A Wavelet AI Algorithm to Automatically Identify Postprandial Glucose Responses From Continuous Glucose Monitoring...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Metabolic Health Monitoring

Background:

• Automated identification of postprandial glucose responses (PPGR) from continuous glucose monitoring (CGM) data is crucial for detecting early dysglycemia in non-diabetic individuals.
• A standardized method for automated PPGR identification from CGM data is currently lacking.

Purpose of the Study:

• To develop and evaluate a wavelet transform-based artificial intelligence (AI) algorithm for automated PPGR identification using only CGM data.
• To assess the algorithm's performance in predicting PPGR characteristics and its association with clinical markers of glucose metabolism.

Main Methods:

• A wavelet transform-based AI algorithm was developed to identify PPGRs from CGM data.
• The algorithm was validated on a public CGM dataset and three independent cohorts, including individuals with normoglycemia and prediabetes.
• Performance was evaluated using mealtime prediction error and PPGR curve area metrics (tAUC, iAUC), with associations to HbA1c and fasting glucose examined.

Main Results:

• The AI algorithm demonstrated accurate PPGR start time prediction with a median error of 10 minutes.
• AI-derived PPGR parameters (tAUC, iAUC) closely matched those derived from ground-truth mealtimes (P > .1).
• AI-derived PPGR iAUC showed significant independent associations with HbA1c (β = 0.57, P = .006) and fasting glucose (β = 0.52, P = .013).

Conclusions:

• The wavelet AI algorithm reliably identifies postprandial glucose responses from CGM data in individuals without diabetes.
• This novel automated approach offers a promising tool for monitoring early signs of postprandial dysglycemia.
• The algorithm's consistent performance across validation cohorts supports its potential clinical utility.