NCT03898076

Predictive A1c Based on CGM Data Using CGM Data

The Prediction of A1c Based on CGM Data Through Applying Machine Learning Approaches

• Status: Completed
• Phase: —
• Study type: Observational
• Enrollment: 60 (actual)
• Sponsor: Sidra Medicine · Academic / Other
• Sex: All
• Age: 2 Years – 18 Years
• Healthy volunteers: Not accepted

Summary

Introduction. The hemoglobin A1C (HbA1c) reflects the average blood glucose level for last two to three months. Recent advancements in the sensor technology facilitate the daily monitoring of the blood glucose using CGM devices. The future prediction of the HbA1C based on the CGM data holds a critical significance in maintaining long term health of diabetes patients. A higher than normal value of the HbA1c greatly increases the likelihood of diabetes related cardiovascular disease. Goal. The aim this study is to predict the HbA1c in advance by utilizing the CGM data through applying machine learning techniques. The outcomes of this research will assist in improving the health of diabetic patients. Methods. This is a retrospective analysis. The investigators will de-identify and analyze 120 patients with T1D who using CGM sensor for last three months. Past 15 days of CGM data will be analyzed and different glucose variability features such as time in range (TIR), coefficient of variation (CV), mean amplitude of glycemic excursion (MAGE), mean of daily differences (MODD), continuous overall net glycemic action (CONGA) will be extracted. A machine learning model will calculate (predict) HbA1c in 2-3 months advance based on these 15 days of CGM data. To evaluate the performance of the proposed prediction model, predicted HbA1c will be compared with the real HbA1c.

Detailed description

This is a retrospective analysis. The investigators will de-identify and analyze 120 patients with T1D using Continuous Glucose Monitoring (CGM) system for last three months. Past 15 days of CGM data will be analyzed and different glucose variability features such as time in range (TIR), coefficient of variation (CV), mean amplitude of glycemic excursion (MAGE), mean of daily differences (MODD), continuous overall net glycemic action (CONGA) will be extracted. A machine learning model will be developed to predict HbA1c in 2-3 months advance based on these 15 days of CGM data. The model is using linear regression, penalized regression (Ridge regression, Lasso regression and Elastic net regression) in combination gradient boosting to calculate predictive A1c

Conditions

• Diabetes Mellitus, Type 1

Interventions

Type	Name	Description
DEVICE	Flash Glucose Monitoring	Continuous Glucose Monitoring (CGM) values will be downloaded from CGM device for a period of 90 days.
OTHER	A1c	A1c levels will be collected from Hospital EMR prior to CGM data downoad
OTHER	Predictive A1c	Predictive A1c will be calculated based on the first 15 days of CGM data using time in range (TIR), coefficient of variation (CV), mean amplitude of glycemic excursion (MAGE), mean of daily differences (MODD), continuous overall net glycemic action (CONGA). Predictive A1c will be correlated with actual A1c.

Timeline

• Start date: 2020-06-01
• Primary completion: 2020-08-31
• Completion: 2020-12-30
• First posted: 2019-04-01
• Last updated: 2021-09-28

Locations

1 site across 1 country: Qatar