The Impact of Carbamylation and Anemia on HbA_1c’s Association With Renal Outcomes in Patients With Diabetes and Chronic Kidney Disease

Diabetes and chronic kidney disease (CKD) are two major public health problems that commonly coexist. Approximately 30% of patients with type 1 diabetes and 40% with type 2 diabetes develop CKD (1). Indeed, diabetes is the leading cause of end-stage kidney disease (ESKD) worldwide, accounting for 50% of all cases (2). While recent breakthroughs, including sodium–glucose cotransporter 2 inhibitors, glucagon-like peptide 1 receptor agonists, and nonsteroidal mineralocorticoid antagonists, provide encouraging kidney protection (3–5), the residual risk of CKD progression and ESKD remains unacceptably high in this population (6). Moreover, although the prevalence and rates of many other diabetic-related complications have declined substantially, the number of patients with diabetes suffering from ESKD is rising (7,8).

Currently, the standard method of monitoring glycemic control in patients with diabetes is by measuring glycated hemoglobin (HbA_1c) (9,10). Poor glycemic control detected by high HbA_1c is one of the most established risk factors for developing diabetic kidney disease (1). Nevertheless, the validity of HbA_1c in patients with CKD has been questioned, and the target HbA_1c level in the population with CKD is controversial (11,12). An initial concern was glycated hemoglobin assay interference from another posttranslational protein modification, carbamylation (13–16). Carbamylation is characterized by the nonenzymatic, spontaneous binding of urea-derived isocyanate to free amino groups on proteins, which can increase in states of high urea burden, such as CKD (17). Assay interference from carbamylation was thought to be a major problem falsely elevating HbA_1c readings (13). While modern-day HbA_1c assays do not appear susceptible to this interference (16), new concerns have since emerged from animal studies suggesting that carbamylation reactions could lead to falsely low HbA_1c values by acting on the same hemoglobin amino groups, thus competitively lowering glycation levels without actual changes in mean glucose levels (18,19). An additional well-studied concern of HbA_1c in patients with CKD is the influence of CKD-related anemia and the lifespans of red cells with subsequent alterations in hemoglobin glycation percentages (14). The degree of influence of carbamylation and anemia on HbA_1c’s predictive abilities for clinical outcomes in CKD remains poorly understood, hampering the interpretation of one of the most widely used laboratory tests in diabetes care. Therefore, we aimed to investigate the association and interaction between HbA_1c and either carbamylation or anemia with CKD progression in a large prospective cohort study of individuals with diabetes and CKD.

The Chronic Renal Insufficiency Cohort (CRIC) is a multicenter prospective observational cohort of patients with mild to severe CKD (defined as estimated glomerular filtration rate [eGFR] of 20 to 70 mL/min/1.73 m² at the screening visit), followed longitudinally to study risk factors for CKD progression, cardiovascular disease (CVD), and mortality. A total of 3,939 patients aged 21 to 74 years were enrolled across 7 U.S. clinical sites from June 2003 to September 2008. The main exclusion criteria for the CRIC study included pregnancy, New York Heart Association class III–IV heart failure, HIV, cirrhosis, multiple myeloma, renal cancer, recent chemotherapy or immunosuppression, polycystic kidney disease, organ transplantation, or previously receiving dialysis for >1 month. Further details of the CRIC study have been previously published (20,21). Participants provided informed consent at the time of enrollment. The CRIC study protocol was approved by the institutional review boards from each participating clinical site and is in accordance with the Declaration of Helsinki.

The year 1 (Y1) CRIC visit (1 year after initial participant enrollment) was the visit for which stored serum samples were available for measurement of carbamylation biomarkers of interest in our study; therefore, this was considered the baseline visit in the current study. For the current analysis, we included a total of 1,516 participants from CRIC who had diabetes with the measurement of HbA_1c available and were free of ESKD at the Y1 mark. Diabetes was defined by the following criteria: fasting plasma glucose level ≥126 mg/dL, random plasma glucose ≥200 mg/dL, or self-reported use of insulin or other diabetes medications. A flowchart illustrating further details on the selection process of participants is shown in Supplementary Fig. 1.

The primary exposure of interest was baseline HbA_1c. The serum samples were collected at the Y1 visit and stored at −80°C until measurement at the CRIC Central Lab (University of Pennsylvania). HbA_1c was measured using an ion-exchange high-performance liquid chromatography (HPLC) method (Bio-Rad Variant II) for samples collected through February 2006 and a Boronate affinity HPLC method for samples collected subsequently. The assays were calibrated using reference materials obtained from the National Glycohemoglobin Standardization Program (NGSP). Carbamylation has been shown to not interfere with these assays (22,23). HbA_1c had a normal distribution and was analyzed both as a continuous variable and by grouping patients into quartiles, with quartile 2 (Q2; 6.4–7.2% [46–55 mmol/mol]) as the reference group considering the clinical target of HbA_1c in patients with CKD and risks associated with values above or below this range (12).

Carbamylated albumin (C-Alb) was reported as the ratio of millimoles of carbamylated albumin per mole of total albumin (24), measured using HPLC and tandem mass spectrometry after a single thaw over 1 week. C-Alb is a commonly used marker of carbamylation load reported in epidemiological studies (24,25). The intraassay coefficient of variance was 4.2%, and the interassay coefficient of variance was 8.3%. Due to the right-skewed distribution, C-Alb was examined both as quartiles and after natural log transformation. The presence of anemia was defined as a hemoglobin level <11 g/dL.

Demographics, lifestyle behaviors, medical history, and current medications were collected at the Y1 visit. All laboratory values were measured using standardized assays performed on samples from the same visit. The eGFR was calculated according to the CRIC-derived equation (26). Proteinuria was measured from 24-h urine collection. Urinary albumin-to-creatinine ratio was unavailable at the Y1 visit; however, urinary albumin-to-creatinine ratio and 24-h proteinuria were closely correlated, and their associations with outcomes were comparable in CRIC (27). All covariates had <1% missing values, except for 24-h proteinuria (9.5%), LDL (4.2%), and triglyceride (4.0%). Missing covariates were imputed by the mean or median of the existing values as appropriate.

The primary outcome was CKD progression (a composite of incident ESKD requiring chronic dialysis or kidney transplant, or 50% decline in eGFR). The participants were followed every 6 months with telephone visits and annually with in-person clinic visits. ESKD status was obtained through semiannual surveillance with further ascertainment supplemented with cross-linkage with the U.S. Renal Data System. Time to 50% eGFR decline was imputed with the assumption of a linear decrease in eGFR between annual measurements (28). The follow-up was censored at the occurrence of death, withdrawal of consent, loss to follow-up, or end of follow-up.

Baseline characteristics were summarized as counts (percentages), mean (SD), or median (interquartile range). We compared baseline characteristics using χ² for categorical variables, ANOVA for normally distributed variables, and Kruskal-Wallis for nonnormally distributed variables. The distributions of HbA_1c per each C-Alb quartile were visualized by constructing box plots. After a global statistically significant difference of HbA_1c levels among C-Alb quartiles was detected by ANOVA, we conducted post hoc analysis with the Tukey honestly significant difference test for multiple pairwise comparisons. Additionally, a linear regression model was applied to further investigate the relationship between HbA_1c and C-Alb quartiles, adjusting for random glucose levels. The Pearson correlation coefficient between HbA_1c and random glucose levels was also evaluated based on the C-Alb quartiles. Locally weighted scatterplot smoothing lines were also used to further assess the relationship of C-Alb with HbA_1c, eGFR, and hemoglobin.

We performed Cox proportional hazard models to assess the associations between HbA_1c and the risk for CKD progression. Subgroup analyses were performed according to each defined subgroup: C-Alb quartiles and the presence/absence of anemia. We also tested the effect modification of HbA_1c impact by these two factors, C-Alb (as quartiles or after natural log transformation) and anemia (as a binary variable or in continuous hemoglobin levels) with interaction testing (assessed by a likelihood ratio test). The assumption of linearity for continuous measures of HbA_1c was tested based on the −2 log likelihood χ², comparing models with the continuous variable versus quartiles. Proportional hazard assumptions for models were evaluated by Schoenfeld residual plots. An assessment of model assumptions was performed and found to be tenable.

The potential confounding variables were selected based on the literature review and clinical relevance. Model 1 was adjusted for age, sex, and race/ethnicity. Model 2 (main model) included model 1 and added BMI, current smoking status, history of CVD, use of ACE inhibitors (ACEIs) or angiotensin II receptor blockers (ARBs), systolic blood pressure, eGFR, 24-h proteinuria (with natural log transformation), serum albumin, LDL, and triglyceride (with natural log transformation). Given the complex relationship among anemia, carbamylation, and glycation, we also performed additional analysis (model 3) by adding hemoglobin to our main model. We conducted several sensitivity analyses for the main findings. First, we categorized HbA_1c based on the Kidney Disease: Improving Global Outcomes (KDIGO) target for patients with diabetes and CKD into three groups: <6.5% (48 mmol/mol), 6.5–8% (reference group [48–64 mmol/mol]), and >8% (64 mmol/mol). Second, we performed multiple imputations for missing covariates, and no substantial differences were observed. Lastly, we conducted Fine and Gray competing risks regression to account for death as competing events for CKD progression.

All reported P values were two-sided, and a cutoff of 0.05 was used to indicate statistical significance. All analyses were conducted using R Studio Version 4.1.2.

Baseline characteristics of the study participants, both overall and according to the quartiles of HbA_1c, are presented in Table 1. The mean age of participants was 60.4 years, and 43.9% were female. The mean baseline HbA_1c was 7.5% (58 mmol/mol), the median C-Alb was 8.4 mmol/mol, and the mean hemoglobin was 12.2 (1.7) g/dL. Overall, the average eGFR was 38.1 mL/min/1.73 m², 58.7% of participants had CKD G3, and 33.1% had CKD G4/G5. Those with higher HbA_1c were younger and more likely to have a history of CVD, to be using insulin, and to have higher 24-h proteinuria. Baseline characteristics of the study participants according to the quartiles of C-Alb are presented in Supplementary Table 1.

The distribution of HbA_1c levels according to the C-Alb quartiles is shown in Fig. 1. There were differences in HbA_1c levels across the C-Alb quartiles by ANOVA (P < 0.001). Using the Tukey honestly significant difference multiple comparison test, HbA_1c was found to be lower in C-Alb Q2 (mean HbA_1c 7.4% [57 mmol/mol]; P = 0.009), Q3 (mean HbA_1c 7.5% [58 mmol/mol]; P = 0.04), and Q4 (mean HbA_1c 7.1% [54 mmol/mol]; P < 0.001) compared with C-Alb Q1 (mean HbA_1c 7.8% [62 mmol/mol]). Even when adjusting for random glucose levels in the linear regression model, HbA_1c was significantly lower in C-Alb Q2 (P < 0.001), Q3 (P = 0.006), and Q4 (P < 0.001), compared with C-Alb Q1. As expected, HbA_1c and random glucose levels were significantly correlated (r = 0.55; P < 0.001) in the entire cohort, but the correlation was strongest in C-Alb Q1 (r = 0.66; P < 0.001) and weaker in C-Alb Q4 (r = 0.50; P < 0.001). Locally weighted scatterplot smoothing lines in Supplementary Fig. 2 also further demonstrated the inverse relationship between C-Alb and HbA_1c, eGFR, and hemoglobin, respectively.

During the median follow-up of 6.9 years, 763 (50.3%) individuals experienced CKD progression, 616 (40.6%) developed ESKD, and 669 (44.1%) died. In the fully adjusted models (model 2), every 1% increase in HbA_1c was associated with a 7% increased risk of CKD progression (adjusted hazard ratio [aHR] 1.07 [95% CI 1.02–1.13]). Results appeared similar when death was treated as a competing risk for CKD progression (aHR 1.10 [95% CI 1.01–1.19]). Using HbA_1c quartiles, compared with those participants in Q2 (reference group), patients in Q4 had a 39% higher risk of CKD progression (HR 1.39 [95% CI 1.14–1.69]). Results remained similar in the adjusted model 2 (Table 2) (e.g., compared with Q2, patients in Q4 had a significantly higher risk of CKD progression: aHR 1.28 [95% CI 1.04–1.56]) and after adding hemoglobin in model 3. The association between HbA_1c and renal outcomes when using HbA_1c 6.5–8% (48–64 mmol/mol) as the reference group is also shown in Table 2.

When an increase in HbA_1c was examined within each quartile of C-Alb, the magnitude of HbA_1c’s association with CKD progression was diminished in a graded relation for those with higher C-Alb levels compared with those with lower C-Alb levels (P-interaction = 0.022): among those in C-Alb Q1, the risk of CKD progression was increased by 23% in patients with every 1% increase in HbA_1c (aHR 1.23 [95% CI 1.11–1.36]); however, among those in C-Alb Q4, HbA_1c was no longer associated with risks of CKD progression (aHR 0.98 [95% CI 0.89–1.09]) (Fig. 2A). The interaction of HbA_1c and natural log-transformed C-Alb (P-interaction <0.001) is shown in Fig. 2B. Results remained similar after adding hemoglobin in the model (model 3). Additionally, in a subset of patients without anemia (n = 1,160), C-Alb remained an effect modifier of the association between HbA_1c and CKD progression (data not shown).

In patients with anemia (hemoglobin <11 g/dL), HbA_1c was no longer an independent risk factor for CKD progression, while in patients without anemia, HbA_1c was associated with a higher risk of CKD progression (aHR 1.12 [95% CI 1.06–1.19]) (Fig. 2A). Interaction testing between HbA_1c and the presence of anemia was significant for CKD progression (P-interaction = 0.025). The interaction of HbA_1c and hemoglobin (P-interaction <0.001) is shown in Fig. 2C.

In a large prospective national cohort of U.S. patients with diabetes and CKD, we found that the association between HbA_1c and CKD progression was modified by CKD-related factors, including high carbamylation levels and the presence of anemia. If any of the two factors existed, the association between HbA_1c and CKD progression was significantly diminished. However, as long as there were low carbamylation levels and no anemia, HbA_1c remained an independent risk factor for adverse renal outcomes in patients with diabetes and CKD. Our results carry particular significance considering HbA_1c is one of the most commonly used biomarkers to guide diabetes management and predict microvascular complications in patients with CKD and diabetes (10). While generally acknowledged as relevant, understanding the magnitude and direction of impact of CKD-related factors on HbA_1c interpretation has been lacking. To the best of our knowledge, this is the first study to rigorously quantify the complex relationship between glycation and carbamylation biomarkers in relationship to important clinical outcomes among this vulnerable patient population.

It has long been reported that carbamylation can cause interference with some HbA_1c assays, leading to a falsely elevated value. Flückiger et al. (13) described in 1981 that the increase in chromatographically measured HbA_1c levels in uremia was strongly correlated with urea levels and primarily caused by carbamylation of hemoglobin from urea-derived cyanate, while HbA_1c measured by the thiobarbituric acid method had no such interference with carbamylation. Since then, additional HbA_1c assays affected by carbamylation leading to overestimation of HbA_1c levels were identified: electrophoresis, cation-exchange chromatography, and fast protein liquid chromatography (14,16). There appears to be no interference from carbamylation if HbA_1c is measured by most HPLC methods, immunoassay, or capillary electrophoresis (14,16). In our study, HbA_1c was measured by Boronate affinity HPLC assay or ion-exchange HPLC method (Bio-Rad Variant II). Our findings of lower HbA_1c levels in higher C-Alb quartiles suggest that the HbA_1c assays used in our study did not provide falsely increased values because of interference from carbamylation; otherwise, the opposite relationship between HbA_1c and C-Alb levels should have been observed.

The relationship between glycation and carbamylation is far more complicated than assay interference. Glycation and carbamylation are chemical reactions that can compete for the same amino groups on proteins. In a well-designed animal study, Nicolas et al. (18) demonstrated that when the carbamylation reactions were intensified by either subtotal nephrectomy or cyanate consumption in diabetic mice, HbA_1cdecreased significantly (from 11.0% [97 mmol/mol] to 6.4% [46 mmol/mol] in subtotal nephrectomy mice, and from 11.3% [100 mmol/mol] to 7.3% [56 mmol/mol] in cyanate consuming mice) despite no significant change in glucose levels. This work clearly demonstrates an inhibited glycation process on hemoglobin under conditions of high carbamylation. Therefore, despite HbA_1c assays without analytical interference from carbamylation, it appears that HbA_1c can be misleadingly decreased due to competition from carbamylation in vivo. Indeed, in our study, participants with greater carbamylation load (reflected by higher C-Alb quartiles) had lower HbA_1c levels. An alternative explanation for such a finding could be that patients with higher carbamylation levels truly had lower mean glucose levels, perhaps due to factors such as altered nutritional status or insulin sensitivity. However, even after controlling for glucose levels, HbA_1c was still lower in high carbamylation groups. Moreover, the correlation between HbA_1c and glucose was attenuated at higher carbamylation levels. Taken together, our study shows that high carbamylation levels modified the association between HbA_1c and CKD progression, possibly due to the competitive relationship between the two posttranslational protein modifications.

As an additional layer of complexity of HbA_1c use in patients with CKD, anemia and its treatment are commonly cited factors complicating the interpretation of HbA_1c (14,15). Compared with patients with CKD but without diabetes with similar eGFR, people with diabetes had double the risk of anemia due to the damage to peritubular interstitial cells that produce erythropoietin (29). Previous work from our group has demonstrated that C-Alb is independently associated with erythropoietin resistance and likely contributes to anemia in patients with ESKD (30). In our current study, we showed that the presence of anemia significantly modified the association between HbA_1c and CKD progression, possibly due to altered red cell production and hemoglobin glycation accumulation. Furthermore, the interaction of C-Alb and HbA_1c remained significant, both after adjusting for hemoglobin levels and in a subset of patients without anemia, suggesting that carbamylation modifies the association of HbA_1c and adverse renal outcomes independent of anemia.

Optimized glycemic control in patients with diabetes has been shown to reduce kidney-related complications. The Diabetes Control and Complications Trial (DCCT) demonstrated that intensive glycemic control (HbA_1c 7% [53 mmol/mol]) versus the control group (HbA_1c 9% [75 mmol/mol]) in patients with type 1 diabetes reduced the occurrence of albuminuria by 54% (31). Follow-up of DCCT showed the persistence of these renal benefits over two decades (“legacy effect”) (32). Nevertheless, despite decreases in macrovascular complications such as myocardial infarction in diabetes attributed to effective therapies, including statins and antiplatelet medications, the incidence and prevalence of diabetic kidney disease are rising (7,8). Our results reinforce this epidemiology, as over half of our study population demonstrated CKD progression and >40% developed ESKD. One could speculate that the high level of observed renal complications could be partly attributed to falsely low HbA_1c values, potentially misleading patients and providers to inaccurate glycemic targets. Both American Diabetes Association and KDIGO recommended that the target HbA_1c in patients with diabetes and CKD should be individualized and account for the severity of CKD, macrovascular complications, comorbidities, life expectancy, and hypoglycemia risk (9,11,12).

Alternative methods to evaluate glycemic control, such as glycated albumin, fructosamine, and continuous glucose monitoring, have been proposed (10,14,15,33,34). Although free from the influence of anemia and its treatment, glycated albumin and fructosamine are still impacted by the competition between glycation and carbamylation, and their prognostic significance is not as clear as HbA_1c yet. Continuous glucose monitoring is another appealing option, but its cost has limited wide adoption, and the variability in glycemic patterns and lack of standardization complicate the analysis and interpretation of results (10,35). HbA_1c remains a good risk assessment with considerable evidence and a universally available reference measurement system (34), as long as clinicians are aware of its limitations, including possible discrepancies between HbA_1c and true mean blood glucose levels in some patients with CKD. It is prudent that clinicians consider the influence of CKD-related factors such as carbamylation and anemia on the relationship between HbA_1c and CKD progression when caring for patients with coexisting diabetes and CKD.

Our study has several limitations. First, lack of information on diabetes type precludes analysis of differences in the relationship between type 1 and type 2 diabetes. HbA_1c was measured with two different assays in the CRIC study, and we did not have data at the individual level on which assay was used. Nevertheless, neither assay has interference with carbamylation. Time variations of the biomarkers were not accounted for, as C-Alb was only measured at one time point. Lastly, CRIC is an observational cohort, and as such, the possibility of residual confounding cannot be excluded. Nevertheless, CRIC has collected a robust list of covariates, including sociodemographic characteristics, lifestyle behaviors, and extensive clinical data, which allow us to use powerful multivariable regression analysis to minimize confounding. Additional strengths of the study include the large racially and ethnically diverse cohort, adding to the generalizability of our findings.

In conclusion, our study demonstrated that the association between high HbA_1c levels and adverse renal outcomes among patients with diabetes and CKD was modified by carbamylation levels and the presence of anemia. This finding may explain why HbA_1c is less reliable in patients with CKD compared with the general population with diabetes. Future studies should evaluate whether incorporating carbamylation biomarkers and hemoglobin levels can improve the risk prediction of HbA_1c in patients with CKD.

Acknowledgments. The authors thank the CRIC investigators and participants.

Funding. S.K. is supported by National Institute of Diabetes and Digestive and Kidney Diseases grants R01DK124453 and U01DK123818. A.B. is supported by National Heart, Lung, and Blood Institute grant R01HL133399.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. M.T. and S.K. contributed to the project administration, conceptualization, study design, statistical analysis, and data interpretation. M.T. and S.K. wrote the original draft of the manuscript. A.B., E.P.R., A.S.A., S.N., S.A.K., and J.P.L. contributed to the review and editing of the manuscript. A.B. and S.A.K. took blood measurements. S.K. acquired funding. All authors reviewed and approved the final version of the manuscript. M.T. and S.K. are the guarantors of this work and, as such, had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.