounded in real-world usage contexts, proposing design guidelines related to user experience.

• Implementation Steps and Key Technologies:

  • Developed an app interface allowing users to log food intake, insulin injections, and physical activity.
  • Used a deep learning model (based on the existing OhioT1DM dataset) to generate simulated blood glucose predictions and enabled users to interact with and record their evaluations.
  • Conducted Reflexive Thematic Analysis to code and analyze data, extracting key themes for design optimization.

Research Findings

• Specific Findings:

  • BGP was found to provide significant support in specific contexts, such as physical activities, unplanned high activity levels, and sleep.
  • Four key themes were identified:
    1. BGP alleviates constraints in certain management activities.
    2. Balancing trust and reliance on predictions.
    3. Emotional impacts (positive, such as confidence, and negative, such as stress) of BGP on management behaviors.
    4. Individual differences in interaction and engagement influence system responses.

• Advantages Compared to Existing Solutions:

  • Focuses on user experience and practical needs, in contrast to studies solely emphasizing algorithmic prediction accuracy, offering a more human-centered approach.
  • Considers emotional and psychological impacts of predictions, such as how users cope with uncertainty.

• Experimental or Evaluation Results:

  • Overall experience with MOON-T1D was positive, with 98.3% of ESM responses indicating that BGP was helpful.
  • Proposed new design suggestions, such as improving safety and presenting prediction uncertainty.

• Limitations and Future Directions:

  • Limitations:
    • Simulated data was used, which may differ from real-world application scenarios.
    • The study sample was concentrated on Western European patients, and cultural and healthcare system heterogeneity was not fully addressed.
  • Future Directions:
    • Evaluate the impact of BGP on health metrics over longer time spans and in real-world data contexts.
    • Investigate user acceptance and utility of various prediction visualization methods.
    • Explore innovative ways to integrate BGP into existing diabetes management support systems.