GLSTM: On Using LSTM for Glucose Level Prediction

Area of Science:

• Endocrinology and Metabolic Diseases
• Biomedical Data Science
• Artificial Intelligence in Healthcare

Background:

• Prediabetes affects one in three individuals, with a 10% annual risk of progressing to type 2 diabetes without intervention.
• Effective glycemic health management is crucial for preventing type 2 diabetes progression.
• Limited noninvasive methods exist for tracking glycemic status, hindering prediabetes self-management.

Purpose of the Study:

• To develop and evaluate a personalized Long Short-Term Memory (LSTM) model for forecasting glucose levels in prediabetes.
• To provide a tool for improved self-management of prediabetes and glycemic health.

Main Methods:

• Utilized personalized prediabetes data including continuous interstitial glucose levels, heart rate, and weekly dietary records.
• Employed a Long Short-Term Memory (LSTM) neural network model for glucose level forecasting.
• Assessed model performance using Root Mean Square Error (RMSE), Mean Squared Error (MSE), Mean Absolute Error (MAE), and R2 coefficient.

Main Results:

• The LSTM model demonstrated efficacy in forecasting glucose levels using personalized prediabetes data.
• Quantitative evaluation metrics (RMSE, MSE, MAE, R2) were used to assess the model's predictive accuracy.

Conclusions:

• The proposed LSTM model shows promise as a noninvasive tool for monitoring glycemic status in prediabetes.
• Personalized glucose forecasting can support self-management strategies for individuals with prediabetes, potentially deterring progression to type 2 diabetes.