Pathophysiology/Complications ORIGINAL ARTICLE

Association of Glycation Gap With Mortality and Vascular Complications in

1,2 4 ANANTH U. NAYAK, MRCP PAUL BASSETT, MSC fructosamine reflects blood glucose attain- 3 1 ALAN M. NEVILL, PHD BALDEV M. SINGH, MD ment over a much shorter time frame than HbA1c and may more readily be influenced by very short-term changes in blood glu- OBJECTIVEdThe “glycation gap” (G-gap), an essentially unproven concept, is an empiric cose levels. It may simply be that the G-gap measure of disagreement between HbA1c and fructosamine, the two indirect estimates of glyce- is no more than an empiric and potentially mic control. Its association with demographic features and key clinical outcomes in individuals spurious measure of disagreement between with diabetes is uncertain. the two indirect estimates of glycemic con- RESEARCH DESIGN AND METHODSdThe G-gap was calculated as the difference trol, with each having a number of con- between measured HbA1c and a fructosamine-derived standardized predicted HbA1c in 3,182 founders to the direct relationship with individuals with diabetes. The G-gap’s associations with demographics and clinical outcomes blood glucose. (retinopathy, nephropathy, macrovascular disease, and mortality) were determined. Although we have demonstrated that RESULTSd fi the G-gap is a consistent phenomenon Demographics varied signi cantly with G-gap for age, sex, ethnic status, smoking within individuals over time (1), there re- status, type and duration of diabetes, insulin use, and obesity. A positive G-gap was associated with retinopathy (odds ratio 1.24 [95% CI 1.01–1.52], P = 0.039), nephropathy (1.55 [1.23– mains doubt as to whether the G-gap is a 1.95], P , 0.001), and, in a subset, macrovascular disease (1.91 [1.18–3.09], P = 0.008). In Cox real phenomenon or if it has any signif- regression analysis, the G-gap had a “U”-shaped quadratic relationship with mortality, with both icant sequelae (15). Hypothesizing that negative G-gap (1.96 [1.50–2.55], P , 0.001) and positive G-gap (2.02 [1.57–2.60], P , 0.001) the G-gap is an inconsequential nonsys- being associated with a significantly higher mortality. tematic event, irrelevant to diabetes out- CONCLUSIONSd fi comes, it would not then be expected to We con rm published associations of G-gap with retinopathy and ne- be associated with distinct subpopula- phropathy. We newly demonstrate a relationship with macrovascular and mortality outcomes tions of human diabetes or to have any and potential links to distinct subpopulations of diabetes. sequelae in clinical outcomes. This pa- per explores the association of the G-gap with diabetic population demographic factors and with crucial clinical out- he glycation gap (G-gap) refers to the key deglycating enzyme, fructosamine- comes to determine if such associations T potential deviation of glycated 3-kinase, has isoforms and a genetic poly- exist. HbA1c away from the other indirect morphism suggested to influence HbA1c estimate of blood glucose attainment variability, but any impact on HbA1c glyca- such that it might read substantially tion is unknown; although it seems un- RESEARCH DESIGN AND lower or higher than expected (1–3). Gly- likely that glycated HbA1c is a substrate METHODS cated HbA1c represents the net effect of for this enzyme since it has been shown several mechanisms, which may shift its that there is no evidence that it plays any Patient selection direct glycation relationship with overall role in HbA1c deglycation at the relevant We reviewed all HbA1c and fructosamine levels of glycemia (4–6). Many factors glycation site (11,12). To add to the poten- estimations undertaken at New Cross are known to influence HbA1c, includ- tial for a spurious generation of a G-gap, Hospital over 4 years (2006–2009), iden- ing various erythrocytic processes (6–9). many factors, including variability in tifying and selecting all adults with dia- Protein glycation is a nonenzymatic reac- protein turnover and obesity, may affect betes ($18 years of age) who had paired tion dependent on glucose concentrations, fructosamine estimation (1,13,14). The estimations of HbA1c and fructosamine but intracellular enzymatic deglycation of evidence concerning the effects of urinary performed on the same day from the proteins has also been identified (10). The protein loss are mixed (1,13). Even then, same sample set. Thereafter, clinical in- formation was taken from our diabetes ccccccccccccccccccccccccccccccccccccccccccccccccc registry and linked to this dataset. The From the 1Wolverhampton Diabetes Centre, New Cross Hospital, Wolverhampton, U.K.; the 2University diabetes register is validated to be Hospital of North Staffordshire National Health Service Trust, Stoke on Trent, U.K.; the 3Research In- . fi 4 99% accurate for the identi cation of stitute in Health Care Sciences, University of Wolverhampton, U.K.; and Statsconsultancy Ltd., known diabetes and for mortality status Amersham, U.K. Corresponding author: Ananth U. Nayak, [email protected]. in linkage with the National Health Ser- Received 30 May 2012 and accepted 29 April 2013. vice Strategic Tracing Service. Pregnant DOI: 10.2337/dc12-1040 women, those with a creatinine .200 This article contains Supplementary Data online at http://care.diabetesjournals.org/lookup/suppl/doi:10 mmol/L, those with a known hemoglo- .2337/dc12-1040/-/DC1. © 2013 by the American Diabetes Association. Readers may use this article as long as the work is properly binopathy, or those with an abnormal cited, the use is educational and not for profit, and the work is not altered. See http://creativecommons.org/ electrophoretic pattern on HbA1c testing licenses/by-nc-nd/3.0/ for details. were excluded. care.diabetesjournals.org DIABETES CARE 1 Diabetes Care Publish Ahead of Print, published online July 8, 2013 Glycation gap and mortality in diabetes

Retinopathy grading, not derived from HbA1c by correlation/ means was by one-way ANOVA and the microalbuminuria, and regression methods. The normalized differences between frequencies/propor- macrovascular risk standard deviate reallocation of fructos- tions by x2 test. Binary logistic regression Digital retinal screening was in accor- amine levels yields fructosamine-based was used to determine the association of dance with the English National Screen- HbA1c equivalent results with the same various independent factors with dichot- ing Program for Diabetic Retinopathy distribution, mean, and SD as HbA1c omized variables. Survival analysis was (ENSPDR) (16). Retinopathy was cate- without altering the rank position of the undertaken using Cox regression. In gorized into a dichotomized variable fructosamine-derived value. A negative each case, a stepwise backward extraction (with or without any retinopathy). Urine G-gap denotes the true HbA1c appearing method excluded all nonsignificant vari- albumin-creatinine ratio (UACR) was to read lower than the FHbA1c, and a pos- ables (P . 0.05) to determine the sim- assessed as dichotomous variable divid- itive G-gap denotes the true HbA1c appear- plest, most parsimonious model. Data ing into lower risk or higher risk for pro- ing to read higher than that predicted by are presented as the mean 6 SD. All sta- gressive microalbuminuria (,10 or .10 fructosamine. Among those with a second tistical tests were considered significant at mg/mmol) (17). Individuals were catego- paired HbA1c-fructosamine estimation, in P , 0.05. rized as having established macrovascular order to identify those with a consistent disease depending on the presence or ab- G-gap direction, the product of two G-gaps Ethical committee approval sence of any previous cardiac, cerebral, or was calculated. If consistent, the G-gap The use of the clinical database for this peripheral macrovascular event. product would be positive (positive 3 pos- study was approved by the relevant local itive = positive; negative 3 negative = pos- U.K. National Health Service Research Analytical methods itive), but any discordance in direction of Ethical Committee. HbA1c International Federation of Clini- the G-gap over time in two paired readings cal Chemistry values were available only would yield a negative G-gap product (neg- RESULTSdOf 4,757 patients identi- from 1 June 2009. Hence we have used ative 3 positive = negative). fied, 3,182 had complete demographic the Diabetes Control and Complications For the whole cohort (n =3,182), data and were included. Their follow-up – fi r2 fi Trail (DCCT) aligned HbA1c in our anal- HbA1c was associated signi cantly ( = from the rst paired HbA1c-fructosamine ysis. HbA1c was measured using high- 0.10, F = 26.9, P , 0.001) with age, estimation to the time of death or study performance liquid chromatography on sex, diabetes duration, smoking status, end point was 38 6 16 months. a Tosoh G7 analyzer (Tosoh Bioscience retinopathy status, serum albumin, and Table 1 shows the glycation esti- Ltd.,Worcestershire,U.K.).Theperfor- UACR but not with ethnicity, diabetes mates. The correlation between HbA1c mance scores in the UK National External type, BMI, serum creatinine, estimated and fructosamine in the first HbA1c- Quality Assurance Scheme (UK NEQAS) glomerular filtration rate (eGFR), macro- fructosamine pair is r = 0.75, P , 0.001 were as follows: A (accuracy) score ,100 vascular disease, or mortality status. Fruc- (n = 3,182). Nevertheless, the G-gap and B (bias) score ,2%, which were tosamine (r2 = 0.20, F = 62.4, P , 0.001) range demonstrates the substantial mag- within the acceptable limits of the UK was associated with age, ethnicity, diabe- nitude of variation between HbA1c and NEQAS for glycated hemoglobins (maxi- tes type, diabetes duration, BMI, retinop- FHbA1c, both of which indicated com- mum limits: A score ,200 and B score athy grade, and UACR but not with sex, pletely differing assessments of attainment less than 67.5%). The between-batch co- smoking status, serum creatinine, eGFR, of glycemic control. The distribution of efficient of variation was 1.8 and 1.4% for serum albumin, macrovascular disease, or G-gap status for the whole group varied 2 an HbA1c of 5.7% (39 mmol/mol) and 9.5% mortality status. The actual versus regres- significantly by HbA1c quintile (x = (80 mmol/mol), respectively. Fructosamine sion model–predicted values (incorporat- 505.8, P , 0.001), noting the striking in- was measured by nitrotetrazolium-blue ing the statistical effect of the factors crease in negative G-gap status in the reduction on a Roche Modular P analyzer outlined above) for HbA1c (8.6 6 1.8 vs. lowest HbA1c quintile, whereas the pos- (Roche Diagnostics Ltd., West Sussex, 8.6 6 0.6%) and fructosamine (313 6 74 itive G-gap prevalence was graded across U.K.) using a Cobas kit with between- vs. 313 6 33 mmol/L) were not signifi- ascending quintiles (Fig. 1). Repeat batch coefficient of variation 3.1% at a cantly different and the crude values HbA1c-fructosamine estimations were level of 263 mmol/L and 2.2% at 518 were thus used throughout. undertaken 11 6 10 months after the mmol/L (18). first in 1,609 patients. There was a qua- G-gap categorization dratic relationship (r2 =0.67,P , 0.001) Calculation of the fructosamine- The G-gap (unit = HbA1c %) was catego- between the first and second G-gap (as predicted HbA1c and the G-gap rized as negative, neutral, or positive described in RESEARCH DESIGN AND METHODS) 2 As published (1), a predicted HbA1c when less than or equal to 1 (i.e., with only 47 (3%) and 17 (1%) of the 1,609 (FHbA1c) was calculated from the simul- more negative than 21), greater than patients discordant at a G-gap product taneously measured fructosamine stan- 21 to less than +1, or greater than or more negative than 20.5 and 21.0, dardized to the HbA1c distribution equal to +1, respectively. This categoriza- respectively. according to the following equation: tion was taken from our previously pub- Thereweresignificant differences FHbA1c = {[(fructosamine – mean fruc- lished clinical error grid analysis of the between G-gap categories in a number tosamine)/SD fructosamine] 3 SD impact of G-gap on assessment of glyce- of relevant demographic characteristics HbA1c} + mean HbA1c.TheG-gapwas mic control (19). (Table 1). The key clinical outcomes of thedifferencebetweenthetrueHbA1c retinopathy (borderline significance), ne- and the fructosamine-derived standard- Statistical analysis phropathy (UACR), established macro- – ized predicted FHbA1c (G-gap = HbA1c Data were analyzed on SPSS version 19. vascular disease, and mortality also varied FHbA1c). Importantly the FHbA1c was Comparison between multiple group significantly with G-gap status (Table 1).

2 DIABETES CARE care.diabetesjournals.org Nayak and Associates

Table 1dThe biochemical, demographic, and clinical characteristics of the cohort, categorized according to G-gap status

G-gap category (HbA1c %) Negative (# 21) Neutral (.21to, +1) Positive ($ +1) Number (%) 586 (18) 1,945 (61) 651 (21) HbA1c (%) 7.9 6 1.8 8.2 6 1.5 9.9 6 2.0 P , 0.001 Fructosamine (mmol/L) 365 6 91 301 6 66 298 6 75 P , 0.001 FHbA1c (%) 9.7 6 2.1 8.2 6 1.5 8.2 6 1.7 P , 0.001 G-gap (HbA1c %) 21.8 6 0.8 0.0 6 0.5 +1.7 6 0.8 P , 0.001 Age (years) 61 6 18 66 6 14 64 6 13 P , 0.001 Sex (% male) 63 54 47 P , 0.001 Ethnicity (% white, Asian, black) 68, 6, 16 62, 28, 10 62, 32, 6 P , 0.001 Smoking status (% never, ex, current) 64, 27, 9 62, 29, 9 57, 28, 15 P =0.006 Diabetes type (% type 2) 60 85 90 P , 0.001 Duration of diabetes (years) 20 6 11 17 6 9166 8 P , 0.001 On insulin (%) 76 68 76 P , 0.001 BMI (kg/m2) 28.8 6 5.6 31.8 6 6.1 34.4 6 7.4 P , 0.001 Systolic blood pressure (mmHg) 136 6 21 138 6 21 137 6 12 P =0.09 Diastolic blood pressure (mmHg) 73 6 12 73 6 12 74 6 11 P =0.38 Cholesterol (mmol/L) 4.4 6 1.1 4.3 6 1.0 4.4 6 1.2 P =0.05 Retinopathy (% any) 62 61 66 P =0.05 UACR (% .10 mg/mmol) 15 17 24 P , 0.001 UACR (mg/mmol) 10 6 56 12 6 48 22 6 70 P , 0.001 Serum creatinine (mmol/L) 93 6 30 92 6 29 87 6 28 P =0.001 eGFR (mL/min) 80 6 27 76 6 26 79 6 27 P =0.003 Macrovascular risk (% established) 24 30 33 P =0.001 Vital status (number [%] dead) 87 (15) 196 (10) 108 (16) P , 0.001

Binary logistic regression analyses macrovascular disease (1.91 [1.18–3.09], categorized according to G-gap status were undertaken to determine the rela- P = 0.008). (negative vs. neutral: OR 1.96 [95% CI tionship between the absence or presence The mortality pattern with G-gap dif- 1.50–2.55], P , 0.001; positive vs. neu- ofthesediabetesoutcomesandtheG-gap fered and was clearly not linear but rather tral: 2.02 [1.57–2.60], P , 0.001). Thus categories (negative, neutral, and posi- “U” shaped. In Cox regression analysis, the for mortality, in contrast to retinopathy, tive, as defined), taking into account G-gap association was significant only as a nephropathy, and macrovascular disease other identified relevant significant fac- quadratic nonlinear U-shaped relationship where the neutral and negative G-gap tors (age, sex, ethnic status, smoking (overall x2 =307.3,P , 0.001). The signif- groups did not significantly differ from status, diabetes type, duration of diabetes, icant factors were age (P , 0.001), smoking each other, both the negative and positive insulin use, and BMI). The overall models (P , 0.001), ethnicity (P = 0.003), and G-gap groups had a significantly worse were all significant (P , 0.001) for each G-gap(squaredterm)(P , 0.001) but not outcome than the neutral group. outcome (Table 2). Within that, indepen- sex, BMI, type or duration of diabetes, and dent of the other significant factors, the insulin use. Introducing the prevailing CONCLUSIONSdThe entire notion fi G-gap effect was signi cant for retinopathy HbA1c (latest value) into the model had no of a G-gap must be treated with skepti- and UACR, and the outcomes were worse significant effect (P = 0.082). cism and caution. There are extensive with a positive G-gap category (Table 2). Similarly the G-gap retained its signif- confounding factors that are real caveats The G-gap status did not retain significance icant association with mortality indepen- to its meaning. Whether they relate to with macrovascular disease prevalence (P = dent of proteinuria. Indeed proteinuria lost erythrocytic function, biochemical path- 0.28) after regression model adjustment for its significant association with mortality ways, or pure statistical error, mecha- other factors. when the most heavily proteinuric subjects nisms for the G-gap are, as yet, not at all When considering only those 1,609 were excluded (UACR .200 mg/mmol) understood. It would be appropriate to with repeat estimates and consistency in whereas G-gap retained significance. Fur- clearly state that there is no known their G-gap over time and repeated mea- thermore, G-gap continued to be signifi- genetic or biochemical mechanism that sures, 549 had a consistently negative (less cantly associated with mortality even provides anything remotely close to a than or equal to 21HbA1c %[n =235]) when a completely normoalbuminuric definitive explanation. or positive (greater than or equal to +1 population subset was analyzed (n = Yet, we now show that the G-gap n x2 fi HbA1c %[ = 314]) G-gap. Belonging to 2,077; 1,873 alive and 204 dead; = varies signi cantly with demographic the consistently positive G-gap group was 179.02, P , 0.001; G-gap OR 1.05 characteristics and is associated with the significantly associated with worsening [1.02–1.09], P , 0.01). See Supplemen- key diabetes outcomes: retinopathy, ne- retinopathy (odds ratio [OR] 1.96 [95% tary Data online for Supplementary Table phropathy, macrovascular disease, and CI 1.31–2.9], P , 0.001), increasing 1 and annotation. mortality. Variation in demographics UACR (1.85 [1.14–3.01], P =0.012) Fig. 2 clearly shows the differing mortal- with the G-gap status has not been pre- but now also the presence of established ity outcome between the groups (P , 0.001) viously reported. The G-gap is consistent care.diabetesjournals.org DIABETES CARE 3 Glycation gap and mortality in diabetes

reports of any relationship between the G-gap and mortality. The long-term follow-up of the UK Prospective Diabetes Study (UKPDS) co- hort suggested some benefit for macro- vascular outcome and mortality with lower HbA1c levels (28), but the conclu- sion of other studies and meta-analyses has demonstrated little or no impact of HbA1c on either macrovascular events or mortality (29–34). The ACCORD trial of intensification of therapy to a target , HbA1c 6.0% (42 mmol/mol) stands out as having led to increased mortality with tighter glycemic targets for uncertain reasons (35). In a retrospective cohort study using the U.K. General Practice Re- search Database, Currie et al. (36) showed increased risk of all-cause mortality with both lower and higher HbA1c levels d x2 proposing a U-shaped association with Figure 1 The prevalence of negative and positive G-gap status by HbA1c quintile ( =505.8, , – the lowest risk at an HbA level of 7.5% P 0.001). The HbA1c (mean [range]) for the quintiles of HbA1c: 1) 6.4% (3.8 7.0), 2)7.5% 1c (7.1–7.8), 3) 8.3% (7.9–8.6), 4) 9.2% (8.7–9.7), and 5)11.3%(9.8–19.0). (58 mmol/mol). Given the general failure to link HbA1c levels with mortality out- comes, our observation of an increased over time (1–3), and twin studies suggest mean of 6.5 years, dividing the cohort prevalence of a negative G-gap at lower it to have significant inheritability (20). into tertiles based on the average of all HbA1c levels and of a positive G-gap at G-gap consistency, potential inheritability, individual G-gaps, and showed that the higher HbA1c levels, both associated and the now-reported demographic link- mean G-gap predicts the progression of with adverse mortality outcomes in a ages tantalizingly point toward human di- nephropathy. In an alternative non– U-shaped pattern that mirrors the obser- abetic subpopulations with biological fructosamine-based approach, the hemo- vations of Curie et al. (36), clearly offers variation in any underlying pathophysio- globin glycation index (HGI), the G-gap an avenue for further exploration. logical mechanisms. was calculated as the difference between It has been argued that the any asso- Hyperglycemia is central to the de- the measured HbA1c minus an HbA1c pre- ciationoftheG-gapwithoutcomes, velopment of diabetes complications dicted from date-matched mean blood whether calculated from fructosamine or (21–23). Hyperglycemia-induced protein glucose estimations (3). In a study by HGI, is a statistically spurious outcome of glycation is an unequivocally important McCarter et al. (26) analyzing the data regression analysis with the anchor HbA1c pathophysiological mechanism (21,24). from DCCT, HGI was shown to be a sig- value (15). In all other published meth- Any factors significantly altering glycation nificant predictor of retinopathy and ne- odologies of the ascertainment of the may theoretically alter the relationship phropathy. To our knowledge, ours is the G-gap (2,3,25,26,37), an HbA1c equivalent between glucose and the development of first published study confirming some po- from fructosamine or blood glucose data diabetes complications. Glycated HbA1c tential association between the G-gap and has been derived by regression analysis. has been shown to correlate with the macrovascular disease. Our methodology specifically avoids this. risk for developing microvascular compli- With a relationship between G-gap With our methodology, we have previously cations in diabetes (22,23). The G-gap is and diabetes vascular complications, shown that over time and repeated meas- proposed as a measure of the deviation of mortality would be expected to follow a ures, the G-gap remains consistent within fi glycated HbA1c away from its expected similar pattern. This was not so. Adjusted subjects despite signi cant within-subject value, such that a negative G-gap is taken all-cause mortality was higher both in the variations in HbA1c and fructosamine and as meaning a lesser level of glycation than negative and positive G-gap groups. The that the variation away from HbA1c is larger expected and a positive G-gap more so. limitations of our study are manifest with than statistically expected by Altman-Bland Our observations, that the micro- and it being a cross-sectional, retrospective analysis (1,19). macrovascular complications of diabetes study that was neither designed nor pow- There is a great need to be cautious are directly associated with a positive ered to address mortality, and we have no about the G-gap, and many caveats must G-gap, are logically consistent with the data on cause of death, noting that diabetes be attached to this concept. As well as the glycation mechanism for complications. is associated with increased morality from concern that the G-gap may well be a spu- Others have reported a relationship be- both vascular and a variety of nonvascular rious statistical phenomenon, there is tween the G-gap and retinopathy and causes (27). It would be tempting to con- concern about the use of fructosamine. nephropathy (2,25,26). Cohen et al. (2) sug- jecture on reasons why a positive G-gap Fructosamine represents the glycation of gested that the G-gap increased the risk of might be associated with mortality, given a number of proteins although predomi- more advanced nephropathy 2.9-fold. the macrovascular association, but we can nantly albumin, the time frame of repre- Rodrıguez-Segade et al. (25) studied offer no true explanation in light of the over- sentation of glycemic attainment may be 2,314 patients with for a all effect. There are no previous published shorter than that of HbA1c (remembering

4 DIABETES CARE care.diabetesjournals.org Nayak and Associates

Table 2dBinary regression analysis showing the association of G-gap status and various other factors with adverse diabetes outcomes

Any retinopathy UACR .10 mg/mmol Any macrovascular P , P , P , Overall model (x2) 324.5, 0.001 81.6, 0.001 481.1, 0.001 OR (95% CI)* for individual factors G-gap category P =0.016 P , 0.001 P = 0.28 Negative vs. neutral 0.84 (0.68–1.05), P =0.12 0.93(0.70–1.23), P = 0.61 0.96 (0.75–1.23), P = 0.74 Positive vs. neutral 1.24 (1.01–1.52), P = 0.039 1.55 (1.23–1.95), P , 0.001 1.17 (0.95–1.45), P = 0.14 Age (years) 0.99 (0.98–0.99), P , 0.001 1.02 (1.01–1.02), P , 0.001 1.06 (1.05–1.07), P , 0.001 Sex (male vs. female) 1.26 (1.07–1.49), P = 0.006 1.41 (1.15–1.73), P = 0.001 1.55 (1.29–1.85), P , 0.001 Ethnicity P =0.002 P =0.001 P , 0.001 Asian vs. Caucasian 1.35 (1.10–1.66), P = 0.004 1.59 (1.25–2.01), P , 0.001 1.42 (1.14–1.76), P =0.002 Black vs. Caucasian 1.45 (1.10–1.91), P = 0.008 1.21 (0.87–1.67), P = 0.25 0.61 (0.45–0.81), P =0.001 Smoking (ever vs. never) 0.86 (0.72–1.02), P =0.08 1.38(1.12–1.71), P = 0.003 1.54 (1.27–1.85), P , 0.001 Diabetes type (type 2 vs. type 1) 1.31(0.98–1.75), P =0.07 1.11(0.77–1.59), P = 0.57 2.05 (1.46–2.89), P , 0.001 Diabetes duration (years) 1.08 (1.07–1.09), P , 0.001 1.02 (1.01–1.03), P = 0.001 1.02 (1.01–1.03), P =0.001 On insulin (yes vs. no) 1.50 (1.24–1.81), P , 0.001 1.03 (0.81–1.31), P = 0.84 1.97 (1.60–2.44), P , 0.001 BMI (kg/m2) 1.01 (1z00–1.02), P =0.22 1.02(1.01–1.04), P = 0.003 1.03 (1.01–1.04), P , 0.001

*For continuous variables (age, diabetes duration, and BMI), the OR represents the change in risk per unit change in the variable (e.g., increase per year of age), whereas for categorical variables, it is with highest risk category.

that HbA1c glycation itself is most influ- terms, the relationship of fructosamine to studies that have linked HbA1c to mortal- enced by glucose levels over the preced- (log)UACR was nonsignificant (P . 0.4). ity, if any link actually exists, the relation- ing 30 days), the glycation product Although in end-stage renal failure there is ship is complex and the factors of linkage assessed is not as specificasdefined for well-known unreliability of HbA1c (indeed ill understood (41,42). In any case, the HbA1c, its glycation may mirror protein fructosamine may be the better estimate), it G-gap itself is not fully associated with turnover rates and protein loss as pro- does not seem appropriate to extend these glycemic control, as indicated by the teinuria, and it is influenced by shorter concerns into the G-gap outcomes in our weak correlation with HbA1c (r = 0.38, time frame changes. Thus, subjects who cohort tested (39). Finally, we can clearly P = 0.001, variance explained [r2]= tightened up on diet, lifestyle, and other state that the relationship of G-gap to mor- 14%) and fructosamine (r = 20.33, P = treatment prior to their blood testing or tality was independent of proteinuria. That 0.001, variance explained = 11%). Fi- those who had intercurrent illness with is to say that although proteinuria and nally, introducing HbA1c into the model short-term deterioration could well have protein turnover themselves may be asso- in the Cox regression analysis had no as- fl P introduced a gap between HbA1c and ciated with mortality and may in uence sociation with all-cause mortality ( = fructosamine, which would have trans- fructosamine, the associations of G-gap 0.082) and did not alter the significant lated into a G-gap. to mortality are statistically independent association of G-gap with mortality. To counter this anxiety, it should be of that as far as we can determine. In conclusion, evidence is mounting pointed out that in general, as confirmed Thus, it seems that fructosamine is an around the G-gap but significant caveats in this study, many have shown a good acceptable measure of glycemia attain- remain, and it may yet turn out to be a relationship between HbA1c and fructos- ment but it cannot be considered a gold spurious phenomenon, especially since amine (1,2,25). Fructosamine is known standard measure. In this regard, it would no defining mechanism is as yet pro- to be well associated with preceding be the validation of the G-gap by blood posed. We do now know that it appears blood glucose levels (38). A concern re- glucose that would best reflect the de- to vary with certain demographic charac- lating to the possible association of fruc- flection of HbA1c glycation. A study of the teristics, it is consistent in direction over tosamine levels with proteinuria has been G-gap and the HGI has confirmed that the time, and, when positive, it is associated published to be significant (13). In our two indices are highly correlated and con- with the key diabetes vascular complica- own regression analysis of fructosamine sistent (40). tions in a manner coherent with one key with multiple relevant factors, we show It is important to stress that the G-gap underlying pathophysiological mecha- that, all in all, they account for no more is not truly independent of HbA1c since nism, protein glycation. We now further than 20% of the variance in fructosamine, the G-gap is computed as the difference show an unexpected association of com- which is to say 80% of fructosamine is not between a measured and a predicted pletely uncertain etiology with mortality in any association with any known influ- HbA1c, and so independence is impos- with both a negative and a positive G-gap encing factor. Among that medley of as- sible. In that regard, any association of inaU-shapedrelationshipwithnoasso- sociated factors as presented, UACR was G-gap with outcomes such as mortality ciated effect of the prevailing HbA1c. the last entered variable being the statisti- will always be difficult to dissect away Our paper confirms the reported cally weakest independent association, from an association with glycemia. How- G-gap association with retinopathy and with an r2 progression of 0.002 thus rep- ever, it would not be expected that a sin- nephropathy, and the findings for demo- resenting only 0.2% of the accountable gle point HbA1c would have any casual graphics, macrovascular disease, and variance of fructosamine. In direct bivariate bearing on mortality. Furthermore, in mortality are previously unreported. care.diabetesjournals.org DIABETES CARE 5 Glycation gap and mortality in diabetes

hematologically normal people is suffi- cient to alter HbA1c. Blood 2008;112: 4284–4291 8. Khera PK, Joiner CH, Carruthers A, et al. Evidence for interindividual heterogene- ity in the glucose gradient across the hu- man red blood cell membrane and its relationship to hemoglobin glycation. Diabetes 2008;57:2445–2452 9. Soranzo N, Sanna S, Wheeler E, et al. Common variants at 10 genomic loci in- fluence hemoglobin A1c levels via glyce- mic and nonglycemic pathways. Diabetes 2010;59:3229–3239 10. Delpierrre G, Vertommen D, Communi D, Rider MH, Van Schaftingen E. Identifica- tion of fructosamine residues deglycated by fructosamine-3-kinase in human he- moglobin. J Biol Chem 2004;279:27613– 27620 11. Mohas M, Kisfali P, Baricza E, et al. A polymorphism within the fructosamine- 3-kinase gene is associated with HbA1c levels and the onset of type 2 diabetes mellitus. Exp Clin Endocrinol Diabetes 2010;118:209–212 12. Delpierre G, Veiga-da-Cunha M, Vertommen D, Buysschaert M, Van Schaftingen E. Variability in erythrocyte fructosamine 3-kinase activity in humans correlates with polymorphisms in the FN3K gene and impacts on haemoglobin glycation at specific sites. Diabetes Metab 2006;32: 31–39 13. Chan JC, Yeung VT, Cheung CK, d fi Figure 2 Cox regression analysis for cumulative survival from time of rst testing of the whole Swaminathan R, Cockram CS. The inter- cohort categorized according to G-gap status (for positive G-gap, n = 651; neutral G-gap, n = relationships between albuminuria, plasma 1,945; and negative G-gap, n =586). albumin concentration and indices of gly- caemic control in non-insulin-dependent diabetes mellitus. Clin Chim Acta 1992; 210:179–185 d 2. Cohen RM, Holmes YR, Chenier TC, Acknowledgments This study was sup- Joiner CH. Discordance between HbA 14. Perry HM 3rd, Morley JE, Miller DK. Se- ported by a grant from the South Staffordshire 1c rum fructosamine and obesity. Diabetes and fructosamine: evidence for a glyco- – Medical Foundation Trust, U.K. sylation gap and its relation to diabetic Care 1999;22:1596 1597 No potential conflicts of interest relevant to nephropathy. Diabetes Care 2003;26: 15. Sacks DB, Nathan DM, Lachin JM. Gaps in this article were reported. the glycation gap hypothesis. Clin Chem 163–167 A.U.N. and B.M.S. researched the data, 2011;57:150–152 3. Hempe JM, Gomez R, McCarter RJ Jr, performed the statistical analysis, and wrote 16. Retinopathy grading recommendations Chalew SA. High and low hemoglobin the draft of the manuscript. A.M.N. and P.B. by the English National Screening Pro- glycation phenotypes in : supported the statistical analysis. All authors gramme for Diabetic Retinopathy (ENSPDR). participated in data interpretation and com- a challenge for interpretation of glycemic Available from http://www.retinalscreening. mented on the draft of the manuscript and control. J Diabetes Complications 2002; – nhs.uk/pages/). Accessed 27 September approved the final version. A.U.N. is the 16:313 320 2011 guarantor of this work and, as such, had full 4. Hudson PR, Child DF, Jones H, Williams 17. Marshall SM. Screening for micro- access to all the data in the study and takes CP. Differences in rates of glycation (gly- albuminuria: which measurement? Diabet fi responsibility for the integrity of the data and cation index) may signi cantly affect in- Med 1991;8:706–711 the accuracy of the data analysis. dividual HbA1c results in type 1 diabetes. 18. Schleicher ED, Mayer R, Wagner EM, – The authors acknowledge help kindly pro- Ann Clin Biochem 1999;36:451 459 Gerbitz KD. Is serum fructosamine assay vided by Dr. Alex Wright (University of Bir- 5. Gould BJ, Davie SJ, Yudkin JS. In- specific for determination of glycated se- mingham, Birmingham, U.K.) toward editing vestigation of the mechanism underlying rum protein? Clin Chem 1988;34:320–323 the final draft prior to its submission. the variability of glycated haemoglobin in 19. Macdonald DR, Hanson AM, Holland MR, non-diabetic subjects not related to gly- Singh BM. Clinical impact of variability – caemia. Clin Chim Acta 1997;260:49 64 in HbA1c as assessed by simultaneously References 6. Dagogo-Jack S. Pitfalls in the use of HbA1c measuring fructosamine and use of error 1. Nayak AU, Holland MR, Macdonald DR, as a diagnostic test: the ethnic conundrum grid analysis. Ann Clin Biochem 2008;45: Nevill A, Singh BM. Evidence for consis- Nat Rev Endocrinol 2010;6:589–593 421–425 tency of the glycation gap in diabetes. 7. Cohen RM, Franco RS, Khera PK, et al. 20. Cohen RM, Snieder H, Lindsell CJ, et al. Diabetes Care 2011;34:1712–1716 Red cell life span heterogeneity in Evidence for independent heritability of

6 DIABETES CARE care.diabetesjournals.org Nayak and Associates

the glycation gap (glycosylation gap) and risk of cause-specific death. N Engl J effects of intensive glucose lowering on fraction of HbA1c in nondiabetic twins. Med 2011;364:829–841 cardiovascular outcomes. N Engl J Med Diabetes Care 2006;29:1739–1743 28. Holman RR, Paul SK, Bethel MA, 2011;364:818–828 21. Brownlee M. Biochemistry and molecular Matthews DR, Neil HAW. 10-year follow-up 36. Currie CJ, Peters JR, Tynan A, et al. Sur- cell biology of diabetic complications. of intensive glucose control in type 2 di- vival as a function of HbA(1c)inpeople Nature 2001;414:813–820 abetes. N Engl J Med 2008;359:1577–1589 with type 2 diabetes: a retrospective co- 22. UK Prospective Diabetes Study (UKPDS) 29. Ray KK, Seshasai SR, Wijesuriya S, et al. hort study. Lancet 2010;375:481–489 Group. Intensive blood-glucose control Effect of intensive control of glucose on 37. Lachin JM, Genuth S, Nathan DM, with sulphonylureas or insulin compared cardiovascular outcomes and death in Rutledge BN. The hemoglobin glycation with conventional treatment and risk of patients with diabetes mellitus: a meta- index is not an independent predictor of complications in patients with type 2 di- analysis of randomised controlled trials. the risk of microvascular complications in abetes (UKPDS 33). Lancet 1998;352: Lancet 2009;373:1765–1772 the Diabetes Control and Complications 837–853 30. Boussageon R, Bejan-Angoulvant T, Trial. Diabetes 2007;56:1913–1921 23. The Diabetes Control and Complications Saadatian-Elahi M, et al. Effect of intensive 38. Baker JR, Metcalf PA, Holdaway IM, Trial Research Group. The effect of in- glucose lowering treatment on all cause Johnson RN. Serum fructosamine con- tensive treatment of diabetes on the de- mortality, cardiovascular death, and mi- centration as measure of blood glucose velopment and progression of long-term crovascular events in type 2 diabetes: control in type I (insulin dependent) di- complications in insulin-dependent di- meta-analysis of randomised controlled abetes mellitus. Br Med J (Clin Res Ed) abetes mellitus. N Engl J Med 1993;329: trials. BMJ 2011;343:d4169 1985;290:352–355 977–986 31. Kelly TN, Bazzano LA, Fonseca VA, 39. Vos FE, Schollum JB, Coulter CV, Manning 24. Goh SY, Cooper ME. Clinical review: the Thethi TK, Reynolds K, He J. Systematic PJ, Duffull SB, Walker RJ. Assessment of role of advanced glycation end products review: glucose control and cardiovascu- markers of glycaemic control in diabetic in progression and complications of di- lar disease in type 2 diabetes. Ann Intern patients with chronic kidney disease using abetes. J Clin Endocrinol Metab 2008;93: Med 2009;151:394–403 continuous glucose monitoring. Nephrol- 1143–1152 32. Brown A, Reynolds LR, Bruemmer D. In- ogy (Carlton) 2012;17:182–188 25. Rodrıguez-Segade S, Rodrıguez J, tensive glycemic control and cardiovas- 40. Chalew SA, McCarter RJ, Thomas J, Cabezas-Agricola JM, Casanueva FF, cular disease: an update. Nat Rev Cardiol Thomson JL, Hempe JM. A comparison of Camiña F. Progression of nephropathy in 2010;7:369–375 the glycosylation gap and hemoglobin type 2 diabetes: the glycation gap is a 33. Duckworth W, Abraira C, Moritz T, et al.; glycation index in patients with diabetes. J significant predictor after adjustment for VADT Investigators. Glucose control and Diabetes Complications 2005;19:218– glycohemoglobin (HbA1c). Clin Chem vascular complications in veterans with 222 2011;57:264–271 type 2 diabetes. N Engl J Med 2009;360: 41. Zoungas S, Chalmers J, Ninomiya T, et al.; 26. McCarter RJ, Hempe JM, Gomez R, 129–139 ADVANCE Collaborative Group. Associ- Chalew SA. Biological variation in HbA1c 34. ADVANCE Collaborative Group, Patel A, ation of HbA1c levels with vascular com- predicts risk of retinopathy and ne- MacMahon S, Chalmers J, et al. Intensive plications and death in patients with type phropathy in type 1 diabetes. Diabetes blood glucose control and vascular out- 2 diabetes: evidence of glycaemic thresh- Care 2004;27:1259–1264 comes in patients with type 2 diabetes. N olds. Diabetologia 2012;55:636–643 – 27. The Emerging Risk Factors Collaboration, Engl J Med 2008;358:2560 2572 42. Rutter MK. Low HbA1c and mortality: Seshasai SR, Kaptoge S, Thompson A, 35. ACCORD Study Group; Gerstein HC, causation and confounding. Diabetologia et al. Diabetes mellitus, fasting glucose, Miller ME, Genuth S, et al. Long-term 2012;55:2307–2311

care.diabetesjournals.org DIABETES CARE 7