Clinical Value of HbA1c/_updatedGMI Ratio in Identifying Early Retinopathy Risk in Type 1 Diabetes

Abstract

Background:

Discrepancies between HbA1c and glucose management indicator (GMI) may reflect individual variations in glycation rate, independent of mean glycemia, and could influence complication risk stratification in type 1 diabetes (T1D). We evaluated the phenotype of individuals with T1D using continuous glucose monitoring (CGM), identified as high glycators based on HbA1c/_updatedGMI ratio, and assessed retrospectively their risk of diabetic retinopathy (DR) and the time to DR diagnosis. The secondary aim was to identify clinical correlates of high glycation. Primary outcome: time to first diagnosis of DR. Secondary outcomes: clinical factors associated with high glycation.

Methods:

A retrospective study of 411 individuals with T1D using CGM and concurrent HbA1c values. Patients with conditions affecting red blood cell (RBC) lifespan were excluded. Participants were divided into 3 subgroups based on current HbA1c/_updatedGMI ratio ≤0.95 (low glycators), >0.95 and <1.05 (concordant glycators), and ≥1.05 (high glycators). Time to diagnosis of DR was retrieved.

Results:

High glycation is associated with shorter time to first diagnosis of DR (adjusted hazard ratio 1.60). Non-HDL-C, RBC indices, and metformin were associated with high glycation.

Conclusion:

Among individuals with T1D, an HbA1c/_updatedGMI ratio ≥1.05 is associated with higher odds of DR. Non-HDL-C and RBC indices are correlates of high glycation. These results underscore the relevance of HbA1c and _updatedGMI discrepancy in cardiometabolic risk assessment, but cutoffs remain to be set.

Introduction

Laboratory-measured glycated hemoglobin (HbA1c) is the gold standard for assessing glycemic control and predicting microvascular outcomes, although continuous glucose monitoring (CGM)-derived metrics are considered more informative for recent glucose exposure. ¹ Measures derived from mean glycemia—eHbA1c (from seven-point capillary profiles), glucose management indicator (GMI) and other surrogates (eg, fructosamine)—highlighted a “glycation gap” issue as potential driver of vascular damage. These indices do not necessarily match with corresponding HbA1c. Statistical interdependence among these variables remains poorly understood.^2-5

McCarter et al. showed that an elevated high glycation gap (identified by hemoglobin glycation index [HGI], the difference between HbA1c and eHbA1c) associated with higher risk of microvascular complications in type 1 diabetes (T1D). ⁶ Subsequent studies exploring this association, using ratios or absolute difference between HbA1c and GMI, were inconsistent,^4,5,7-9 suggesting that glycation may not always capture glycemic exposure. After CGM was introduced, it became clear that many patients show discrepancies between HbA1c and GMI that impact patient management,^10-14 with high (HbA1c > GMI) and low glycators (GMI > HbA1c) at risk of over-treatment/under-treatment based on targeting HbA1c.

Glycated hemoglobin is a weighted average, more influenced by glycemia exposure over the last month, while GMI is a linear regression from mean glucose, these two surrogates being not equivalent.^12,15 Changes in HbA1C and GMI after glucose-lowering intervention showed modest correlation across clinical trials. ¹⁶ GMI also overestimated HbA1c among 71% of non-diabetic individuals, compared with 39% of participants in cohort used to develop the GMI equation. ¹⁷

Despite HbA1c considered an established predictor of microvascular risk, GMI represents the true glucose exposure, the critical factor driving microvascular complications. ¹⁸ Since HbA1c is influenced by numerous biological variables, ¹⁹ it is unclear whether microvascular damage in discordant patients is driven by hyperglycemia reflected by HbA1c or GMI, or whether reduced glycation afford protection against microangiopathy.

Some factors contributing to HbA1c and GMI discordance are inherent to red blood cell (RBC) lifespan (eg, hemoglobin variants) or interindividual variations in RBC characteristics, ethnicity and CGM characteristics, and wear conditions.^20,21 Furthermore, the first GMI equation was derived from a mostly White European population wearing DexCom sensors. ¹² Changes in MARD (mean absolute relative difference) between sensors may also add within-patient and between-patient bias.^22,23

An updated GMI (_uGMI) equation demonstrates stronger correlation with HbA1c than both eHbA1c and the original GMI. ²⁴ It is derived from a kinetic model that incorporates RBC turnover and overall hemoglobin glycation constants, including transmembrane glucose transport, validated in adult/pediatric cohorts.^25,26

We aimed to characterize individuals with T1D wearing CGM according to HbA1c/_uGMI ratio, to determine whether phenotypes differ between low, concordant, and high glycators, and to assess whether discordance relates to microangiopathies, focusing on diabetic retinopathy (DR) diagnosis. The secondary aim was to identify clinical correlates of high glycation.

Methods

Study Population

This retrospective study used the electronic medical record from adults with T1D (January 2024-December 2024) for patients with HbA1c contemporaneous with CGM (synchronous/within 1 week). Exclusion criteria were: conditions affecting RBC lifespan ¹¹ ; acute anemia; vitamin B₁₂/B₉/iron deficiency; mean corpuscular volume (MCV) <70 fL; estimated glomerular filtration rate (eGFR) ²⁷ <30 mL/min/1.73 m²; hemodialysis; pregnancy; hemoglobin variants; new insulin pump users, organ transplantation, and sensor wear <70%.

Patients’ Characteristics

These included demographics, diabetes duration (DD), hypertension history, smoking, insulin pump use, CGM metrics, glucose-lowering, and cardiometabolic drugs. Glucose management indicator was calculated using the recently published equation derived from the kinetic model that accounts for RBC turnover and overall hemoglobin glycation, which includes cross-membrane glucose transport: updated GMI = 1/(15.36/average glucose (mg/dL) + 0.0425. ²⁴ The _uGMI demonstrates the strongest correlation with HbA1c, outperforming other indicators of glucose exposure. The screening schedule for DR is proposed in accordance with the recommendations of the American Academy of Ophthalmology. Time to diagnosis of diabetic DR was retrospectively retrieved, defined as interval from T1D diagnosis to first DR diagnosis, with staging from Diabetic Retinopathy Preferred Practice Pattern. ²⁸ Clinical presentations were as follows: no apparent or presence of DR, DR (mild, moderate, severe, and proliferative) and/or diabetic maculopathy. Albuminuria was defined as albumin/creatinine ratio ≥30 mg/g. Decreased eGFR was <60 mL/min/1.73 m². Chronic kidney disease (CKD) was diagnosed as albuminuria, decreased eGFR or both. Only CKD occurring in long-standing hyperglycemia not preceding diabetes diagnosis in patients without (micro)albuminuria at T1D diagnosis was considered microangiopathic. Diabetic neuropathy, diabetic foot disease (DF) erectile dysfunction, coronary artery disease (CAD), cerebrovascular disease (CeVD), peripheral artery disease (PAD), and heart failure (HF) were recorded.

Assessment of High Glycation

Discrepancy between HbA1c and GMI was assessed according to current HbA1c (%)/_uGMI (%) ratio. We choose a ratio since absolute differences may not carry similar clinical meaning across glycemic exposure. Such ratio also reduces heteroscedasticity, enhancing comparability across glycemic strata (White test, P < .001) and accounts for nonlinearities in HbA1c and GMI relationship, whereby original GMI tends to underestimate at high HbA1c and overestimate at low HbA1c. A Delphi panel (n = 15) identified clinically meaningful discordance thresholds, due to lack of consensual cutoff, with median cutoff was HbA1c/_uGMI ratio of ≤0.95 or ≥1.05.

A cutoff of 1.10 was reported^3,4 and absolute cutoffs, as per McCarter, ⁶ were used. While a difference of 0.5% between two consecutive HbA1c is often regarded as clinically relevant, such differences can also arise from inherent analytical variability. External Quality Assessment programs show that a single sample can produce substantially different HbA1c results across laboratories (interlaboratory variability up to 0.8%). Accordingly, a difference of 0.8% between HbA1c and GMI may be considered a pragmatic threshold for discordance. This threshold accounts for both analytical variability and clinical relevance, while underscoring that individual clinical judgment remains essential, particularly when differences are consistently observed. ²¹ To ensure robustness, multiple Cox models were used to explore additional HbA1c/_uGMI ratio thresholds and association with time to DR diagnosis. Absolute differences (HbA1c – GMI) were also calculated following McCarter’s HGI definition and examined in Cox regression analyses.

For the primary outcome, participants were stratified into three groups according to current HbA1c/_uGMI ratio: ≤0.95 (low glycators; G1), >0.95 and <1.05 (concordant glycators; G2), and ≥1.05 (high glycators; G3). These designations should, however, be interpreted with caution, as the biological mechanisms driving HbA1c and GMI discordance are poorly understood. Same stratification was made according to current HbA1c/_uGMI ratio: ≤0.90 (low glycators; G1), >0.90 and <1.10 (concordant glycators; G2), and ≥1.10 (high glycators; G3).

We performed a sub-analysis restricted to patients with at least two pairs of HbA1c and _uGMI, thereby averaging the ratio across a longer time window.

HbA1c was measured on Tosoh G8, with performance established against HLC-723G7: slope = 0.9, y-intercept = –0.46, correlation coefficient = 0.996; coefficient of variability: 0.2% to 0.6%. All patients wore FDA/EMA (Food and Drug Administration/European Medicines Agency)-approved CGMs: DexcomG7, FreeStyle Libre 1/2/3, GuardianConnect, or Guardian4. Last 60-day CGM metrics were used. Other measurements included eGFR, hemoglobin, red cell distribution width-coefficient of variation (RDW-CV), RDW-standard deviation, MCV, routine lipids, and lipoprotein (a).

Statistical Analyses

These were performed with STATA (StataCorp. Stata Statistical Software: Release 18. College Station, Texas: StataCorp LLC; 2023). To assess differences between groups, mean comparison tests were conducted using Student t test or analysis of variance (ANOVA) if more than two groups were compared. Kaplan-Meier estimates on DR as event and DD as observation window were also computed, patients with DR of unknown event date were excluded. Estimates were carried out and compared across G1, G2, and G3. Cox regressions were used to compute hazard ratios and P values, both with and without additional covariates. Multivariate logistic regressions estimated the impact of high-glycation phenotype on presence of DR. Multivariate logistic regressions estimated the impact of several variables on high-glycation probability.

Results

Flow chart

Of 439 patients, 28 were excluded: 14 pre-end-stage renal disease, two transplantation, three corpuscular volume below 70 fL, one splenectomy, one pregnancy, two not consistently wearing sensors, and five without HbA1c matching CGM, leaving a cohort of 411 patients (Figure 1: flow chart).

Figure 1.

Study flowchart. Of the 439 patients identified, 28 were excluded for the following reasons: two did not consistently wear their sensor, five had no HbA1c values corresponding with the CGM, 14 had pre-end-stage renal disease, two had undergone organ transplantation, three had a corpuscular volume below 70 fL, one had undergone a splenectomy, and one was pregnant.

Abbreviations: CGM, continuous glucose monitoring; HbA1c, glycated hemoglobin A1c.

Patients’ Characteristics

Mean age was 49.1 years (17.5), men represented 57.4%; and DD was 26.2 years. Continuous glucose monitoring brands were FreeStyle3 (n = 232; 56.4%), FreeStyle1 (n = 77; 18.7%), Guardian4 (n = 53; 12.9%; in patients on Medtronic 780G closed-loop), FreeStyle2 (n = 31; 7.5%), DexComG7 (n = 17; 4.1%), GuardianConnect (n = 1; 0.2%). Mean HbA1c was 7.8%, GMI 7.8%, and HbA1c/GMI ratio 1.01. The general characteristics of the entire cohort and glycation groups G1 (n = 94), G2 (n = 203), and G3 (n = 114) are shown in Table 1. Twenty-eight percent were high glycators based on our definition. Glycation groups differed in gender, metformin and RAAS blocker use, RBC indices, and pre-lipid-lowering drugs (pre-LLD)/current non-HDL-C and pre-LLD remnant cholesterol (RC).

Table 1.

Clinical and Biological Characteristics According to Glycation Status.

		Cutoff 0.95 and 1.05					Cutoff 0.9 and 1.10
	Total	G1	G2	G3	P (ANOVA)	P (G3 vs 1-2)	G1	G2	G3	P (ANOVA)	P (G3 vs 1-2)
Age, y	49.1 (17.5)	46.6 (17.4)	49.4 (17.5)	50.5 (17.5)	0.245	0.292	45.5 (17.7)	49.4 (17.3)	48.7 (18.5)	0.245	0.292
White Caucasian, %	92.0	92.5	91.1	94.7	0.866	0.601	92.8	92.1	94.4	0.978	0.981
Men (birth-assigned sex, %	57.4	59.6	62.6	46.5	0.019	0.005	53.6	59.3	48.1	0.284	0.140
Age at diabetes diagnosis, y	23 (15)	22 (17)	22 (14)	24 (16)	0.429	0.196	25 (16)	22 (15)	24 (15)	0.466	0.411
Diabetes duration, y	26 (15)	24 (16)	27 (14)	26 (15)	0.230	0.834	20 (13)	27 (15)	25 (13)	0.048	0.378
Diabetes duration in non-DR, y	19 (1)	16 (1)	21 (1)	19 (2)	0.100	0.900	15 (2)	20 (1)	18 (3)	0.182	0.456
Smoking (%)	30.0	27.6	28.6	34.5	0.471	0.224	32.1	29.0	35.2	0.639	0.378
Current smoking (%)	11.4	7.4	11.3	14.9	0.243	0.170	10.7	10.3	18.5	0.215	0.080
HbA1c (%)	7.8 (1.2)	7.3 (0.9)	7.7 (1.0)	8.5 (1.5)	<0.001	<0.001	7.0 (0.6)	7.7 (1.0)	8.9 (1.7)	<0.001	<0.001
HbA1c (mmol/mol)	62 (13)	56 (10)	61 (11)	69 (16)	<0.001	<0.001	53 (7)	60.7 (11)	73.8 (19)	<0.001	<0.001
uGMI (%)	7.8 (1.0)	8.1 (0.9)	7.7 (1.0)	7.6 (1.2)	0.005	0.045	8.2 (0.7)	7.8 (1.0)	7.7 (0.5)	0.074	0.546
% CSII	23.8	14.9	24.6	29.8	0.039	0.078	10.7	23.7	31.5	0.111	0.158
% with HbA1c <7%	23.8	41.5	21.7	13.1	<0.001	0.002	50.0	24.6	5.5	<0.001	0.001
% with uGMI <7%	23.6	8.5	23.6	36.0	<0.001	<0.001	3.5	22.2	42.6	<0.001	<0.001
MBG (mg/dL) from CGM	181 (38)	191 (34)	179 (36)	176 (43)	0.013	0.100	194 (28)	180 (36)	180 (49)	0.133	0.799
TIR (%)	51 (18.0)	47 (15)	52 (16)	53 (22)	0.021	0.170	45 (11)	51 (17)	52 (24)	0.233	0.812
TBR (%)	5 (5)	4 (4)	4 (4)	5 (6)	0.188	0.122	4 (4)	5 (5)	5 (6)	0.561	0.576
Glycemic variability reported as CV (%)	40.4 (9.3)	40.5 (6.3)	40.5(8.0)	40.3 (13.1)	0.975	0.823	42.0 (5.2)	40.5 (9.6)	39.1 (8.1)	0.446	0.322
HbA1c/uGMI	1.01 (0.09)	0.9 (0.04)	1.0 (0.03)	1.1 (0.06)	<0.001	<0.001	0.85 (0.01)	1.0 (0.00)	1.16 (0.01)	<0.001	<0.001
HbA1c-uGMI (%)	0.1 (0.7)	-0.8 (0.4)	0.0 (0.2)	0.9 (0.5)	<0.001	<0.001	-1.2 (0.4)	0.0 (0.4)	1.2 (0.6)	<0.001	<0.001
Statins (%)	50.8	34.0	54.7	57.9	<0.001	0.077	28.6	52.3	53.7	0.050	0.654
Ezetimibe (%)	30.0	21.2	31.0	35.1	0.086	0.157	14.3	30.1	33.3	0.161	0.599
Fibrate (%)	10.2	8.5	9.8	12.2	0.653	0.393	7.1	10.3	11.1	0.844	0.817
Metformin (%)	5.8	1.1	6.0	9.6	0.031	0.041	3.6	5.5	9.3	0.476	0.251
RAAS blocker (%)	30.4	21.3	29.5	39.5	0.016	0.013	25.0	29.5	38.9	0.310	0.147
Hypertension (%)	52.8	45.7	52.7	58.8	0.173	0.133	46.4	52.9	55.5	0.734	0.664
BMI (kg/m²)	26.0 (4.6)	26.1 (4.6)	25.5 (4.7)	26.5 (4.5)	0.174	0.114	26.2 (4.7)	25.8 (4.6)	26.7 (4.7)	0.424	0.212
Hemoglobin (g/dL)	14.1 (1.4)	14.4 (1.4)	14.1 (1.3)	13.8 (1.4)	0.003	0.003	14.5 (1.3)	14.1 (1.4)	13.8 (1.4)	0.061	0.065
RDW-CV (%)	12.9 (1.0)	12.7 (0.8)	12.8 (1.1)	13.1 (1.0)	0.016	0.006	12.9 (1.0)	12.8 (1.0)	13.3 (1.3)	0.018	0.005
RDW-SD (fL)	40.9 (3.7)	40.2 (3.6)	41.0 (4.0)	41.3 (3.0)	0.125	0.216	41.2 (5.3)	40.8 (3.6)	41.6 (3.5)	0.443	0.243
MCV (fL)	87.3 (5.9)	87.8 (4.8)	87.8 (4.4)	85.9 (8.6)	0.033	0.008	88.1 (6.9)	87.4 (6.0)	86.2 (5.2)	0.385	0.200
Pre-LLD non-HDL-C (mg/dL)	131 (43)	125 (43)	128 (37)	142 (52)	0.019	0.006	133 (44)	130 (42)	138 (49)	0.597	0.331
Current non-HDL-C (mg/dL)	104 (34)	106 (31)	100 (33)	109 (39)	0.106	0.085	111 (34)	103 (34)	107 (38)	0.345	0.420
Pre-LLD remnant cholesterol (mg/dL)	22 (18)	22 (24)	19 (11)	26 (14)	0.051	0.032	22 (10)	21 (18)	26 (23)	0.387	0.168
Current remnant cholesterol (mg/dL)	20 (16)	23 (25)	18 (11)	22 (14)	0.019	0.174	23 (30)	20 (15)	21 (13)	0.663	0.688
Pre-LLD LDL-C (mg/dL)	110 (38)	105 (36)	109 (34)	118 (48)	0.114	0.056	111 (40)	110 (38)	114 (40)	0.849	0.588
Current LDL-C (mg/dL)	84 (31)	84 (29)	82 (31)	86 (34)	0.538	0.337	92 (31)	83 (31)	86 (34)	0.316	0.541
Atherogenic dyslipidemia (%)	7.8	8.5	5.9	10.5	0.332	0.203	10.7	6.7	13.0	0.239	0.130

HbA1c/_uGMI ratio threshold of 1.05: G1: n = 94; G2; n = 203 G3; n = 114.

HbA1c/_uGMI ratio threshold of 1.10: G1: n = 28; G2; n = 329 G3; n = 54.

Abbreviations: BMI, body mass index; CV, coefficient of variation; CSII, continuous subcutaneous insulin infusion; G1, group 1 defined as low glycators; G2, group 2 defined as concordant glycators: G3, group 3 defined as high glycators; _uGMI, updated glucose management indicator; pre-LLD, value before lipid-lowering drug initiation; RAAS, renin-angiotensin-aldosterone system; TBR, time below range; TIR, time in range; y, years.

Microvascular Target Organ Damages

Overall prevalence of all-cause microangiopathy and DR were 52.5% (n = 216) and 47.2% (n = 194), respectively. There was a positive gradient in frequency of microangiopathy from G1 to G3. Time from diabetes onset to first diagnosis of DR was significantly shorter in G3 (18.5 vs 23.0 (G1) and 22.7 years (G2), respectively; P = 0.02). G3 patients had higher urinary albumin-creatinine ratio (mg/g) (Table 2).

Table 2.

Micro and Macrovascular Diseases According to Glycation Status.

		Cutoff 0.95 and 1.05					Cutoff 0.9 and 1.10
	Total	G1	G2	G3	P (ANOVA)	P (G3 vs 1-2)	G1	G2	G3	P (ANOVA)	P (G3 vs 1-2)
Overall microvascular disease (%)	52.5	46.8	52.7	57.0	0.342	0.263	42.8	51.7	63.0	0.174	0.101
Diabetic retinopathy (%)	47.2	41.5	49.7	47.4	0.416	0.967	35.7	47.7	50.0	0.432	0.659
Time to first diagnosis of DR, y	22 (10)	23 (12)	23 (11)	18 (8)	0.076	0.023	24 (13)	22 (11)	18 (7)	0.165	0.072
Polyneuropathy (%)	20.8	17.0	22.4	21.2	0.569	0.901	17.8	19.6	30.2	0.195	0.072
Diabetic foot disease (%)	2.9	1.1	3.2	3.9	0.476	0.470	3.7	2.3	6.2	0.296	0.133
Erectile dysfunction in men (%)	24.1	22.6	21.7	31.4	0.385	0.169	20.0	22.3	40.0	0.142	0.049
CKD (%)	19.7	18.1	17.8	24.6	0.318	0.130	18.1	19.2	22.2	0.854	0.626
UACR (mg/g)	28.6 (153)	17.5 (48.5	19.5 (69.7)	54.2 (273)	0.119	0.039	22.7 (49.9)	22.3 (86.0)	71.1 (369)	0.101	0.032
UACR >30 mg/g (%)	13.9	12.8	11.8	18.4	0.245	0.099	14.3	12.5	22.2	0.158	0.057
eGFR (mL/min.1.73 m²)	88.5 (22.4)	93.5 (23.8)	88.0 (20.6)	85.2 (23.8)	0.028	0.066	98.2 (25.0)	87.6 (21.3)	88.8 (26.4)	0.054	0.905
eGFR, 30-60 mL/min.1.73 m² (%)	10.2	9.6	8.7	13.1	0.470	0.224	7.1	9.7	14.8	0.447	0.232
Overall macrovascular disease (%)	12.9	9.6	12.3	16.7	0.299	0.158	10.7	12.1	18.5	0.409	0.187
CAD (%)	9.2	8.5	8.4	11.4	0.647	0.351	10.7	8.5	13.0	0.558	0.313
CeVD (%)	4.4	2.1	4.4	6.1	0.373	0.281	0	4.2	7.4	0.291	0.244
PAD (%)	2.9	1.1	2.0	4,9	0.219	0.083	3.6	2.4	5.6	0.442	0.218
Heart failure (%)	1.8	2.2	0.0	4.8	0.011	0.007	3.6	1.3	3.9	0.317	0.216

HbA1c/_uGMI ratio threshold of 1.05: G1: n = 94; G2; n = 203 G3; n = 114.

HbA1c/_uGMI ratio threshold of 1.10: G1: n = 28; G2; n = 329 G3; n = 54.

Abbreviations: CAD, coronary artery disease; CeVD, cerebral vascular disease; CKD, chronic kidney disease; DR, diabetic retinopathy; G1, group 1 defined as low glycators; G2, group 2 defined as concordant glycators: G3, group 3 defined as high glycators; PAD, peripheral artery disease; UACR, urinary albumin-creatinine ratio.

In Cox regressions, patients with HbA1c/_uGMI ratio ≥ 1.05 had higher probability of DR, with hazards ratio (HR) of 1.6 (two-sided P = 0.057) after adjustment for multiple variables: _uGMI, hypertension, smoking, sex, time in range (TIR) ≥ 70%, hemoglobin, non-HDL-C pre-LLD, MBG, type of sensor, CSII use, and _uGMIpump (Figure 2). A ratio of 1.10 was associated with a significantly higher risk (HR = 1.90; two-sided P = 0.041) (Figure 3 and Table 3). When modeled as a continuous variable, the HbA1c/_uGMI ratio was associated with a hazard ratio of 16, with a two-sided P value of 0.037.

Figure 2.

Kaplan-Meier curve time to DR according to low—concordant and high-glycation status, respectively (log rank G3 vs G1 and G2: 0.342). According to Cox regression analysis, an HbA1c/GMI ratio of 1.05 was significantly associated with a shorter time to DR diagnosis, with HR of 1.6 (two-sided P = 0.057) after controlling for _uGMI, hypertension, smoking, sex, TIR ≥ 70%, hemoglobin, non-HDL-C pre-LLD, MBG, type of sensor, CSII, _uGMIpump. Kaplan-Meier estimates are based on DR as event and DD as observation window. Legend: G1, group 1; G2, group 2; G3, group 3.

Figure 3.

Kaplan-Meier curve time to DR diagnosis according to low-concordant and high-glycation status, respectively (log rank G3 vs G1 and G2: 0.140). According to Cox regression analysis, an HbA1c/_uGMI ratio of 1.10 was significantly associated with a shorter time to DR diagnosis, with HR of 1.9 (two-sided P = 0.042) after controlling for _uGMI, hypertension, smoking, sex, TIR ≥ 70%, hemoglobin, non HDL-C pre LLD, MBG, type of sensor, CSII, _uGMIpump. Kaplan-Meier estimates are based on DR as event and DD as observation window. Legend: G1, group 1; G2, group 2; G3, group 3.

Table 3.

Cox Regression Hazard Ratios for Diabetic Retinopathy According to Different HbA1c/GMI Ratio Thresholds.

HbA1c/uGMI ratio
	≥1.10	≥1.05	≥1.04	≥1.03	≥1.02	≥1.01
Non adjusted OR G3	1.44	1.18	1.33	1.41	1.35	1.27
Non-adjusted two-sided P value	(0.140)	(0.346)	(0.086)	(0.040)	(0.062)	(0.142)
Adjusted OR G3	1.9	01.6	1.60	1.50	1.50	1.46
Adjusted two-sided P value	(0.041)	(0.057)	(0.041)	(0.088)	(0.071)	(0.090)

A total of 274 patients were included in the regression model.

Excluded: patients with an unknown date of diabetes onset.

Covariables for adjustment: _uGMI, hypertension, smoking, sex, TIR ≥ 70%, hemoglobin, non-HDL-C pre LLD, MBG, type of sensor, CSII, _uGMI × pump.

Abbreviations: CSII, continuous subcutaneous insulin infusion; G3, group 3 defined as high glycators; indicator; MBG, mean blood glucose; pre-LLD, value before lipid-lowering drug initiation; TIR, time in range, _uGMI, updated glucose management.

Table 4 shows multiple Cox regressions as regard absolute differences (HbA1c – _uGMI). A cutpoint ≥0.8% was associated with a significantly higher risk (HR = 1.80; two-sided P = 0.090). Among 112 patients with absolute difference ≥0.4%, all had ratio ≥1.05 except for four (ratio between 1.04 and 1.05). Among 299 patients with absolute difference <0.4%, only six had a ratio ≥1.05.

Table 4.

Cox Regression Hazard Ratios for Diabetic Retinopathy According to Different Hemoglobin Glycation Index Thresholds.

	Absolute difference between HbA1c and uGMI
	≥0.80%	≥0.50%	≥0.40%	≥0.30%	≥0.20%	≥0.10%
Non adjusted OR G3	1.4	1.2	1.3	1.4	1.5	1.3
Non-adjusted two-sided P value	(0.223)	(0.312)	(0.169)	(0.035)	(0.021)	(0.124)
Adjusted OR G3	1.8	1.3	1.5	1.5	1.7	1.5
Adjusted two-sided P value	(0.090)	(0.236)	(0.115)	(0.074)	(0.021)	(0.075)

A total of 274 patients were included in the regression model.

Excluded: patients with an unknown date of diabetes onset.

Covariables for adjustment: _uGMI, hypertension, smoking, sex, TIR ≥ 70%, hemoglobin, non-HDL-C pre-LLD, MBG, type of sensor, CSII, _uGMI × pump.

Abbreviations: CSII, continuous subcutaneous insulin infusion; G3, group 3 defined as high glycators; indicator; MBG, mean blood glucose; pre-LLD, value before lipid-lowering drug initiation; TIR, time in range, _uGMI, updated glucose management.

Sub-analysis With Smoothed HbA1c/_uGMI Ratio

To assess within-individual variability in a subgroup of 127 patients, we compared the maximum and minimum HbA1c/_uGMI ratios recorded for each participant. Mean interval between first and last HbA1c/_uGMI ratio was 2 years (±2 years). Overall, 59 patients changed glycation category when applying the 1.05 cutoff, including 29 who entered or left the high-glycation group (G3). Using the 1.10 cutoff, 27 patients shifted categories, of whom 15 transitioned into or out of G3.

When comparing the current ratio with the mean ratio derived from repeated paired measurements, 23 patients changed glycation category under the 1.05 cutoff (including 13 who moved into or out of G3), and 13 patients changed category under the 1.10 cutoff (including 7 transitioning into or out of G3). A paired t-test showed no significant difference between the current and mean HbA1c/_uGMI ratios (1.00 [95% confidence interval [CI] = 0.98-1.01] vs 1.04 [95% CI = 0.97-1.10]; P = 0.24).

In Cox regressions, patients with average HbA1c/_uGMI ratio ≥1.05 had higher probability of DR, with HR of 4.8 (86 observations; two-sided P = 0.003) after adjustment for above-mentioned variables. A mean ratio of 1.10 was associated with a significantly higher risk (HR = 6.9; two-sided P = 0.001). When modeled as a continuous variable, the average HbA1c/_uGMI ratio was associated with a hazard ratio of 1.8, with a two-sided P value of 0.078.

High-Glycation Correlates

Multiple logistic regression models were performed, including different sets of covariates and variables of interest to adjust for confounders. RBC indices, and pre-LLD non-HDL-C were significantly associated with odds of being high glycator. Metformin use showed a trend toward association with high glycation (Table 5).

Table 5.

Odds Ratios of Adjusted Logistic Regressions on High-Glycation Probability (Based on an HbA1c/_uGMI Ratio of ≥1.05 and 1.1).

HbA1c/uGMI ratio	1.05		1.10		Adjusted for
	OR	Two-sided P value	OR	Two-sided p-value
					Sex	RAAS	Hemoglobin (g/dL)	eGFR (mL/min.1.73 m3)	Age, y	BMI (kg/m2)	Current smoking	Hypertension
Pre-LLD non-HDL-C (mg/dL)	1.01	0.008	1.01	0.253	x			x	x	x
MCV (fL)	0.94	0.023	0.98	0.356	x		x	x	x		x
Pre-LLD remnant-C (mg/dL)	1.01	0.057	1.01	0.181	x			x	x	x
RDW-CV (%)	1.26	0.062	1.37	0.040	x		x	x	x		x
Metformin	2.05	0.100	1.57	0.404						x
Hemoglobin (g/dL)	0.85	0.104	0.82	0.130	x			x	x		x
RAAS	1.6	0.123	1.78	0.157				x	x			x
Current smoking	1.62	0.141	2.08	0.063	x				x
eGFR (mL/min.1.73 m3)	0.99	0.467	1.01	0.516	x	x			x	x		x

Abbreviations: MCV, mean corpuscular volume; pre-LLD, value before lipid-lowering drug initiation; RAAS, renin-angiotensin-aldosterone system; RDW-CV, red blood cell distribution width-coefficient of variation.

Discussion

This study assessed the association between high glycation defined as HbA1c/_uGMI ratio and DR diagnosis in T1D patients. Three findings emerge: (1) dysglycation is common, notwithstanding within-subject variability; (2) high glycation is associated with shorter time to first diagnosis of DR; and (3) high glycation correlates with RBC indices and higher non-HDL-C.

A high-glycation phenotype was associated with microvascular complications.^4,6,8 Prior to CGM, such increased risk was reported from Diabetes Control and Complication Trial, which stratified participants into tertiles of HGI. At 7 years of follow-up, third HGI tertile (≥0.42%) patients had earlier diagnosis of DR compared with second and first HGI tertile patients.^6,19,29 A recent study reported no difference in incident DR between high glycators (absolute HbA1c–GMI difference ≥ 0.4%) and non-high-glycators in a relatively small cohort of younger participants, shorter DD (~18 years), and lower prevalence of complications, but was possibly underpowered to detect a difference. ⁹ Sager la Ganga et al. report no association between GMI/HbA1c ratio and DR prevalence in logistic regression models. ⁵ We obtained similar result after adjustment to their variables. Extending our analysis to account for the temporal dimension—time to DR diagnosis—showed, however, that high-glycation phenotype associates with shorter time to DR diagnosis.

Glycated hemoglobin is a surrogate marker of protein glycation underlying target organ damage. GLUT1 is a facilitative glucose transporter predominantly expressed in erythrocytes and endothelial cells, including retinal microvasculature.^30,31 Intracellular hyperglycemia in target organs drives mitochondrial overproduction of reactive oxygen species, activating the polyol pathway, advanced glycation end product (AGE) formation, protein kinase C and nuclear factor kappa B signaling, and the hexosamine pathway. These processes collectively promote oxidative stress and vascular injury. ¹⁸ Resulting intraretinal hypoxemia/pseudohypoxia fosters neovascularization and upregulates GLUT1, increasing glucose uptake in retinal endotheliocytes and further amplifying pseudohypoxic signaling implicated in DR. ³² Increased rate of glucose entry through GLUT1 into RBC may be a surrogate of enhanced glucose entry into retinal endotheliocytes. Patients with faster glucose transport via GLUT1 may exhibit enhanced pseudohypoxic signaling. Moreover, several studies reported an association between high glycation and AGEs. Notably, higher HGI or HbA1c/GMI ratio were linked to increased skin AGEs and microvascular complications, particularly DR.^4,7,8 These findings suggest that a positive discordance (HbA1c > surrogate_HbA1c) reflects systemic protein glycation. We excluded patients with dysglycation from abnormal RBC lifespan, assuming that patients with HbA1c > GMI were genuine high glycators, with greater glucose uptake via GLUT1 in retinal endothelial cells, promoting pseudohypoxic signaling and AGE formation.

A kinetic model, from which the _uGMI was subsequently derived, computed theoretical HbA1c values adjusted for RBC glucose uptake and lifespan in T1D and support this hypothesis. These computations were compared with HbA1c and concurrent GMI. Theoretical HbA1c showed stronger correlation with HbA1c than GMI, highlighting significant contribution of erythrocyte-related factors to hemoglobin glycation, supporting the concept of “individual glucose-HbA1c relationship.” Microvascular complications were associated with discordant relationship between glucose levels and HbA1c, likely reflecting increased RBC glucose uptake, and appeared to correlate with RBC kinetic parameters supporting the concept of “individual glucose-HbA1c relationship.”^26,33,34 Individuals with microvascular complications exhibited higher glycation constants. These observations highlight the relationship between intracellular RBC glucose levels and the development of microvascular complications, suggesting that glucose uptake rates may be coordinated between RBCs and non-erythrocytic cells within target organs susceptible to microangiopathy.^26,35 Recent evidence indicates that theoretical HbA1c is a slightly stronger predictor of DR compared with other glucose exposure metrics. Furthermore, the finding that a combined model incorporating both HbA1c and _uGMI provides slightly superior predictive performance reinforces the hypothesis that discordance between these metrics may further refine DR risk stratification. ²⁴

Thus, high glycation of hemoglobin for comparable given glucose exposure would confer microangiopathy risk, confirming the added value of HbA1c as biomarker of enhanced microvascular risk, even among patients with T1D whose glycemic control is deemed satisfactory from current metrics such as the GMI or TIR.

As for HbA1c and GMI mismatch, two scenarios arise. First, a negative difference (HbA1c < GMI) from shortened RBC lifespan may lead to both under-treatment and underestimation of vascular risk, as it does not reflect genuine hyperglycemia. Accordingly, TIR may be preferred to both HbA1c and GMI as primary metric of optimal glucose control. Second, a positive difference (HbA1c > GMI) unrelated to abnormal RBC lifespan may raise vascular risk as a result of whole-body proteins glycation and pseudohypoxic signaling. ²⁰ For such patients, HbA1c may be a preferred measure to GMI and/or TIR for assessing risk of long-term vascular complications. Caution should be exercised to avoid overtreatment and hypoglycemia risk among patients with HbA1c > GMI, while integrating patient self-management and real-time glucose metrics, aiming for the highest feasible TIR, with low average normal glucose and short time below range.

Previous studies reported that age, MCV, and creatininemia were associated with over- or underestimation of HbA1c.^10,13,36,37 Factors affecting hematopoiesis are key confounders of HbA1c level, yet they ought not to affect incident microvascular complications at similar glucose exposure. Non-HDL-C, elevated RDW-CV, low MCV, and metformin were associated with hemoglobin high-glycation phenotype.

Higher HGI is linked to total cholesterol, triglycerides, and other metabolic markers in non-diabetic populations at risk of hepatic steatosis,,^38,39 as well as in T1D. ⁸ The mechanisms remain unclear, but associations with obesity and central adiposity suggest that visceral fat and triglycerides may influence HGI via inflammation and oxidative stress-driven glycation.^8,39,40 Some studies have reported an association between RC, increased variability of RC and microvascular damage.^41-43

Red cell distribution width-coefficient of variation, a measure of RBC size variability (anisocytosis), is elevated in conditions such as iron deficiency, B12/folate deficiency, or post-transfusion and is correlated with RBC longevity. It is linked to inflammation, oxidative stress, and predicts CV risk, nephropathy, and mortality, including in people with diabetes.^44,45 Red cell distribution width correlates positively with HbA1c independently of glycemia,^46,47 and has been associated with differences between HbA1c and GMI, reflecting a potential non-glycemic association with hemoglobin glycation. ¹³ Although RBC indices, such as RDW, were correlated with HbA1c and differed by race, these factors did not fully account for the higher HGI observed in individuals of African ancestry, and ethnicity remained an independent predictor in multivariable models. ⁴⁸

Metformin enhances glucose transport across erythrocytes, promoting HbA1c glycation^49,50 In vitro data are echoed by a clinical study showing independent association between metformin use and HbA1c glycation. ³⁶ Patients on metformin may have lower glucose exposure than inferred from HbA1c, yet it is not known whether metformin promotes glucose transport across non-erythrocytic cells. In newly diagnosed T2D treated with metformin, the ratio of glycated albumin to HbA1c significantly decreased over 24 weeks, suggesting metformin may shift glycation markers. ⁵¹

Smoking was not associated with high glycation in this study. However, smoking generates pseudo-hypoxia in (pre)erythrocytes, with upregulation of GLUT1 in RBC membrane observed by immunoblotting and immunocytochemistry. Glucose uptake in whole blood was also higher among smokers. ⁵² Compared with healthy controls, patients with T2D—regardless of complications—had increased erythrocyte fragility, lipid peroxidation, and GLUT1 expression. ⁵³

This study has several limitations. First, there is currently no universally accepted definition of discordance between HbA1c and _(u)GMI. We tested several alternative cutoffs to challenge the Delphi panel threshold selected. Establishing clinically meaningful cutoffs will require additional studies in larger cohorts and ultimately rely on expert consensus.

Second, there are few data on longitudinal consistency of HbA1/_uGMI ratio. There is no evidence that the observed level of discordance in our study remains consistent over time. Whereas the gaps between HbA1c and fructosamine-derived HbA1c, and between HbA1c and eHbA1c appear stable over time,^54,55 whether HbA1c/_uGMI ratio exhibits fair temporal stability is unknown. One study demonstrated that the difference between HbA1c and GMI varied over time in 30% to 50% of participants suggesting fluctuations in concordance between laboratory-measured and sensor-derived metrics in many patients. ⁹ A recent study, using the same ratios as ours, showed substantial within-subject variability over 2 years. ⁵⁶ It remains unclear whether variability evens out over the long term, and whether patients ultimately remain within their respective categories. Finally, date of DR diagnosis can only serve as an approximation of DR onset, being subject to error.

Conclusion

In conclusion, CGM introduced challenges in interpreting HbA1c. While interstitial glucose, and by extension GMI, is a true reflection of glucose exposure with a strong correlation to capillary glucose, it remains debated whether high glycation of hemoglobin reflects that of target organs (kidney, retina, and nerves). This study provides evidence that hemoglobin high glycation compared with GMI is associated with early DR. These two parameters provide complementary insights into diabetes management and should therefore be considered jointly in everyday clinical care.