BACKGROUND: Monitoring of glycemic control with hemoglobin A1c (A1c) in hemodialysis patients may be compromised by anemia and erythropoietin therapy. Glycated albumin (GA) is an alternative measure of glycemic control but is not commonly used because of insufficient evidence of association to clinical outcomes. We tested whether GA measurements were associated with mortality in hemodialysis patients with diabetes mellitus.
METHODS: The German Diabetes and Dialysis Study (4D) investigated effects of atorvastatin on survival in 1255 patients with diabetes mellitus receiving hemodialysis. We measured GA during months 0, 6, and 12. Cox proportional hazards analysis was used to measure associations between GA and A1c and all-cause mortality.
RESULTS: Patients with high baseline GA (fourth quartile) had a 42% higher 4-year mortality compared to those in the first quartile (HR 1.42; 95% CI, 1.09–1.85, P = 0.009). Repeated measurements of GA during year one also demonstrated that individuals in the top quartile for GA (analyzed as a time-varying covariate) had a 39% higher 4-year mortality (HR 1.39; 95% CI, 1.05–1.85, P = 0.022). The associations between high A1c and mortality using similar analyses were less consistent; mortality in individuals with baseline A1c values in the 3rd quartile was increased compared to 1st quartile (HR 1.36; 95% CI, 1.04–1.77, P = 0.023), but risk was not significantly increased in the 2nd or 4th quartiles, and there was a less consistent association between time-varying A1c values and mortality.
CONCLUSIONS: High GA measurements are consistently associated with increased mortality in patients with diabetes mellitus on hemodialysis.
Currently, the standard method of monitoring glycemic control in patients with diabetes mellitus is by periodic measurement of percentage of hemoglobin A1c in patients' blood (referred to as A1c in this article).4 A1c is hemoglobin that has been modified by glucose, and the proportion of hemoglobin carrying this modification is proportional to patients' recent time-averaged blood glucose concentrations. Not only is A1c a biomarker of time-averaged glucose, but high A1c values have been shown [by the German Diabetes and Dialysis Study (4D) study group as well as by others] to be associated with increased mortality in patients with diabetes mellitus on hemodialysis (HD) (1–3). Despite these associations, some authors have questioned the validity of A1c measurements in patients with diabetic renal disease because of the association between kidney disease and anemia. Kidney failure patients are often anemic and dependent on erythropoietin therapy, resulting in unpredictable changes in erythrocyte life span; these changes in red cell kinetics in turn result in proportional changes in A1c that are independent of glucose (4–8).
Because of the confounding effects of kidney disease on the relationship between mean glucose and A1c values, it has been suggested that the percentage of serum albumin that is glycated (referred to as GA in this article) may be a reasonable alternative marker for glycemic control in patients on HD (9). Both A1c and GA are spontaneously glycated over their circulating lifespans, and thus both are biomarkers for time-averaged blood glucose concentrations. In patients with diabetes mellitus and normal red cell kinetics, A1c values provide an assessment of mean blood glucose over approximately the past 100 days; GA values, in contrast, represent mean blood glucose over approximately the past 40 days (10). In contrast to A1c, however, GA values are not influenced by renal anemia or erythropoietin therapy, and thus may provide a more accurate assessment of glycemic control in this patient population and in other patients with disorders affecting erythrocyte life span (4, 7, 9, 11–18). In patients without end-stage renal disease (ESRD), there is growing evidence of the association between GA and mortality, similar to the association observed for A1c (19). Multiple clinical studies have observed that GA in patients on HD tends to be more strongly correlated with mean glucose concentrations compared to A1c. (4, 15–17, 20). In addition, there is also mounting evidence that GA is a better predictor of adverse outcomes in patients with diabetes mellitus on HD. There have been 3 moderate-sized clinical studies demonstrating an association between GA and mortality in HD patients (21–23). Furthermore, studies have shown that in comparison to A1c, GA is a better predictor of cardiovascular hospitalizations, length of stay, and risk of contrast-induced acute kidney injury in patients with severe kidney disease (18, 24). Despite the mounting data in support of GA as a biomarker of glycemic control and risk in patients on dialysis, these studies have not yet been deemed sufficient to warrant recommendations for its use in clinical practice (9, 21). In the present study, we sought to test whether the mortality risk associated with high GA could be corroborated in a larger cohort of patients with diabetes mellitus receiving HD. Using samples and clinical study data collected from 1053 study participants of the German 4D study (a clinical trial that tested the survival benefits of atorvastatin therapy), we measured GA and A1c in study participants' blood samples collected during study months 0, 6, and 12. We analyzed the results for comparative associations between GA and A1c with future risk of death in 4D study participants over a median of 4 years of follow-up (25).
Materials and Methods
This was a retrospective study of previously collected frozen serum samples from the 4D clinical trial. The 4D trial was a prospective randomized controlled trial that included a consecutive series of 1255 ESRD patients on maintenance HD with diabetes mellitus enrolled from 178 dialysis units throughout Germany (25). These patients ranged from ages 18–80 and were on dialysis for <2 years. After 4 weeks, patients were randomly assigned to atorvastatin 20 mg daily or placebo daily. All study participants who had a complete set of frozen samples collected at 0, 6, and 12 months were included (n = 1053). For a diagrammatic description of the study design, see Fig. S1 in the Data Supplement that accompanies the online version of this article at http://www.clinchem.org/content/vol63/issue2. The study was conducted in accordance with ethical standards and approved by the institutional medical ethical committee, and all patients gave written informed consent before inclusion (25).
The patients were seen by study staff 3 times before randomization and then every 6 months afterward until date of death, censoring, or the end of the study in March, 2014. The primary endpoint of this original study was a composite of cardiovascular events including cardiac death (sudden death, fatal myocardial infarction, death caused by congestive heart failure, death resulting from coronary heart disease within 28 days after an intervention), fatal or nonfatal stroke and nonfatal myocardial infarction (MI), whichever occurred first. All-cause mortality was a secondary endpoint. The current ancillary study focused on all-cause mortality and included analysis of blood samples collected at study initiation and during the 6 months and 12 months follow-up visits. Age, sex, and smoking history were obtained through patient interviews with smoking status categorized as current, former, or never. Comorbidities and past medical history were reported by the patients' nephrologists. Coronary artery disease was defined by history of MI, coronary artery bypass graft surgery, percutaneous coronary intervention, or documentation of coronary artery disease by angiography. Blood pressure was measured while sitting down.
A1c and all other laboratory analyses (except for GA measurements) were performed at the time of collection between March, 1998 and October, 2002 by the Department of Clinical Chemistry, University of Freiburg, Freiburg, Germany as previously published (1). A1c was measured using automated cation exchange HPLC (Tosoh HLC-723 GHb V, A1c2.2; reference interval, 3.4%–6.0%) in a clinical laboratory certified by the College of American Pathologists. At the time of the original 4D study, this assay was not NGSP certified, as NGSP certification was not yet available (26). The CV of A1c measurements was <5%. Additional details describing the measurement and collection of clinical covariates have been previously published (25, 27).
Measurement of the proportional concentrations of glycated and nonglycated albumin for calculating % glycated albumin (GA) were performed using an LC-MS/MS assay method similar to previously described methods (28, 29). Measurements were performed on deidentified 4D patient serum samples stored in −80 °C freezers and were thawed to room temperature immediately before analysis. The proportion of albumin glycated on the second lysine within the proteolytic peptide RQIKKQTALVE was assayed by digesting serum with Glu-C protease followed by measurement by LC-MS/MS using methods similar to those previously described (29). This site has been shown to be the most common site of glycation on human albumin (28). For digestion, 2.5 μL of serum was diluted with 50 μL of digestion buffer; the digestion buffer contained 25 mmol/L Tris-HCl pH 7.4, 10 mmol/L dithiothreitol, 9.5 μmol/L isotopic peptide internal calibrator [purified synthetic peptide, amino acid sequence RQIKKQTALVE with isotopic l-leucine (13C6, 15 N) incorporated at position 9, synthetized by Primm Biotech], and 0.1 g/L glutamyl endoproteinase (Worthington Biochemical). Samples and digestion buffer were incubated at 37 °C for 3 h and frozen at −80 °C until analysis. The digested samples were then thawed and analyzed using HPLC/tandem mass spectrometry to determine the percentage of albumin that was glycated. The assay method was calibrated using commercial calibrators for glycated albumin from Asahi Kasei Pharma (see additional details in the online Supplemental Materials). LC-MS/MS measurements of glycated albumin were validated by comparison to measurements using the Lucica GA-L assay (Asahi Kasei Pharma). Analytical validation studies used discarded serum samples from 100 random (consecutively collected) patients from Beth Israel Deaconess Medical Center who had also had hemoglobin A1c measured for clinical purposes; these samples were deidentified at the time of collection before research testing, and all procedures used conformed to a protocol approved by the human subjects research institutional review board. Additional assay method details and validation studies are included in the online Supplemental Materials.
Participant characteristics data were tested for normality with the Shapiro–Wilk test; normally distributed variables were expressed as mean (SD), and other variables were expressed as median and interquartile range, or percentage of total, as appropriate. Survival analyses were performed with the total follow-up time available (median follow-up 4 years), defined as longer-term outcome analyses. The associations between all-cause mortality and baseline measurements of GA and A1c were evaluated using Cox proportional hazards (PH) analysis. Study participants were stratified into quartiles according to their GA or A1c, and the risks associated with increased GA and A1c were estimated by calculating the unadjusted hazard ratios (HRs) and their corresponding 95% CIs for each quartile compared to participants in the bottom quartile as reference group. Multivariable-adjusted HR values were also calculated after adjusting for confounders including age, sex, body mass index, diabetes duration, atorvastatin treatment, history of coronary artery disease, history of congestive heart failure, systolic blood pressure, BMI and blood concentrations of calcium, phosphate, hemoglobin, low density lipoprotein cholesterol, triglycerides, and C-reactive protein. Lastly, to determine whether longitudinal measurements of GA and A1c were also associated with mortality risk, we conducted a standard time-varying Cox-regression approach, modelling A1c and GA as time-varying predictor variables, respectively (30). We fitted categorical models including A1c and GA as time-varying variables with categories defined by quartiles. Therefore, participants were stratified into quartiles by their GA and A1c measurements at 0, 6, and 12 months, and these variables were included in the Cox PH models as time-varying covariates to analyze risk associated with different GA and A1c quartiles. All P values reported are 2 sided, and a P value <0.05 was considered statistically significant. The P value testing the significance of the difference between GA vs A1c correlation coefficients was determined using previously described methods (31).
4D study patients were recruited between March 1998 and October 2002. In our analysis, 1053 of the 1255 originally recruited patients had samples from 0, 6, and 12 months available for this study. The mean follow-up period was 3.96 years on atorvastatin and 3.91 years on placebo. To analyze risk associated with increasing GA, study participants were stratified into quartiles on the basis of their baseline GA values; baseline clinical characteristics of the complete cohort as well as GA-stratified subgroups are shown in Table 1. Baseline percent GA in study participants was 18 +/− 5% (mean +/− SD). Baseline clinical characteristics of participants stratified by A1c quartile are shown in online Supplemental Table S1.
CHARACTERISTICS OF LC-MS/MS ASSAY FOR GLYCATED ALBUMIN
LC-MS/MS GA measurements agreed well with measurements using a commercially available clinical assay, and had a coefficient of variation of 1.9%. Additional experiments validating this LC-MS/MS assay method are presented in the online Supplemental Materials.
Multiple studies have shown that A1c and GA are strongly correlated in patients without renal disease (14). To corroborate this previous finding and confirm the association between GA and glycemic control using our newly developed LC-MS/MS assay, we measured the correlation between GA and A1c in simultaneously collected serum and whole blood samples from 100 random hospital patients with normal renal function [defined by the Modification of Diet in Renal Disease estimated glomerular filtration rate values >90 mL · min−1 · (1.73 m2)−1]. GA values were tightly correlated with A1c values in nonuremic individuals (R2 = 0.738, P <0.0001, Fig. 1). In contrast, although GA values also correlated with A1c values among 4D study participants on HD (R2 = 0.457, P <0.0001), the strength of the correlation was decreased and more diffuse in comparison to nonuremic individuals (P value for difference between correlation coefficients <0.0001). Furthermore, it has been previously shown that the ratio of GA relative to A1c is higher among patients with decreased renal function, possibly due to the effects of renal disease on red cell dynamics (15). When we calculated the GA:A1c ratios of our study participants, we also found mean GA:A1c ratios to be the higher among 4D participants compared to nonuremic patients (GA:A1c ratio of 2.74 ± 0.56 vs 2.21 ± 0.52, P < 0.0001). Together, these findings suggest that patient variables other than time-averaged glucose may be affecting A1c values in the participants with end-stage kidney disease, as has been previously reported (4, 5, 7, 15, 32).
GA AND A1c AND FUTURE RISK OF ALL-CAUSE MORTALITY
During the follow-up period, 499 of 1053 patients expired. To contrast the risk of death associated with both GA and A1c using comparable analyses, multivariable-adjusted Cox-proportional hazard analysis was used to relate the longer-term risk of death associated with increases in study participants' GA and A1c values (median follow-up period 4 years). As shown in Table 2, when we analyzed the risk associated with GA and A1c measurements in participants stratified into quartiles, we observed a significant increase in mortality risk in participants with glycated albumin in the top quartile compared to those in the bottom quartile, who served as the reference group [HR = 1.42 (95% CI, 1.09–1.85, P = 0.009); HR adjusted for other mortality risk factors = 1.32 (95% CI, 1.01–1.73, P = 0.04)]. When we performed the same analysis on the association between A1c values and mortality, we found that unadjusted risk in participants in the 3rd A1c quartile was increased compared to those in the lowest quartile (reference group) [HR = 1.44 (95% CI, 1.11–1.86, P = 0.0.005); HR = 1.27 (95% CI, 0.98–1.64, P = 0.0.067); HR adjusted for other risk factors = 1.36 (95% CI, 1.04–1.77, P = 0.024)]. Surprisingly, however, the estimated survival risk associated with participants in the top baseline A1c quartile was not statistically significant, suggesting a discontinuous relationship between higher A1c values and mortality.
In addition to our interest in the association between isolated measurements of baseline GA and subsequent 4-year survival, we were also interested in whether multiple longitudinal measurements of GA were associated with long-term mortality using statistical methods for analysis of time-varying risk factors. GA was measured in 4D study participants during months 0, 6, and 12 of the study. As shown in Table 3, time-dependent analyses again found a 39% increase in 4-year mortality in participants with repeated GA values in the top quartile [HR adjusted for age and sex = 1.39 (95% CI, 1.05–1.85), P = 0.022]. In contrast, when we performed the same analysis on repeat measurements of A1c, we found no significant risk in participants in the top quartile for time-varying A1c [HR adjusted for age and sex = 0.87 (95% CI, 0.70 – 1.08, P = 0.196)]. Instead, we found that individuals in the 2nd and 3rd quartiles for time-varying A1c had decreased mortality compared to those in the bottom quartile. These results suggest there is a U-shaped association between A1c values and mortality, where individuals are at risk when their A1c values are very high or very low, as has been previously reported (3, 33, 34).
Lastly, to graphically evaluate the associations between GA, A1c, and mortality risk, we plotted Kaplan–Meier survival curves stratified by baseline GA and A1c quartiles. As illustrated in Fig. 2, we were able to clearly see the difference in mortality of individuals in the top quartile for baseline glycated albumin compared to those in the lower quartiles.
In this study of hemodialysis patients with type 2 diabetes mellitus, we observed that high concentrations of glycated albumin measured singly or repeatedly during the first year of a 4-year study were consistently associated with future risk of death due to all causes, even after adjusting for other significant risk factors. In contrast, we found an inconsistent association between higher A1c values and mortality risk, suggesting a more complex relationship between A1c and survival. Although these analyses do not constitute a true head-to-head comparison of the predictive value of GA vs A1c measurements, and cannot be used to determine prognostic superiority in a clinical setting, our findings do support the hypothesis that poor glycemic control, as reflected by GA values in the top quartile, is a significant survival risk factor in this patient population.
DISCREPANCY BETWEEN A1c AND GA
There have been multiple studies demonstrating the discord between A1c and GA in patients with renal insufficiency. Multiple observational studies have found that HD patients have lower A1c levels despite higher random glucose and GA levels (4–7). These studies also found that the ratio of GA to A1c was increased in HD patients relative to patients without renal insufficiency. High ratios of GA relative to A1c were also noted in patients on peritoneal dialysis and those with stage V predialysis chronic kidney disease (5, 6). Furthermore, these studies observed that A1c correlated positively with hemoglobin and negatively with erythropoietin stimulating agents (4, 5, 7, 32). In the present study, we observed similar trends, finding that the ratio of GA compared to A1c was higher in HD patients compared to individuals with normal renal function. Furthermore, the correlation between GA and A1c was significantly more diffuse among patients on HD compared to the correlation found in nonuremic patients. Together, our new findings combined with these previous cited studies suggest that hematologic factors independent of glycemic control alter A1c values in a manner that is independent of mean glucose concentrations.
GA AND OUTCOMES IN DIALYSIS PATIENTS
Compared to A1c, there have been few large clinical studies evaluating outcomes associated with glycemic control measured by GA in patients on HD. One prospective follow-up study by Fukuoka, et al. followed 98 patients with diabetic nephropathy and ESRD for 27 months after entry (22). After adjustment for age, sex, total cholesterol, CRP, protein, and albumin, high mean GA was a significant predictor of cardiovascular death (HR of 1.042 per 1.0% increment of GA, 95% CI, 1.014–1.070, P <0.05); in contrast, these investigators did not observe a significant association between A1c and cardiovascular mortality. In a similar study of patients with diabetes mellitus on HD (n = 401) and peritoneal dialysis (n = 43) by Freedman et al., multivariate analysis demonstrated that GA was significantly associated with mortality while both A1c and random glucose were not (21). To our knowledge, this study is the largest published study to date showing an association between GA and mortality in ESRD patients, and our results are in general agreement with prior studies testing the association between GA and mortality in patients with ESRD (21, 22). There are additional ongoing studies, such as the Glycemic Indices in Dialysis Evaluation (GLIDE) study that are continuing to evaluate the association of GA with outcomes in dialysis patients (32).
It is important to note that the focus of this report is intended to test the relationship between GA and survival risk in patients on HD, and our analyses do not constitute a true head-to-head comparison of the prognostic values of GA compared to A1c. However, it is relevant to discuss the importance of the associations that we observed between A1c and mortality in this cohort. Although we found that individuals with baseline A1c values in the 3rd quartile had increased mortality compared to the lowest quartile, the risk paradoxically decreased in individuals in the 4th quartile. Furthermore, our analysis of time-varying A1c measurements suggested that individuals in the highest and lowest quartiles were actually at increased risk compared to the middle 2 A1c quartiles. Although these results may at first be difficult to reconcile, other previous studies associating A1c with survival in patients with renal insufficiency have shown similar results. A number of studies have observed inconsistent and sometimes paradoxical inverse associations between A1c and survival in this patient population (21, 22, 35, 36). The Dialysis Outcomes and Practice Patterns Study, which included 9201 patients with diabetes mellitus on HD and used longitudinal analysis similar to the time-dependent analysis used in this report, observed a U-shaped relationship between time-varying A1c measurements and mortality risk, finding that patients with either very high or very low A1c values were at significant risk of death compared to patients in the middle (3). This U-shaped association between A1c and mortality has also been corroborated by several other large studies (2, 3, 33, 34).
It has been suggested that the apparently discontinuous relationship between A1c and mortality may be due to the fact that very low A1c values in patients with ESRD may be the consequences of anemia or malnutrition and abnormal red blood cell dynamics (3). The poor correlation between GA and A1c values observed in patients with ESRD, together with mortality risk associated with both very high and very low A1c values supports the hypothesis that patients on HD may have factors other than glycemic control that influence their A1c values, confounding the relationship between A1c measurements and outcomes. Using a more suitable marker for evaluation of glycemic control, such as GA, may help us determine the actual effect of glycemic control on morbidity and mortality in this population.
This study has several limitations. The original 4D clinical trial was designed to test benefits of atorvastatin therapy in patients with type 2 diabetes mellitus on HD, and not specifically designed to test benefits of glycemic control. Prospective randomized controlled studies specifically designed to test the effects of targeting glycemic control based upon GA measurements are needed to prove its clinical benefits, to define GA thresholds for optimal glycemic therapeutic targets, and to compare this alternative therapeutic target to A1c. Secondly, this study did not include any patients with diabetes mellitus and chronic kidney disease who were not on HD, nor patients receiving peritoneal dialysis. Additional studies are needed to determine whether these patient populations, who also suffer from varying degrees of anemia and altered A1c values, may also benefit from GA monitoring. Lastly, although the automated HPLC assay used to measure A1c for this study was a widely used commercial method, it was not NGSP certified or otherwise DCCT-harmonized at the time, and thus it is possible that the association between A1c and mortality may have been affected by the analytical performance of this assay method.
There are several issues pertaining to measurement of glycated albumin in dialysis patients that should be mentioned. Measurements of GA by non–mass spectrometric assays may be affected by lipemia, hyperbilirubinemia, and hemolysis. GA measurements also have been shown to be affected by hyperuricemia, uremia, high doses of aspirin, hypoproteinemia, age, albuminuria, cirrhosis, thyroid dysfunction, and smoking. Other studies have shown that glycated albumin concentrations are inversely influenced by BMI and fat mass. Lastly, different assays for GA display varying reference intervals, indicating that there are still issues of assay standardization and harmonization remaining that need to be resolved (9).
In summary, there is growing evidence that monitoring and optimization of glycemic control in patients with diabetic kidney disease is important in reducing their risks of cardiovascular disease and death. To do this, we need to make sure that the most accurate marker of glycemic control is used. This study provides additional evidence supporting the association between GA and survival outcomes, and thus indirectly supports the application of GA as a test for monitoring glycemic control in patients with diabetes mellitus on HD in the research and clinical setting.
↵† C.W. Chen and C. Drechsler contributed equally to the work, and both should be considered as first authors.
↵4 Nonstandard abbreviations:
- whole blood % hemoglobin A1c;
- the German Diabetes and Dialysis Study;
- serum % glycated albumin;
- end-stage renal disease;
- myocardial infarction;
- proportional hazards;
- hazard ratio.
(see editorial on page 447)
Author Contributions: All authors confirmed they have contributed to the intellectual content of this paper and have met the following 3 requirements: (a) significant contributions to the conception and design, acquisition of data, or analysis and interpretation of data; (b) drafting or revising the article for intellectual content; and (c) final approval of the published article.
Authors' Disclosures or Potential Conflicts of Interest: Upon manuscript submission, all authors completed the author disclosure form. Disclosures and/or potential conflicts of interest:
Employment or Leadership: None declared.
Consultant or Advisory Role: None declared.
Stock Ownership: None declared.
Honoraria: None declared.
Research Funding: C.W. Chen, National Institutes of Health award T32 DK007199-38; C. Drechsler and C. Wanner, German Federal Ministry for Education and Research (BMBF01EO1004) and University Hospital Wuerzburg Gundausstattung grant program; A.H. Berg, National Institutes of Health award K08 HL121801, American Diabetes Association Innovation award 1-15-IN-02.
Expert Testimony: None declared.
Patents: None declared.
Role of Sponsor: The funding organizations played no role in the design of study, choice of enrolled patients, review and interpretation of data, and final approval of manuscript.
- Received for publication April 3, 2016.
- Accepted for publication August 17, 2016.
- © 2016 American Association for Clinical Chemistry