Beyond the Coefficient of Variation: Reframing Glycemic Variability as Dispersion and Volatility

The adoption of the coefficient of variation (%CV) and the ambulatory glucose profile (AGP) as consensus measures of glycemic variability (GV) represented a major advance in diabetes care. These parameters identify individuals at heightened risk of hypoglycemia and are associated with micro- and macrovascular complications. ¹ However, they are often treated as synonymous with GV itself, when in fact they capture only one dimension: the dispersion of glucose values around the mean, irrespective of temporal order.

A second, equally important component of GV is the temporal dimension—volatility: how frequently and rapidly glucose levels change. Emerging evidence shows that volatility is clinically significant. Rapid glucose fluctuations, distinct from dispersion, are linked to acute hypoglycemia-like symptoms during normoglycemia, ² intensive care mortality, ³ future severe hypoglycemia, ⁴ and long-term diabetes progression. ⁵ Various measures have been proposed to capture this temporal aspect, including the recently introduced time with rapid change (TRC) ⁶ and earlier metrics such as Mean-Absolute Glucose change (MAG) and Continuous Overlapping Net Glycaemic Action (CONGA).^7–9 These temporal metrics are highly correlated, ¹⁰ and so any single metric can effectively reflect volatility. However, none have transitioned into routine clinical reporting, where consensus interpretation of GV relies exclusively on dispersion-based metrics like %CV and the AGP. This letter aims to: (1) visually illustrate how consensus reports obscure volatility, (2) present an AGP-analogous visual summary for volatility, and (3) explicitly frame GV along two distinct axes: dispersion and volatility.

The limited visibility of volatility in standard reports is illustrated in Figure 1, which compares two 14-day continuous glucose monitoring (CGM) traces from individuals with type 1 diabetes in the T1DiabetesGranada dataset ¹¹ (see Supplementary Data for data processing methodology). Visual inspection of the raw traces (Panel A) immediately conveys a clinical distinction: Subject A exhibits infrequent, gradual excursions, whereas Subject B shows frequent and rapid fluctuations. Intuitively, Subject B’s control is far more volatile. However, standard metrics (Panel B) do not reflect this. With nearly identical %CV values (30.6% vs. 32.7%) and visually similar AGP plots, both appear to have equivalent and satisfactory control. The metrics correctly report similar glucose dispersion but do not adequately convey the starkly different glucose volatilities.

OPEN IN VIEWER

FIG. 1.

Glycemic variability comprises two dimensions (dispersion and volatility) the latter of which is not captured by standard CV and AGP metrics. (A) Raw glucose traces: Subject A (left) exhibits infrequent, gradual changes, whereas Subject B (right) shows frequent, rapid fluctuations. (B) Standard GV visuals: Despite clear visual differences in their glucose traces, both subjects have comparable CV values (30.6% vs. 32.7%) and nearly identical AGP plots, highlighting that standard metrics capture dispersion but not volatility. (C) Dynamic GV visuals: the AVP and time with rapid change (TRC) charts reveal Subject B’s volatility, with outer shaded regions (95th percentile range) reaching ±2 mg/dL/min. (D) Dispersion versus volatility: a scatterplot of dispersion (CV) against volatility (TRC) reveals only modest correlation (Pearson R² = 0.25). Each blue marker represents one of n = 15,704 individual 14-day glucose traces. Dashed lines indicate the median value for each parameter (CV_median = 36.2%, TRC_median = 3.7%), used to delineate the Tight/Broad and Stable/Volatile ranges. CV_median in this cohort happens to approximate the clinical consensus threshold of 36%. Notably, nearly one-third (32.7%) of individuals with CV <36% still exhibited above-median TRC, as shown by those occupying the lower right quadrant. (E) Correlation matrices between standard CGM metrics and volatility metrics. Volatility metrics (TRC, TRR, and TRF) exhibit only modest Pearson R² correlations with standard level-based metrics (TAR and TBR) or dispersion metrics (SD and CV), indicating that they capture distinct aspects of glycemic behavior. Curly braces indicate subgroups of relatively correlated metrics, corresponding to hyperglycemia, hypoglycemia, dispersion, and volatility. Subjects A and B correspond to patients LIB194126 and LIB193522 respectively in the T1DiabetesGranada dataset. ¹¹ AGP, ambulatory glucose profile; AVP, ambulatory volatility profile; BG, blood glucose CV, coefficient of variation; GV, glucose variability; RoC, rate of change; SD, standard deviation; TAR-L1, time above range (>180 mg/dL); TAR-L2, time above range (>250 mg/dL); TBR-L1, time below range (<70 mg/dL); TBR-L2, time below range (<50 mg/dL); TIR, time in range; TRC, time with rapid change (>±2 mg/dL/min over 15 min); TRF, time with rapid fall (<−2 mg/dL/min over 15 min); TRR, time with rapid rise (>2 mg/dL/min over 15 min).

To illustrate this hidden dimension, I present two complementary visualizations (Fig. 1, Panel C). First, the Ambulatory Volatility Profile (AVP), analogous to the standard AGP, plots the distribution (median, IQR, and 95% range) of glucose rate of change (RoC) throughout the day. This highlights periods of routine rapid rises or falls that are otherwise invisible in standard AGPs. Second, the TRC stacked bar chart, mirrors the familiar Time in Range chart but quantifies the proportion of time spent with rapid glucose changes over 15-min windows: rapid rise (>2 mg/dL/min), very rapid rise (>3 mg/dL/min), rapid fall (<−2 mg/dL/min), and very rapid fall (<−3 mg/dL/min). This visual extension of the TRC metric provides an intuitive signature of volatility, conveying both the direction and intensity of rapid changes, as seen in Subject B’s prominent yellow and orange segments. The proposed thresholds (±2 and ±3 mg/dL/min) approximate the 1–99th and 0.1–99.9th percentiles of glucose RoC in nondiabetic individuals, ¹² and thus represent changes rarely observed in normal physiology, though further work is needed to quantify their clinical significance. Together, these tools make the dynamic dimension of GV visible and quantifiable, and their alignment with established AGP and Time in Range formats facilitates familiarity and ease of adoption by clinicians and patients.

I further suggest refining terminology to reflect the two principal axes of GV (Fig. 1, Panel D). Dispersion (the spread of values, measured by %CV) can be described as Tight (low) or Dispersed (high), while Volatility (the rates of change, reflected by TRC) can be described as Stable (low) or Volatile (high). This yields a simple 2 × 2 taxonomy distinguishing, for instance, a Tight–Volatile glycemic phenotype (small but rapid swings) from a Dispersed–Stable one (slow, wide excursions). This framework highlights potential therapeutic trade-offs. For example, over-bolusing insulin may reliably mitigate upward drift and thereby limit dispersion, but at the expense of inducing rapid glucose declines that increase volatility.

The distinction between Subjects A and B is representative, not exceptional. Across the n = 15,704 14-day glucose traces in the T1DiabetesGranada dataset, dispersion (CV) and volatility (TRC, >±2 mg/dL/min) showed only a modest association (Pearson R² = 0.25, Fig. 1 Panel E). Notably, nearly one-third (32.7%) of individuals meeting the consensus definition of “low GV” (CV < 36%) still exhibited above-median TRC (Fig. 1, Panel D), indicating that low dispersion does not imply stable glucose dynamics. A similar pattern was observed in a type 2 diabetes cohort (Supplementary Fig. S1), supporting the generality of this finding. Importantly, low CV should not be equated with overall stability when volatility is high. Misclassifying “tight-volatile” phenotypes as “stable” risks overlooking the need for targeted intervention, as highlighted by a recent illustrative case. ¹³ More broadly, these findings support adopting a parsimonious, nonredundant set of CGM metrics. This should include at least one dispersion metric (e.g., the consensus %CV) and one volatility metric. I recommend TRC, as its structural similarity to time below and above range facilitates easy clinical adoption and visualization.

In conclusion, dispersion-based metrics such as CV have advanced diabetes care but represent only one axis of GV. Adopting terminology that distinguishes dispersion (tight/dispersed) from volatility (stable/volatile) and integrating visualizations such as the Ambulatory Volatility Profile could help align future treatment goals across both axes of variability, promoting therapy that achieves not only tight but also stable glycemic control.

Author’s Contributions

R.R. performed all aspects of the study, including design, analysis, and article preparation. R.R. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.