Capturing Physician Perception

A Consensus Perceived Glycemic Variability Metric

Glycemic variability is an important aspect of blood glucose control that is not routinely assessed in practice. Excessive glycemic variability is associated with poor glycemic control and is a strong predictor of hypoglycemia, which has been linked to excessive morbidity and mortality. Its automated detection would enable routine clinical screening to identify potentially at-risk patients.

There is no definitive metric for glycemic variability; nor is there any clear cut criteria for when it is excessive. However, diabetes specialists readily recognize excessive glycemic variabilty when they see it in blood glucose plots. The goal of this project is to capture that clinician perception, via machine learning algorithms, in order to measure and assess glycemic variability.

A new Consensus Perceived Glycemic Variability (CPGV) metric has been developed that could supplement HbA1c as a measure of overall glycemic control in clinical practice. To develop this metric, 12 physicians managing patients with type 1 diabetes rated 250 24-hour continuous glucose monitoring (CGM) plots as exhibiting low, borderline, high or extremely high glycemic variability. When physician ratings were not unanimous, they were averaged to obtain a consensus. Descriptive features derived from the CGM plots were used as input to machine learning algorithms, which were trained to match consensus ratings.

The best performing model, a support vector regression (SVR) model with a Gaussian kernel, was selected as the new CPGV metric. When judged by the root mean square error (RMSE), this model performed comparably to individual physicians at matching consensus ratings. When applied to 262 different CGM plots as a screen for excessive glycemic variability, this model had accuracy, sensitivity and specificity of 90.1%, 97.0%, and 74.1%, respectively. This was a significant improvement in performance over other measures of glycemic variability, including mean amplitude of glycemic excursion (MAGE), standard deviation, distance traveled and excursion frequency.

24-Hour Continuous Glucose Monitoring Plots Illustrating Excessive and Low Glycemic Variability