Glucodensities: A new representation of glucose profiles using distributional data analysis

Abstract

Biosensor data have the potential to improve disease control and detection. However, the analysis of these data under free-living conditions is not feasible with current statistical techniques. To address this challenge, we introduce a new functional representation of biosensor data, termed the glucodensity, together with a data analysis framework based on distances between them. The new data analysis procedure is illustrated through an application in diabetes with continuous-time glucose monitoring (CGM) data. In this domain, we show marked improvement with respect to state-of-the-art analysis methods. In particular, our findings demonstrate that (i) the glucodensity possesses an extraordinary clinical sensitivity to capture the typical biomarkers used in the standard clinical practice in diabetes; (ii) previous biomarkers cannot accurately predict glucodensity, so that the latter is a richer source of information and; (iii) the glucodensity is a natural generalization of the time in range metric, this being the gold standard in the handling of CGM data. Furthermore, the new method overcomes many of the drawbacks of time in range metrics and provides more in-depth insight into assessing glucose metabolism.

Keywords: CGM technology; biosensor data; diabetes; distributional data analysis.

Figure 1.

Glucodensities are estimated from a random sample of the AEGIS study including normoglycemic and patients with diabetes. For each patient, our glucose representation estimates the proportion of time spent at each glucose concentration over a continuum, representing a more sophisticated approach to assess glucose metabolism. Currently, the time in range metrics that are the gold standard CGM data representation in diabetes only quantify glycemic distributional differences along the previously predefined target zones that correspond to coarsely defined intervals, resulting in information loss.

Figure 2.

Real values vs. estimated values when glucodensity is predictor.

Figure 3.

Residuals in quantile space when predicting glucodensities.

Figure 4.

Real values vs. estimated values when time in range metric is the predictor. Blue, time in range metric with cut-offs calculated with normoglycemics from the AEGIS database. Red, time in range metric using of cut-offs suggested by ADA.

Figure 5

(Left two panels) Glucodensities for men and women of the AEGIS study, plotted as quantile functions; (Third panel) 2-Wasserstein mean quantile functions for each group (Fourth Panel). Cross-sectional standard deviation curves for quantile functions in each group.

Figure 6.

Clustering analysis of diabetes patients in AEGIS study.