📄 medicine

Glycemic Persistence Index: a threshold-free CGM-derived measure of persistent hyperglycemia

The paper introduces the Glycemic Persistence Index (GPI), a novel, threshold-free metric derived from continuous glucose monitoring that quantifies sustained hyperglycemia by jointly capturing its magnitude and duration, demonstrating superior performance in predicting HbA1c levels and distinguishing treatment benefits compared to existing metrics.

Original authors: Ren Zhang

Published 2026-09-11

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Ren Zhang

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

For millions of people living with diabetes, the body's ability to manage sugar is broken. The goal of treatment is to keep blood glucose levels within a safe, narrow band, avoiding the dangerous highs that damage organs over time and the dangerous lows that can cause immediate harm. To track this, doctors and patients now rely on continuous glucose monitors, small sensors worn on the skin that take a reading every few minutes, creating a massive, complex stream of data. This stream tells a story of the day's sugar levels, but the story is often too long and detailed to read easily. Clinicians need a way to summarize thousands of data points into a single, clear number that captures not just how high the sugar went, but how long it stayed high. Without such a summary, it is difficult to compare treatments or understand the true burden of the disease.

A new study introduces a fresh way to solve this problem, offering a single number that describes the persistence of high blood sugar without relying on arbitrary cut-off points. Researchers analyzed data from two large groups of people: one group of 168 individuals with type 1 diabetes participating in a clinical trial, and another independent group of nearly 1,900 people ranging from healthy individuals to those with advanced diabetes. They developed a metric called the Glycemic Persistence Index. This number works by finding the largest integer where the blood sugar stayed at or above that level for at least that many minutes in a single day. For instance, if a person's index is 125, it means their blood sugar was at least 125 mg/dL for at least 125 minutes. It is a simple, threshold-free way to say that the sugar was high and stayed high, combining the height of the spike and the length of time it lasted into one intuitive score.

The researchers first tested this new index in a randomized trial comparing two treatments for type 1 diabetes. One group used an advanced system that automatically adjusted insulin delivery, while the other used a standard pump that required manual input. The study found that the group using the automatic system had a significantly lower Glycemic Persistence Index than the control group. This result confirmed that the new metric could successfully detect when a treatment was working to reduce the burden of sustained high blood sugar. Furthermore, the index proved to be very stable from day to day, showing less fluctuation than other common measures, which suggests it is a reliable tool for tracking changes over time without being thrown off by normal daily variations.

When the researchers applied this index to the larger, more diverse group of nearly 1,900 people, the results revealed a clear pattern. The index values rose steadily as the participants' health status moved from healthy to prediabetes, and then to diabetes requiring medication or insulin. Healthy individuals had the lowest scores, while those with insulin-dependent diabetes had the highest. The index also showed a strong connection to the standard blood test used to measure long-term sugar control, known as HbA1c. In fact, the new index was better at predicting a person's HbA1c level than any other single measure derived from the glucose monitor data, including average blood sugar and the percentage of time spent in a target range. This suggests that the index captures the essence of chronic sugar exposure more effectively than existing tools.

The study concludes that this new measure offers a practical advantage because it does not require doctors or patients to agree on specific numbers that define "high" blood sugar, which can vary between different groups of people. Instead, the index adapts naturally to the data, providing a consistent way to describe how long and how high blood sugar remains elevated. While the researchers note that this index is meant to complement existing tools rather than replace them, and that future studies are needed to see if it can predict long-term health complications, the findings provide a clear, unified way to understand the persistence of high blood sugar. By turning a complex history of glucose levels into a single, understandable number, the Glycemic Persistence Index offers a new lens through which to view the daily reality of diabetes management.

Technical Summary: Glycemic Persistence Index (GPI)

Problem Statement
Sustained hyperglycemia is a primary driver of diabetes complications. While Continuous Glucose Monitoring (CGM) provides high-resolution temporal data, summarizing this complexity into a single, interpretable parameter remains challenging. Existing metrics have limitations:

Threshold-based metrics (e.g., Time-in-Range, Time-Above-Range) require prespecified glucose cutoffs that vary by population and lose gradation once a threshold is crossed.
Mean glucose summarizes overall exposure but fails to distinguish between transient spikes and sustained elevation.
Variability measures describe dispersion but do not integrate magnitude and duration into a unified representation of persistence.

There is a need for a parameter-free, threshold-free scalar that jointly captures the magnitude and duration of sustained hyperglycemia.

Methodology
The author introduces the Glycemic Persistence Index (GPI), a nonparametric metric derived directly from CGM data.

Definition: GPI is defined as the largest integer $k$ $k$ such that the cumulative time spent at glucose levels $\ge k$ $\geq k$ mg/dL is at least $k$ $k$ minutes within a 24-hour period.
- Mathematically: $GPI = \max \{k \in \mathbb{N} : \sum_{\{i: G_i \ge k\}} \Delta t_i \ge k\}$ .
- This formulation is conceptually analogous to the $h$ -index in bibliometrics and the Durfee square in integer partition theory.
- Units: The index is unitless but interpretable as "minutes spent at or above $k$ mg/dL" (e.g., GPI = 125 implies glucose was $\ge 125$ mg/dL for at least 125 minutes).
Data Sources:
1. DCLP3 Trial: A randomized controlled trial (RCT) involving 168 participants with Type 1 Diabetes (T1D) comparing closed-loop therapy vs. sensor-augmented pump therapy. CGM data (Dexcom G6) were analyzed at the subject-day level.
2. AI-READI Cohort: An independent, multimodal cohort of 1,898 participants spanning healthy individuals, prediabetes, oral/non-insulin diabetes, and insulin-dependent diabetes. This cohort was used to validate GPI across the glycemic spectrum and anchor it to HbA1c.
Statistical Analysis:
- DCLP3: Analysis of Covariance (ANCOVA) was used to compare post-randomization GPI between groups, adjusting for baseline. Variance decomposition assessed the relationship between GPI and Time-in-Range (TIR).
- AI-READI: Group differences were assessed via Kruskal-Wallis tests. Associations with HbA1c were evaluated using Pearson and Spearman correlations. Receiver Operating Characteristic (ROC) analysis determined GPI cutoffs for HbA1c-defined states ( $\ge 5.7\%$ and $\ge 6.5\%$ ). Single-metric prediction of HbA1c severity was benchmarked using repeated 10-fold cross-validation ( $R^2$ ).

Key Contributions

Novel Metric: Introduction of GPI as a simple, threshold-free scalar that integrates glucose magnitude and duration into a single value without modeling assumptions or smoothing.
Geometric Interpretation: GPI is defined as the intersection of the ranked glucose curve and the identity line ( $y=x$ ), providing a direct visual and mathematical interpretation of persistence.
Validation Across Spectrums: Demonstration of GPI's utility in both a controlled T1D RCT and a broad observational cohort ranging from normoglycemia to insulin-dependent diabetes.
Benchmarking: Comprehensive comparison of GPI against standard CGM metrics (Mean Glucose, GMI, TIR, TAR, etc.) regarding treatment sensitivity, day-to-day stability, and HbA1c prediction.

Results

Treatment Sensitivity (DCLP3): GPI successfully captured the benefit of closed-loop therapy. The closed-loop group showed a significantly lower mean GPI compared to the control group (adjusted difference: $-15.04$ points; $P < 0.0001$ ).
Differentiation within TIR: While GPI is inversely associated with TIR, it captures residual variation. Two days with identical TIR (68%) exhibited markedly different GPI values (311 vs. 155), indicating GPI can distinguish the severity of hyperglycemic burden even when time-in-range is similar.
Stability: GPI demonstrated low day-to-day variability (median Coefficient of Variation = 0.148), comparable to mean glucose and lower than TIR.
Clinical Distribution (AI-READI): GPI showed a progressive upward shift across glycemic groups: Healthy (Median $\approx 142$ ) $\to$ Prediabetes ( $\approx 148$ ) $\to$ Oral/Non-insulin Diabetes ( $\approx 174$ ) $\to$ Insulin-dependent Diabetes ( $\approx 206$ ).
HbA1c Association: GPI was strongly correlated with HbA1c (Pearson $r = 0.771$ $r = 0.771$ ).
- Discrimination: GPI achieved an AUC of 0.933 for distinguishing HbA1c $\ge 6.5\%$ .
- Optimal Cutoffs: A GPI of 170.8 yielded 88.6% sensitivity and 85.2% specificity for HbA1c $\ge 6.5\%$ .
Predictive Performance: In single-metric prediction of rank-transformed HbA1c, GPI ranked first among all tested CGM metrics in the overall cohort ( $R^2 = 0.470$ ) and in healthy, prediabetes, and oral/non-insulin subgroups. It slightly trailed TAR >180 mg/dL in the insulin-dependent subgroup.

Significance and Claims
The paper claims that GPI offers a concise, intuitive, and clinically anchored summary of sustained glycemic burden. Its primary significance lies in its threshold-free nature, which avoids the arbitrariness of cutoff selection and allows for continuous gradation of severity. By jointly encoding magnitude and duration, GPI provides a unified representation of hyperglycemia that is immediately interpretable (e.g., "glucose was $\ge 120$ mg/dL for 120 minutes").

The author positions GPI not as a replacement for consensus metrics (TIR, TAR, TBR) but as a complementary adjunctive metric. It is particularly valuable for:

Providing a reference range for healthy individuals (centered near 140).
Offering a single scalar that correlates strongly with HbA1c, potentially simplifying the assessment of chronic glycemic exposure.
Capturing nuances of hyperglycemic persistence that threshold-based metrics may miss.

The study concludes that GPI is a mathematically distinct, assumption-free tool that enhances the interpretability of CGM data for both specialists and non-specialists, though future work is needed to link specific GPI levels to long-term clinical outcomes.

Technical Summary: Glycemic Persistence Index (GPI)

More like this