Imputation Model for Glucose Values Above the Upper Detection Limit for Continuous Glucose Monitors.

Area of Science:

• Endocrinology
• Biostatistics
• Medical Technology

Background:

• Continuous glucose monitors (CGMs) are crucial for diabetes management.
• CGMs have an upper detection limit (typically 22.2 mmol/L) that can introduce bias in glucose metrics.
• Accurate assessment of glycemic control is essential, especially in cases of severe hyperglycemia.

Purpose of the Study:

• To develop and validate a statistical model for imputing continuous glucose monitor (CGM) values exceeding the upper detection limit.
• To address the potential bias in CGM metrics caused by the upper detection limit.
• To enable more accurate quantification of glycemic control in individuals with severe hyperglycemia.

Main Methods:

• A Bayesian nonparametric latent Gaussian process regression model was employed.
• The model was applied to CGM data from diverse patient groups: inpatients with type 2 diabetes, and outpatients with type 1 and type 2 diabetes.
• Data were intentionally right-censored at 5%, 10%, 20%, and 30% to simulate exceeding the upper detection limit, with performance evaluated against uncensored data using bias, MSE, and R².

Main Results:

• The imputation model demonstrated low bias across various censoring levels for mean glucose, standard deviation (SD), and coefficient of variation (CV) in all patient groups.
• For instance, bias on mean glucose ranged from -0.012 to 0.485 across groups and censoring levels.
• Mean squared error (MSE) and coefficient of determination (R²) also indicated good performance of the imputation model, confirming its accuracy.

Conclusions:

• A robust statistical imputation model for CGM glucose values above the upper detection limit has been successfully developed and validated.
• This model effectively corrects for bias introduced by the upper detection limit of CGMs.
• The validated imputation method allows for more accurate and unbiased assessment of glycemic control, particularly beneficial for patients experiencing severe hyperglycemia.