Glucose Prediction Using Population-based Models and Genetic Data in Type 1 Diabetes Patients

Area of Science:

• Biomedical Engineering
• Computational Biology
• Endocrinology

Background:

• Type 1 Diabetes (T1D) management involves complex glucose regulation, with dysglycaemia posing long-term risks.
• Technological advancements like continuous glucose sensors and closed-loop systems offer improved T1D management strategies.
• Hypoglycaemia presents an acute, life-threatening challenge impacting T1D patients' quality of life.

Purpose of the Study:

• To investigate the integration of genetic variants into predictive models for subcutaneous glucose concentration in T1D.
• To enhance short-term glucose prediction accuracy, focusing on improving performance in hypoglycaemic ranges.
• To develop and evaluate cluster-wise predictive models based on genetic data for T1D glucose forecasting.

Main Methods:

• Utilized K-mode clustering to group T1D participants based on genetic data.
• Developed separate univariate Long Short-Term Memory (LSTM) models for each genetic cluster.
• Tested cluster-wise population models on a dataset of 47 T1D patients with a 30-minute prediction horizon.

Main Results:

• Cluster-wise models achieved a Mean Absolute Percentage Error (MAPE) of 11.6%, 12.8%, and 12.0% across three identified genetic clusters.
• Predictions from cluster-wise models demonstrated superior performance compared to global models, especially for a cluster with frequent extreme glucose values (Group 0).
• The findings indicate that incorporating genetic data into T1D glucose predictive modeling pipelines can significantly reduce prediction errors.

Conclusions:

• Genetic data integration into predictive modeling pipelines enhances glucose prediction accuracy in T1D.
• Cluster-wise modeling based on genetic profiles offers a promising approach for personalized T1D glucose forecasting.
• This strategy holds potential for improving the safety and efficacy of T1D management technologies.