Mealtime prediction using wearable insulin pump data to support diabetes management

Area of Science:

• Biomedical Engineering
• Endocrinology
• Data Science

Background:

• Post-meal hyperglycemia is a common challenge for individuals with diabetes, often resulting from delayed or missed insulin administrations.
• Inconsistent meal timing significantly complicates blood glucose management in type 1 diabetes.

Purpose of the Study:

• To analyze mealtime patterns in type 1 diabetes patients using data from wearable insulin pumps.
• To create personalized predictive models for mealtimes to facilitate timely insulin dosing.

Main Methods:

• Utilized two independent datasets comprising over 45,000 meal logs from 82 type 1 diabetes patients.
• Developed and evaluated personalized Long Short-Term Memory (LSTM)-based models for mealtime prediction.

Main Results:

• Identified that over 60% of patients exhibit irregular and inconsistent mealtime patterns.
• Demonstrated the feasibility of predicting future mealtimes with personalized LSTM models, achieving over 95% F1 score and fewer than 0.25 false positives daily.
• Observed significant daily and monthly variations in individual meal timing.

Conclusions:

• Personalized predictive models show high accuracy in forecasting mealtimes for individuals with type 1 diabetes.
• This research provides a foundation for a meal prediction system to aid in timely insulin bolus delivery.
• Reducing post-meal glucose spikes through proactive insulin administration is a key potential benefit.