Introduction
Chronic kidney disease (CKD) is a global public health issue with a prevalence of 9.1% in the world’s population and is the ninth leading cause of death worldwide, with the number of deaths increasing by 95% between 2000 and 2021 [1, 2]. The kidneys are responsible for maintaining glucose homeostasis through glucose and insulin metabolism and for medication clearance. Glucose homeostasis is altered in a complicated manner in patients with CKD [3].
Diabetes mellitus (DM) is a leading comorbidity of CKD, with the largest contribution in terms of the number of CKD disability-adjusted life-years [1]. Poor glycemic control is associated with chronic diabetes complications [4]. Therefore, choosing a biomarker that reflects not only glycemic control but also outcomes in CKD patients is critical.
Hemoglobin A1c (HbA1c) is a key biomarker for assessing glycemic control. Associations with vascular outcomes and mortality were identified in both DM patients and individuals without a prior diagnosis of DM [5, 6]. Although HbA1c has been recommended for monitoring glycemic control in diabetic patients with CKD, as outlined by the guidelines from both the American Diabetes Association (ADA) and Kidney Disease Improving Global Outcomes (KDIGO), it has pitfalls that could decrease its accuracy due to carbamylated hemoglobin and a reduced erythrocyte lifespan from uremia, the presence of anemia and the use of erythropoiesis-stimulating agents (ESAs) [4, 7, 8].
A candidate for glycemic monitoring, glycated albumin (GA), has been proposed [4]. Compared with HbA1c, GA provides a more recent reflection of glycemic status (within 2–4 weeks), as it has a shorter half-life in the bloodstream [9]. Previous studies have indicated that GA levels are associated with the risk of CVD and mortality among individuals with and without DM [10].
However, GA may be considered a problematic measure of glycemic control in patients with CKD because its accuracy and precision could be influenced by hypoalbuminemia.
The KDIGO guidelines indicate that the reliability of HbA1c for glycemic monitoring decreases at more advanced stages of CKD [4]. Our prior study demonstrated that GA was linked to an increased risk of cardiovascular (CV) and all-cause mortality in both dialysis and nondialysis patients, and GA has been shown to be superior to HbA1c in assessing blood glucose control among patients with CKD [10–12]. GA has also been demonstrated across multiple studies to outperform HbA1c in predicting the mortality of dialysis patients [13–18].
However, comparisons of the prognostic associations between GA and HbA1c among CKD patients have not been rigorously established. We hypothesized that GA would show a stronger association with mortality compared with HbA1c in CKD patients.
This study sought to compare the associations of GA and HbA1c with mortality in patients with CKD and to explore whether GA provides complementary value in clinical practice.
Methods
Data sources and study subjects
The National Health and Nutrition Examination Survey (NHANES), overseen by the Centers for Disease Control and Prevention (CDC), is designed to assess the well-being and dietary conditions of individuals across the United States. Its protocols undergo rigorous review and approval by the Research Ethics Review Board of the National Center for Health Statistics (NCHS), ensuring ethical compliance and participant protection through informed written consent. To promote openness and foster future studies, the NHANES datasets utilized in this research are made accessible to the public via the official website (https://www.cdc.gov/nchs/nhanes/index.html).
The selection process, as outlined in Fig. 1, began with 31,126 participants from the NHANES datasets spanning 1999–2004, the period during which GA measurements were available. Participants aged 18 years or older were included.
Exclusions were applied sequentially: those younger than 18 years (n = 14,065), individuals with missing GA data (n = 3,258) or HbA1c data (n = 12), and those with missing mortality information (n = 13). Additional exclusions were made for participants without CKD, defined as those with an eGFR ≥ 60 mL/min/1.73 m² and an ACR < 30 mg/g, as well as those undergoing dialysis. By applying these exclusion criteria, 2,433 participants were retained in the final analytic cohort.
Fig. 1Study selection from the NHANES 1999-−2004; workflow and major findings of this study. ACR: albumin-to-creatinine ratio; CKD: chronic kidney disease; eGFR: estimated glomerular filtration rate; NHANES: National Health and Nutrition Examination Survey
Estimates of CKD
In NHANES 1999–2004, serum and urinary creatinine levels were quantified via a solid-phase fluorescent immunoassay and analyzed via the kinetic Jaffe rate reaction method [19]. The estimated glomerular filtration rate (eGFR) was derived from serum creatinine levels via the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation [20]. This formula is expressed as follows: eGFR = 141 × min (Scr/κ, 1) α × max (Scr/κ, 1)−1.209 × 0.993Age × 1.018 [if female] × 1.159 [if Black], where Scr is the serum creatinine level, κ is 0.7 for females and 0.9 for males, and α is −0.329 for females and − 0.411 for males.
“Min” and “max” indicate the minimum and maximum of Scr/κ or 1, respectively. Urinary albumin levels were measured via a fluorescence immunoassay, and the albumin-to-creatinine ratio (ACR) was determined by dividing the urine albumin level by the creatinine concentration. CKD was defined according to current guidelines as an eGFR < 60 mL/min/1.73 m² or an ACR ≥ 30 mg/g [21].
CKD status was primarily categorized into two groups according to the eGFR, following the classification commonly used in nephrology practice and research [22, 23], early CKD was defined as eGFR ≥ 30 mL/min/1.73 m², and advanced CKD was defined as eGFR < 30 mL/min/1.73 m². This classification served as the main exposure variable throughout the analyses. Additionally, participants with CKD were further stratified into moderately increased-risk, high-risk, and very high-risk categories according to the KDIGO prognosis risk classification based on cause, GFR category (G1–G5), and albuminuria category (A1–A3), abbreviated as CGA [21].
Measurements of GA and HbA1c
GA was measured via all available surplus serum samples from NHANES participants between 1999 and 2004. Laboratory testing took place from 2018 to 2020 at the University of Maryland School of Medicine. GA levels were determined with the Lucica-GA-L assay (Asahi Kasei Pharma Corporation), which measures total albumin via the bromocresol purple method and GA via an enzymatic approach involving ketoimine oxidase and an albumin-specific protease.
These measurements were conducted on an automated multichannel analyzer (Siemens Dimension Vista 1500, Siemens Diagnostics Corp.). GA was calculated as a percentage based on the formula provided by the manufacturer: [(glycated albumin, g/dL/serum total albumin, g/dL) × 100/1.14] + 2.9, and the lower limit of detection for GA was 3.0% [24].
HbA1c was measured in whole blood via high-performance liquid chromatography (HPLC) at the University of Missouri-Columbia. The analysis employed the Primus CLC330 and CLC385 systems (Primus Corporation), with standardization based on the DCCT protocol. The minimum detectable value for HbA1c was 4.0%.
Assessment of mortality outcomes
Mortality data for the follow-up period were obtained from NHANES Public-Use Linked Mortality Files, which are connected to National Death Index records up to December 31, 2019. Cause-of-death information was categorized via the 10th Revision of the International Statistical Classification of Diseases (ICD-10), whereas all-cause mortality included deaths from any cause (ICD-10 codes A00–Y89). CV deaths were defined as those attributable to rheumatic heart diseases, hypertensive heart and renal diseases, ischemic heart disease, and heart failure (ICD-10 codes I00–I09, I11, I13, and I20–I51).
Measurements of covariates
Baseline demographic data, including age, sex, race, education, smoking status, marital status, drinking habits, and medication use, were collected through NHANES household interviews. Height and weight were measured by trained staff at the Mobile Examination Center, and body mass index (BMI) was calculated as weight (kg) divided by height (m²). Race was categorized into White, Black, Mexican, or other, while education levels were grouped as less than high school, high school or equivalent, and college or above.
Smoking status was classified as current smokers, former smokers (≥ 100 cigarettes in their lifetime, now quit), and nonsmokers. Drinking status was divided into three categories: current drinkers (≥ 12 drinks consumed in the past year), former drinkers (> 12 lifetime drinks but none in the past year), and nondrinkers. The use of hypolipidemic agents was determined based on participants’ responses to the following question: “Are you currently taking prescribed medication for high lipid levels?” A response of “yes” indicated current use.
Similarly, antihypertensive medication use was identified through the question “Are you currently taking prescribed medication for high blood pressure?” Participants who answered “Yes” to either question were classified as medication users, whereas those who answered “No” to both were categorized as nonusers. Antidiabetic therapy was assessed via participants’ responses to two questions: “Are you now taking insulin?” and “Are you taking diabetic pills to lower blood sugar?” Participants who responded “Yes” to either were classified as users of antidiabetic agents. CKD-related therapies include metabolic and acid‒base regulating agents (e.g., compound α-ketoacid and sodium bicarbonate) and anemia treatments such as ESAs, iron preparations, and folic acid.
Cardiovascular disease (CVD) status was defined by a history of any one of the following conditions: coronary heart disease, congestive heart failure, heart attack, stroke, or angina.
Baseline measurements of total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), and fasting triglycerides (TG) were obtained from blood samples collected at the Mobile Examination Center.
Self-reported medical histories were used to identify cases of heart failure (HF), coronary heart disease (CHD), and stroke among participants. Hypertension was diagnosed if participants had a prior medical history of high blood pressure, were receiving treatment with antihypertensive medications, or had systolic blood pressure readings of ≥ 130 mm Hg or diastolic readings of ≥ 80 mm Hg [25]. The diagnosis of diabetes mellitus (DM) followed standardized criteria, including fasting plasma glucose levels ≥ 126 mg/dL (≥ 7.0 mmol/L), HbA1c levels > 6.5% [26], a prior DM diagnosis, or evidence of antidiabetic medication use, such as insulin or oral glucose-lowering agents.
Anemia was defined on the basis of hemoglobin levels as follows: for nonpregnant women, anemia was classified as hemoglobin < 120 g/L, whereas nonanemia was ≥ 120 g/L; for pregnant women, anemia was < 110 g/L, and nonanemia was ≥ 110 g/L; and for men, anemia was < 130 g/L, and nonanemia was ≥ 130 g/L [27].
Dyslipidemia was considered based on self-reported diagnosis, the use of cholesterol-lowering medications, TC ≥ 6.2 mmol/L, or LDL-C ≥ 4.1 mmol/L [28].
Statistical analysis
Participants were stratified into tertiles on the basis of baseline HbA1c levels (Q1 ≤ 5.4, Q2 5.4–5.9, and Q3 ≥ 5.9) and baseline GA levels (Q1 ≤ 13.4, Q2 13.4–15.6, and Q3 ≥ 15.6). Continuous variables are presented as the means ± standard errors (SEs), and categorical variables are presented as frequencies with percentages. Baseline characteristics across GA and HbA1c tertiles were analyzed via one-way analysis of variance (ANOVA) for continuous variables and Pearson’s chi-square test for categorical variables.
Insulin resistance (IR) is a key factor in the progression of type 2 DM and CKD, contributing to both metabolic disorders and CVD risk. Both GA and HbA1c reflect glucose metabolism, and their associations with IR are clinically important. Currently, the homeostasis model assessment of insulin resistance index (HOMA-IR index), quantitative insulin sensitivity check index (QUICKI index), and homeostasis model assessment of β-cell function index (HOMA-β index) are popular methods for measuring IR and are calculated according to the following formulas:
HOMA-IR index =\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\frac{fasting\:glucose\:(mmol/L)\times\:\:fasting\:insulin\:(\mu\:U/mL)\:}{22.5}$$\end{document} QUICKI index =\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\frac{1}{log(fasting\:insulin,\:\mu\:U/mL)\:+\:log(fasting\:plasma\:glucose,\:mg/dL)}$$\end{document} HOMA-β index =\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\:\frac{20\:\times\:\:fasting\:insulin\:(\mu\:IU/ml)}{fasting\:glucose\:(mmol/ml)\:-\:3.5}$$\end{document}
Age-, sex-, and race-adjusted linear regression analyses were performed to assess the associations between GA/HbA1c and the IR indices.
Survival curves for GA and HbA1c levels were generated with weighted log-rank tests. To investigate potential nonlinear relationships between GA/HbA1c and mortality, restricted cubic spline (RCS) regression models were constructed with three knots at the 10th, 50th, and 90th percentiles. Weighted multivariate Cox proportional hazards models, expressing hazard ratios (HRs) and 95% confidence intervals (CIs), were used to adjust for confounding factors, with GA and HbA1c analyzed as both continuous and categorical variables.
The crude model was unadjusted, whereas Model I included adjustments for age, sex, and race. Model II expanded upon Model I by additionally adjusting for smoking status, alcohol use, hypertension, CVD status, DM status, eGFR, ACR and hypolipidemic and hypotensive medications. Covariates were selected based on established clinical experience in clinical practice and risk factors as specified in relevant clinical guidelines [21, 29].
Additionally, clinical thresholds (GA: 17.8%, HbA1c: 6.5%) were used to dichotomize the participants, and Cox models were used to assess their associations with all-cause and cardiovascular mortality. To examine potential effects on the associations between GA/HbA1c and mortality, subgroup analyses were performed via weighted multivariable Cox models. The subgroups included sex; age (< 60 vs. ≥ 60 years); race (White, Mexican, Black); BMI (< 25 vs. ≥ 25 kg/m²); smoking and alcohol use; hypertension; diabetes; anemia; serum albumin (< 35 vs. ≥ 35 g/L); ACR (< 30, 30–300, ≥ 300 mg/g); eGFR (< 30 vs. ≥ 30 mL/min/1.73 m²); and antidiabetic medication use.
Subgroup models employed age, sex, race, BMI, ACR, and eGFR, with the stratifying variable excluded from its respective model; interactions were tested by including multiplicative interaction terms.
The discrimination of the prognostic performance of GA and HbA1c was analyzed via receiver operating characteristic (ROC) curves, and the C-index was utilized to assess performance. The comparison of C-index values was conducted via the method described by DeLong et al. [30]. Additionally, time-dependent ROC analysis was used to evaluate the predictive accuracy of GA and HbA1c at specific time points during the follow-up period.
This analysis generated ROC curves and C-index values for each time point, providing insight into the models’ ability to discriminate between events and nonevents over time. In addition, we fitted a univariate Cox model for each marker, used the model’s linear predictor as the risk score, and estimated time-dependent C-indices from these scores to quantify discrimination over follow-up while accounting for censoring; time-dependent ROC analyses were used to generate ROC curves and C-index values at specific time points.
Furthermore, the predictive performance of the combined risk models with the addition of GA or HbA1c to the existing risk models (Framingham Heart Age and Framingham CVD risk score) for mortality outcomes was compared. The Framingham Heart Age model, derived from the Framingham Heart Study, estimates an individual’s “heart age” on the basis of their 10-year risk of CVD. This model includes essential risk factors such as age, sex, smoking status, blood pressure, and cholesterol levels.
The calculated heart age represents the chronological age of an individual with the same predicted 10-year CVD risk but with all risk factors within the optimal range. The detailed calculation methods can be found in the original study by D’Agostino et al. [31]. Similarly, the Framingham CVD risk score system assigns numerical scores to various risk factors to estimate a person’s 10-year likelihood of developing CVD.
This system incorporates age, total cholesterol, HDL cholesterol, systolic blood pressure (categorized by whether antihypertensive treatment is used), smoking status, and diabetes status. Each risk factor contributes a specific point value, with the total score corresponding to an estimated percentage risk over a decade. The methodology, including point allocations and risk estimations, is documented in the Framingham Heart Study publications.
The C-index for the combined risk models was calculated. NRI (Net Reclassification Improvement) and IDI (Integrated Discrimination Improvement) were used to quantify the improvement in the predictive performance of the combined models. Confidence intervals (CIs) for the NRIs and IDIs were derived via bootstrap resampling.
Additionally, we quantified the incremental prognostic association of adding GA to clinical models based on eGFR and ACR via changes in the C-index and reclassification metrics.
Subgroup analyses were conducted by anemia status, CKD stage (early vs. advanced), ACR category (< 30, 30–300, and ≥ 300 mg/g), serum albumin level (< 35 vs. ≥35 g/L), and KDIGO prognosis risk classification (moderate, high, very high). In the supplementary analyses, we also summarized confusion matrix metrics, including sensitivity, specificity, accuracy, PPV, and NPV for GA and HbA1c at optimal discrimination thresholds in the overall cohort and across these prespecified subgroups, and visualized the count matrices.
Multiple sensitivity analyses were performed: (1) removing participants with less than 2 years of follow-up; (2) excluding participants under 30 years of age to reduce the influence of genetic diseases; (3) conducting landmark survival analyses at 5 years to assess temporal robustness; and (4) recalculating the eGFR via the Modification of Diet in Renal Disease (MDRD) equation, which is defined as eGFR = 175 × (Scr) −1.154 × (Age) −0.203 × (0.742 if female) × (1.212 if Black), where Scr represents serum creatinine (mg/dL), and age is measured in years [32].
The statistical analyses were carried out via R software (version 4.2.1; https://www.r-project.org). To address the complex sampling design of the NHANES, sample weights, clustering, and stratification were incorporated into the analysis [33]. Each individual’s analytic weight captures the inverse of their probability of selection, adjusted for nonresponse and calibrated to U.S. population totals.
We specified SDMVPSU as the primary sampling unit (PSU) variable and SDMVSTRA as the stratum variable and applied the MEC exam sample weight (WTMEC2YR) (adjusted for combining cycles) to all analyses [33]. P < 0.05 indicated statistical significance.
Results
Baseline characteristics of the study participants
As shown in Table S1, the cohort consisted of 2,433 participants, weighted to represent 28,536,547 individuals (mean age 60.57 years [SE: 0.67]; 57.48% female). The average baseline GA was 15.36% (SE: 0.19), the HbA1c was 5.97% (SE: 0.04), and the eGFR was 72.92 mL/min/1.73 m² (SE: 0.83). The proportions of comorbidities were as follows: hypertension (64.29%), diabetes mellitus (26.08%), hyperlipidemia (81.84%), heart failure (8.25%), coronary heart disease (9.99%), and stroke (7.65%).
The distribution of the participants with CKD grouped by eGFR and ACR categories is presented in Online Table S2; 1,207 participants had reduced kidney function (eGFR < 60 mL/min/1.73 m²), and 1,616 participants had albuminuria (ACR ≥ 30 mg/g). The participants’ baseline characteristics categorized by tertiles of GA and HbA1c are presented in Tables 1 and 2, respectively.
Table 1Baseline characteristics of the participants in the cohort from the National Health and Nutrition Examination Survey (1999−2004) grouped by glycated albumin tertilesVariableOverallGlycated albuminP valueQ1(≤13.4)Q2(13.4–15.6)Q3(≥15.6)N (%)2433805 (33.09)823 (33.83)805 (33.09)Glycated Albumin, %15.36 (0.19)12.22 (0.05)14.34 (0.03)21.08 (0.45)< 0.001Age, years60.57 (0.67)54.15 (1.05)63.97 (1.14)65.60 (0.92)< 0.001Female, n (%)1295 (57.48)448 (57.90)467 (63.16)380 (50.02)< 0.001BMI, kg/m228.90 (0.25)29.34 (0.32)27.56 (0.32)29.91 (0.46)< 0.001ACR, mg/g162.79 (17.04)142.32 (22.35)113.65 (17.20)253.16 (36.69)0.003eGFR, mL/min/1.73 m²72.92 (0.83)80.14 (1.54)68.70 (1.44)67.69 (1.45)< 0.001Race, n (%)< 0.001 Black439 (11.60)108 (8.34)145 (11.63)186 (16.23) Mexican481 (5.76)180 (7.32)118 (3.90)183 (5.77) Other173 (10.88)55 (10.41)49 (8.48)69 (14.47) White1340 (71.76)462 (73.94)511 (75.99)367 (63.54)Education, n (%)< 0.001 Less than high school983 (31.18)270 (27.36)311 (28.18)402 (42.30) High school grad or equivalent543 (25.91)198 (27.72)194 (29.49)151 (20.87) Some college or above803 (41.02)283 (44.93)285 (42.33)235 (36.83)Marital, n (%)0.80 Married1163 (49.32)390 (51.84)382 (49.57)391 (50.71) Other1204 (47.80)398 (48.16)415 (50.43)391 (49.29)Alcohol user, n (%)< 0.001 Former661 (24.80)173 (22.11)206 (25.55)282 (34.55) No431 (17.67)124 (18.15)150 (19.35)157 (20.00) Yes1104 (50.15)403 (59.73)394 (55.09)307 (45.45)Smoke, n (%) Former834 (32.79)216 (28.09)312 (36.17)306 (37.51) Never1119 (46.50)357 (45.12)374 (50.51)388 (46.71) Now380 (18.90)178 (26.79)104 (13.32)98 (15.78)CKD Prognosis Risk Category, n (%)< 0.001 Moderately increased risk1,651 (69.66)632 (79.10)558 (69.40)461 (60.10) High risk440 (18.57)121 (15.14)147 (18.28)172 (22.43) Very high risk279 (11.77)46 (5.76)99 (12.31)134 (17.47)Laboratory results, mean (SE) FPG, mg/dL118.20 (2.06) 99.42 (0.82)104.92 (1.58)162.08 (4.49)< 0.001 TC, mg/dL205.27 (1.49)208.35 (2.20)200.90 (2.00)206.18 (2.09)0.01 TG, mg/dL179.47 (9.18)183.80 (12.38)137.96 (5.04)229.13 (26.88)< 0.001 LDL-C, mg/dL119.31 (1.64)122.53 (2.59)118.99 (2.40)114.80 (2.99)0.21 HDL-C, mg/dL51.58 (0.55)50.55 (0.64)53.65 (0.93)50.55 (0.83)0.01Disease, n (%) HF231 (8.25)43 (4.21)89 (9.57)99 (12.97)< 0.001 CHD262 (9.99)65 (7.55)94 (10.08)103 (14.37)0.001 Angina196 (7.76)49 (5.92)67 (7.87)80 (10.99)0.01 Heart attack269 (9.38)60 (6.44)95 (10.02)114 (13.41)< 0.001 Stroke216 (7.65)55 (6.71)71 (6.77)90 (10.55)0.03 Hyperlipidemia1974 (81.84)650 (82.34)656 (79.78)668 (83.61)0.25 DM730 (26.08)51 (6.04)130 (15.36)549 (67.81)< 0.001 Hypertension1671 (64.29)480 (56.14)580 (65.15)611 (75.14)< 0.001Medication, n (%) Anti-hypertensive agents, n (%)800 (60.06)212 (55.79)250 (56.18)338 (66.67)<0.001 Lipid-lowering agents, n (%)407 (81.12)143 (83.16)160 (78.91)194 (81.01)0.74 Antidiabetic agents, n (%)509 (69.25)28 (54.90)79 (59.40)402 (72.96)< 0.001 CKD-related agents, n (%)14 (0.58)2 (0.25)4 (0.49)8 (0.99)0.08 Data are expressed as the means (SEs) and numbers (percentages) as appropriate. All estimates were weighted to be nationally representativeACR: albumin-to-creatinine ratio; BMI: body mass index; FPG: fasting plasma glucose; TC: total cholesterol; LDL-C: low-density lipoprotein cholesterol; HDL-C: high-density lipoprotein cholesterol; HbA1c: hemoglobin A1c; HF: heart failure; CHD: coronary heart disease; DM: diabetes mellitus; eGFR: estimated glomerular filtration rate
Table 2Baseline characteristics of the participants in the cohort from the National Health and Nutrition Examination Survey (1999–2004) grouped by hemoglobin A1c tertilesVariableOverallHemoglobin A1cP valueQ1(≤ 5.4)Q2(5.4–5.9)Q3(≥ 5.9)N (%)2433994 (40.85)661 (27.17)828 (34.03)< 0.001Hemoglobin A1c, %5.97 (0.04)5.14 (0.01)5.63 (0.01)7.50 (0.09)< 0.001Age, years60.57 (0.67)54.73 (0.88)65.82 (0.95)64.11 (0.83)< 0.001Female, n (%)1295 (57.48)568 (63.16)349 (59.60)378 (47.00)< 0.001BMI, kg/m228.90 (0.25)26.93 (0.28)29.02 (0.34)31.76 (0.45)< 0.001ACR, mg/g162.79 (17.04)125.02 (20.99)118.31 (20.68)261.31 (32.52)0.01eGFR, mL/min/1.73 m²72.92 (0.83)78.07 (1.19)66.83 (1.28)71.20 (1.35)< 0.001Race, n (%)< 0.001 Black439 (11.60)146 (9.30)108 (10.45)185 (16.13) Mexican481 (5.76)176 (5.85)83 (4.42)222 (6.92) Other173 (10.88)56 (9.33)45 (8.83)72 (15.18) White1340 (71.76)566 (75.52)425 (76.30)349 (61.77)Education, n (%)< 0.001 Less than high school983 (31.18)304 (24.99)250 (30.13)429 (43.07) High school grad or equivalent543 (25.91)231 (29.50)155 (27.26)157 (21.16) Some college or above803 (41.02)323 (45.51)245 (42.61)235 (35.77)Marital, n (%)0.28 Married1163 (49.32)414 (48.05)324 (52.68)425 (52.93) Other1204 (47.80)495 (51.95)321 (47.32)388 (47.07)Alcohol user, n (%)0.002 Former 661 (24.80)207 (21.09)190 (29.90)264 (31.66) No431 (17.67)157 (19.80)107 (17.44)167 (19.63) Yes1104 (50.15)441 (59.11)317 (52.66)346 (48.70)Smoke, n (%) Former834 (32.79)278 (28.90)233 (34.70)323 (38.56) Never1119 (46.50)431 (49.94)306 (46.15)382 (44.82) Now380 (18.90)150 (21.16)113 (19.14)117 (16.62)CKD Prognosis Risk Category, n (%)< 0.001 Moderately increased risk1,651 (69.66)691 (74.86)455 (70.32)505 (63.13) High risk440 (18.57)140 (15.17)123 (19.01)177 (22.13) Very high risk279 (11.77)92 (9.967)69 (10.66)118 (14.75)Laboratory results, mean (SE)FPG, mg/dL118.20 (2.06)95.89 (0.71)102.29 (1.18)167.88 (4.45)< 0.001TC, mg/dL205.27 (1.49)201.52 (2.10)209.48 (2.62)206.75 (2.26)0.03TG, mg/dL179.47 (9.18)150.44 (11.06)158.57 (9.54)243.71 (23.68)0.003LDL-C, mg/dL119.31 (1.64)117.62 (2.40)125.67 (3.01)114.03 (2.44)0.03HDL-C, mg/dL51.58 (0.55)54.35 (0.87)51.74 (0.84)47.32 (0.71)< 0.001Disease, n (%) HF231 (8.25)64 (4.70)101 (12.09)66 (12.59)< 0.001 CHD262 (9.99)76 (7.60)84 (11.16)102 (13.30)0.01 Angina196 (7.76)58 (6.83)56 (7.49)82 (10.10)0.17 Heart attack269 (9.38)68 (6.56)84 (9.42)117 (14.01)< 0.001 Stroke216 (7.65)74 (6.18)55 (8.57)87 (9.38)0.13 Hyperlipidemia1974 (81.84)693 (75.43)551 (84.81)730 (88.44)< 0.001 DM730 (26.08)31 (2.81)71 (10.32)628 (75.95)< 0.001 Hypertension1671 (64.29)536 (51.80)486 (70.61)649 (76.89)< 0.001Medication, n (%) Anti-hypertensive agents, n (%)800 (60.06)196 (47.57)212 (55.79)392 (72.59)< 0.001 Lipid-lowering agents, n (%)407 (81.12)101 (74.08)153 (83.70)243 (83.15)0.14 Antidiabetic agents, n (%)509 (69.25)21 (63.64)36 (50.00)452 (71.75)< 0.001 CKD-related agents, n (%)14 (0.58)6 (0.64)4 (0.61)4 (0.48)0.91Data are expressed as the means (SEs) and numbers (percentages) as appropriate. All estimates were weighted to be nationally representativeACR: albumin-to-creatinine ratio; BMI: body mass index; FPG: fasting plasma glucose; TC: total cholesterol; LDL-C: low-density lipoprotein cholesterol; HDL-C: high-density lipoprotein cholesterol; HbA1c: hemoglobin A1c; HF: heart failure; CHD: coronary heart disease; DM: diabetes mellitus; eGFR: estimated glomerular filtration rate
Correlations of GA and HbA1c with IR indices
Figure S1 depicts the correlations between GA/HbA1c and IR indices. GA was significantly linearly correlated with HOMA-IR (r = 0.31, P < 0.001) and QUICKI (r = −0.29, P < 0.01) but not with HOMA-β (r = −0.01, p = 0.28). HbA1c was associated with IR indices comparable in magnitude to those associated with GA (r = 0.39, P < 0.001 for HOMA-IR; r = −0.46, P < 0.01 for QUICKI; and r = −0.03, P = 0.61 for HOMA-β).
Notably, the correlations between HbA1c and both HOMA-IR and QUICKI were significantly stronger than those observed for GA (all P < 0.05).
Associations of GA and HbA1c with mortality
Exposure-effect analysis
Over a 12.46-year median follow-up period, 1532 deaths (62.97%) from all causes and 473 deaths (19.44%) from CVD were reported. The KM survival curves demonstrated that higher GA levels were associated with increased risks of all-cause and CV mortality compared with lower levels (all log-rank P values ≤ 0.0001). The results were similar when patients were stratified by HbA1c levels (log-rank P ≤ 0.0001 for all-cause mortality and log-rank P = 0.002 for CV mortality) (Fig.
2).
Fig. 2Kaplan‒Meier curves of the relationships between Glycated Albumin/Hemoglobin A1c and mortality outcomes in CKD patients. A: Kaplan‒Meier curves in tertiles of Glycated Albumin with all-cause mortality.
B: Kaplan‒Meier curves for the tertiles of Hemoglobin A1c with all-cause mortality. C: Kaplan‒Meier curves for tertiles of Glycated Albumin with CV mortality. D: Kaplan‒Meier curves for tertiles of Hemoglobin A1c with CV mortality CV: cardiovascular
Exposure-effect analysis revealed distinct associations of GA and HbA1c with the outcomes. HbA1c levels showed U-shaped correlations with all-cause and CV mortality (Pnonlinearity = 0.02 for all-cause mortality and 0.04 for CV mortality), with the lowest risk at an HbA1c of approximately 5.4% and a higher risk at both lower and higher HbA1c levels (Online Figure S2A & S2B). However, GA levels were positively linearly correlated with all-cause mortality (Pnonlinearity = 0.18) and CV mortality (Pnonlinearity = 0.29) (Online Figure S2C & S2D).
Categorical and continuous variable analysis
Table 3 presents the crude and adjusted associations between categorical GA/HbA1c baseline levels and the outcomes. According to the fully adjusted model, there were gradually higher risks of outcomes among participants in the three groups stratified by GA level. Specifically, compared with the GA Q1 group (< 13.4), individuals in the GA Q2 group (13.4–15.6) had a 1.16-fold (95% CI 1.02–1.33) greater risk of all-cause mortality.
Those in the GAQ3 group (> 15.6) demonstrated a 1.53-fold (95% CI 1.24–1.89) greater risk of all-cause mortality and a 1.76-fold (95% CI 1.23–2.50) greater risk of CV mortality. When grouped by HbA1c level, individuals in the HbA1c Q3 group (≥ 5.9) presented a 1.36-fold (95% CI 1.02–2.04) greater risk of CV mortality than those in the HbA1c Q1 group (< 5.4). Notably, those in the HbA1c Q2 group (5.4–5.9) presented a reduced risk of all-cause mortality (HR 0.85, 95% CI 0.72–0.99).
In the threshold effect analysis, individuals with a higher level of GA (≥ 17.8%) had increased risks of both all-cause mortality (HR 1.32, 95% CI 1.10–1.58) and CV mortality (HR 1.68, 95% CI 1.17–2.42) compared with those with a lower level (< 17.8%). A similar disparity existed between patients with higher levels of HbA1c (≥ 6.5%) and those with lower levels (< 6.5%) (HR for all-cause mortality 1.39, 95% CI 1.16–1.67; HR for CV mortality 1.60, 95% CI 1.10–2.32) (Online Table S3).
Analyzed as a continuous variable, GA was significantly linked to elevated risks of all-cause and CV mortality. Comparable patterns were observed in the impact of HbA1c on mortality (Table 3).
Table 3Associations of glycated albumin/hemoglobin A1c with all-cause and cardiovascular mortality in patients with chronic kidney diseaseGlycated AlbuminHemoglobin A1cNumber of deaths/totalsModel IHR (95% CI)Model IIHR (95% CI)Number of deaths/totalsModel IHR (95% CI)Model IIHR (95% CI)All-cause mortalityQ1527 (42.03)Ref.Ref.Q1502 (45.70)Ref.Ref.Q2598 (69.24)1.17 (1.04,1.30)1.16 (1.02,1.33)Q2676 (61.07)0.88 (0.75,1.03)0.85 (0.72,0.99)Q3407 (68.64)1.65 (1.42,1.93)1.53 (1.24,1.89)Q3354 (65.12)1.23 (1.02,1.49)1.01 (0.79,1.29)Per 3-unit increase1532 (62.96)1.01 (1.01,1.02)1.01 (1.01,1.02)Per 1 unit increase1532 (62.96)1.11 (1.06,1.17)1.07 (1.00,1.15)CV mortalityQ1148 (12.17)Ref.Ref.Q1146 (13.51)Ref.Ref.Q2191 (21.84)1.05 (0.78,1.40)1.01 (0.71,1.39)Q2218 (19.09)0.89 (0.65,1.22)0.94 (0.69,1.27)Q3134 (25.25)1.98 (1.42,2.75)1.76 (1.23,2.50)Q3109 (23.82)1.52 (1.07,2.18)1.36 (1.02,2.04)Per 3-unit increase473 (19.44)1.02 (1.02,1.03)1.02 (1.02,1.03)Per 1 unit increase473 (19.44)1.18 (1.09,1.27)1.14 (1.03,1.27) Model I: Adjusted for age, sex, and raceModel II: Adjusted for age, sex, race, smoking status, alcohol use, hypertension status, CVD status, DM status, eGFR, ACR, hypolipidemic and hypotensive medications95% CI: 95% confidence interval; HR: hazard ratio; Ref: reference, CV: cardiovascular
Stratified analyses
GA levels were consistently associated with increased risks of all-cause and CV mortality. No significant interactions were found between GA or HbA1c and mortality. A U-shaped association with mortality was evident for HbA1c and remained consistent across subgroup analyses stratified by age, sex, race, BMI, smoking status, alcohol use, hypertension, diabetes mellitus, anemia, albumin levels, ACR, and eGFR.
(Online Tables S4–S7).
Sensitivity analyses
In the sensitivity analysis, after participants with less than 2 years of follow-up and those under 30 years of age were removed to reduce the potential impact of genetic diseases, GA levels continued to be significantly associated with increased risks of all-cause and CV mortality, whereas HbA1c levels demonstrated a U-shaped relationship with mortality outcomes (Online Tables S8 & S9). When further adjusted for antidiabetic agents, CKD-related agents, and BMI, the associations between baseline glycemic markers and mortality remained consistent (Online Table S10). In the landmark analyses, survival curves showed clear separation across GA tertiles for both all-cause mortality and CV mortality, supporting the stability of these findings over time (Online Figure S3).
Comparison of the associations of GA and HbA1c with mortality
Figure 3 compares the performance of GA and HbA1c in detecting all-cause and CV mortality in CKD patients. As expected, GA had higher C-index values than HbA1c in the associations with both all-cause (C-index: 0.60 (95% CI 0.59–0.62) vs. 0.55 (95% CI 0.53–0.57), p < 0.001; NRI, 27% (95% CI 6%−41%); IDI, 0.1% (95% CI 0.1%−0.3%)) and CV mortality (C-index: 0.62 (95% CI 0.60–0.64) vs.
0.56 (95% CI 0.53–0.58), p < 0.001; NRI, 43% (95% CI 14%−55%); IDI, 0.1% (95% CI 0.1%−0.2%)). This discrimination between GA and HbA1c persisted across different follow-up periods, including 3 years, 5 years, and 7 years (Fig. 4).
This pattern was supported by the confusion-matrix plots (Online Figure S4), which show that GA-based classification generally had higher sensitivity, whereas HbA1c-based classification showed higher specificity for both all-cause and CV mortality across prespecified subgroups (Online Table S11). The time-varying discrimination between GA and HbA1c for all-cause and CV mortality was evaluated via time-dependent C-indices (Fig. 5).
Models with GA consistently presented higher C-indices than those with HbA1c across follow-up, indicating better performance of GA in ranking risk over time.
Fig. 3Overall performance of glycated albumin and hemoglobin A1c for detecting all-cause. (A) and CV (B) mortality in chronic kidney disease patients.
CV: cardiovascular
Fig. 4Comparison of time-dependent ROC curves of glycated albumin and hemoglobin A1c in predicting all-cause. (A) and cardiovascular (B) mortality at different follow-up periods CV: cardiovascular, GA: glycated albumin; HbA1c: hemoglobin A1c
Fig. 5Time-dependent C-index for all-cause and cardiovascular mortality based on models including glycated albumin and hemoglobin A1c. CV: cardiovascular, GA: glycated albumin; Hb: hemoglobin A1c
Furthermore, GA and HbA1c were incorporated into the existing risk models (including the Framingham Heart Age and Framingham CVD risk score) as additional parameters to evaluate the predictive ability of the combined models for mortality outcomes. Compared with HbA1c, the addition of GA was associated with a higher C-index and significant improvements in the NRI and IDI. (Framingham Heart Age + GA: C-index, 0.68; Framingham Heart Age + HbA1c: C-index, 0.67, P = 0.04; NRI 12% (95% CI −3%−24%); IDI 1.3% (95% CI 0.5%−2.3%); Framingham CVD Risk Score + GA: C-index, 0.70; Framingham CVD Risk Score + HbA1c: C-index, 0.69, P = 0.04; NRI, 20% (95% CI 1%−36%); IDI, 0.3% (95% CI −3%−0.6%)).
Similar patterns were observed for CV mortality (Framingham Heart Age + GA: C-index, 0.69; Framingham Heart Age + HbA1c: C-index, 0.68, P = 0.04; NRI, 2% (95% CI −20%−33%); IDI, 0.8% (95% CI 0.2%−1.9%); Framingham CVD Risk Score + GA: C-index, 0.70; Framingham CVD Risk Score + HbA1c: C-index, 0.69, P = 0.04; NRI, 17% (95% CI 8%−36%); IDI, 0.3% (95% CI −0.4%−1.7%) (Table 4). In line with these model-based results, adding the GA to the clinical models (eGFR and ACR) modestly increased the C-indices compared with those of the models without the GA (Table S12).
Table 4C-index, NRI, and IDI for the comparison of the predictive performance of glycated albumin and hemoglobin A1c for all-cause and cardiovascular mortality in patients with chronic kidney diseaseC-index, p valueNRI (95% CI)IDI (95% CI)All-cause mortalityHbA1c0.55,RefRefGA0.60, p<0.00127% (6% to 41%)0.1% (0.1% to 0.3%)HbA1c + age, sex, race0.62,RefRefGA + age, sex, race0.63, p<0.00115% (5% to 43%)0.2% (0.1% to 0.3%)HbA1c + Framingham CVD Risk Score0.69,RefRefGA + Framingham CVD Risk Score0.70, p=0.0420% (1% to 36%)0.3% (-3% to 0.6%)HbA1c + Framingham Heart Age0.67RefRefGA + Framingham Heart Age0.68, p=0.0412% (-3% to 24%)1.3% (0.5% to 2.3%)CV mortalityHbA1c0.56RefRefGA0.62, p<0.00143% (14% to 55%)0.1% (0.1% to 0.2%)HbA1c + age, sex, race0.63RefRefGA + age, sex, race0.64, p =0.00115% (7% to 54%)0.1% (-0.1% to 0.2%)HbA1c + Framingham CVD Risk Score0.69RefRefGA + Framingham CVD Risk Score0.70, p=0.0417% (8% to 36%)0.3% (-0.4% to 1.7%)HbA1c + Framingham Heart Age0.68RefRefGA + Framingham Heart Age0.69, p=0.042% (-20% to 33%)0.8% (0.2% to 1.9%) Framingham Heart Age and Framingham CVD Risk Score are predictive models derived from the Framingham Heart Study to estimate 10-year cardiovascular disease risk. Heart age reflects the age of an individual with the same predicted CV risk but optimal risk factors, whereas the CVD Risk Score assigns numerical values to key risk factors to estimate CVD likelihoodIDI, integrated discrimination improvement; NRI, net reclassification improvement; GA: Glycated Albumin; HbA1c: Hemoglobin A1c; Ref: reference
According to the subgroup analyses, stronger associations for GA compared with HbA1c persisted across proteinuria categories and serum albumin levels, and this trend persisted regardless of anemia status (Online Figures S5–S7). In early CKD patients, GA exhibited stronger associations with outcomes over HbA1c (all-cause mortality: C-index: 0.61 (95% CI 0.59–0.62) vs. 0.55 (95% CI 0.54–0.57), P < 0.001; CV mortality: C-index: 0.62 (95% CI 0.59–0.64) vs.
0.57 (95% CI 0.54–0.59), P < 0.001), whereas no notable difference was found in advanced CKD patients (all P > 0.05). (Online Figure S8). Consistently, in analyses based on the KDIGO prognosis risk classification, this superiority was also significant among participants in the moderate-risk and high-risk categories for all-cause mortality (P < 0.001; P = 0.018).
(Online Figure S9)
Discussion
To the best of our knowledge, this is the most comprehensive analysis to investigate the associations of GA and HbA1c with mortality in a nationally representative CKD cohort and the first systematic evaluation of these associations across all CKD stages and under multiple confounding-adjusted models. Our study revealed that GA was consistently associated with increased all-cause and CV mortality in CKD patients. The associations with all-cause mortality and CV mortality were linear for GA and U-shaped for HbA1c, with the lowest risk at an HbA1c of approximately 5.4% and a higher risk at both lower and higher HbA1c levels in CKD patients.
GA showed stronger statistical associations with all-cause and CV mortality than HbA1c among CKD patients, particularly in the early stages. These patterns were also observed in the subgroups of CKD patients with anemia, proteinuria or hypoalbuminemia. These findings suggest that GA exhibit a closer association with mortality than HbA1c in patients with CKD and may provide complementary prognostic information.
The prognostic associations of GA have been compared with those of HbA1c in previous studies. Specifically, Junichi Hoshino et al. reported that the associations with one-year mortality were linear or J-shaped for GA and U-shaped for HbA1c in diabetic hemodialysis (HD) patients in Japan [34]. In addition, GA has been reported to be more strongly associated with mortality than HbA1c in other cohorts of dialysis patients from different countries [13–18].
As expected, our study demonstrated that GA was linearly correlated with all-cause and CV mortality in CKD patients, whereas HbA1c exhibited a U-shaped pattern, with the lowest risk at an HbA1c of approximately 5.4% and a higher risk at both lower and higher HbA1c levels. Similarly, in a subpopulation involving CKD patients in a previous study evaluating the associations of GA and HbA1c with mortality among a cohort of 12,915 adults from the NHANES 1999–2004 by Mary R. Rooney et al., mortality risk increased with increased HbA1c or GA, and low HbA1c was associated with elevated mortality risk, whereas low GA was not [35].
However, in another cohort study including 3,110 CKD patients from 7 US clinical centers, there was a J-shaped association between GA levels and the risk of all-cause mortality, with increased risks observed at both low and high GA levels [29]. This discrepancy may be attributed to the greater proportion of patients with CKD stages 4–5 (26.3% vs. 4.8%) and the greater percentage of individuals with comorbidities than in our study.
A low level of GA in these patients was linked to an increased risk of hypoglycemia, leading to a higher risk of mortality. Second, the disparities in the proportion of each race/ethnic group and observation period probably contributed to the different results.
In our analyses, GA demonstrated higher C-index values for both all-cause and CV mortality than HbA1c, particularly in early-stage CKD. Moreover, the addition of GA improved the prognostic performance of the existing risk models and standard CKD risk markers for both outcomes. This improvement is meaningful given the very large global CKD population, estimated at 697.5 million (95% uncertainty interval 649.2–752.1) cases [1].
There are several potential shared mechanisms. One might speculate that the half-life of GA is shorter than that of HbA1c. Therefore, GA could flexibly represent glycemic variation.
Another possible explanation is that GA, as an extracellularly advanced glycation end product, is also an oxidative and proinflammatory marker that can evoke atherogenesis independently of the presence of DM [36, 37]. Third, our results suggested that the associations between GA and outcomes remained stronger than those for HbA1c in the setting of anemia, which was the most common complication in CKD patients, and that the superiority was also not influenced by proteinuria or hypoalbuminemia.
In advanced CKD patients, no notable differences were observed in the associations with outcomes between GA and HbA1c. Consistently, when individuals were stratified by KDIGO prognosis risk categories, these differences remained nonsignificant among individuals in the very high-risk group. We suggest several reasons for this phenomenon.
First, advanced CKD patients already have several uremia-related complications, including anemia, hypertension, mineral and bone disorder (MBD), inflammation, protein energy wasting (PEW), heart failure, psychological problems, and sleep disorders [38–43]. In addition, there are many traditional risk factors, other than glycemic control, that may have more significant effects on CV complications and survival in advanced CKD patients [44–46]. Therefore, the direct influence of glycemic level on death might be decreased in these patients.
Second, there is insufficient statistical power because of the small number of advanced CKD patients, limiting our capacity to compare GA with HbA1c in this subpopulation. Larger-scale studies are needed to refine and advance this exploration.
In our study, we performed stratified analyses to clarify the associations of GA with outcomes by confounding factors. In detail, previous studies have indicated that obesity is negatively associated with GA, which may be attributable mainly to the effects of obesity on the serum albumin concentration and that of chronic inflammation on serum albumin turnover [47]. As expected, our study revealed that GA associations with mortality were evident and remained consistent across all the subgroup analyses stratified by BMI, and no significant interactions were found (the details are shown in Online Tables S5, S7).
Additionally, GA levels were consistently associated with increased risks of all-cause mortality across all the racial groups. However, relationships with CV mortality existed in White subjects only. The possible explanations are the racial disparity in awareness, treatment and progression of CKD, as well as the ethnic variation in albumin turnover and other biological factors [48–50].
Moreover, only a minority had a BMI < 25 kg/m², whereas the majority had a BMI ≥ 25 kg/m² (709 and 1,590, respectively). Similarly, the numbers of Black and Mexican participants were relatively small (439 and 481, respectively). The present study was underpowered to detect the effect of BMI or race on GA associations.
Therefore, further studies are needed to confirm our results in the future.
The findings of this study should be considered in the context of potential limitations. First, the GA measurement was carried out once at baseline, whereas treatments, glycemic control, CKD stage, and comorbidities likely changed over the median 12.46 years. The incorporation of longitudinal assessments of GA levels may be more representative [11].
Second, previous studies have demonstrated racial differences in HbA1c levels at a given mean glucose concentration [51, 52]. Thus, the findings of this study may have limited generalizability to other ethnic groups. Third, information on medication use was obtained from self-reported household interviews.
Although we adjusted for antihypertensive, lipid-lowering, antidiabetic and CKD-related therapies, detailed information on treatment adherence, duration and dosage was unavailable, which might introduce potential misclassification and residual confounding due to incomplete treatment data. Fourth, some unconventional confounders of GA, such as hypothyroidism, hepatic cirrhosis, inflammation, protein-energy wasting and malnutrition, and those of HbA1c, including hemoglobinopathies and other conditions, which might lead to variations in GA and HbA1c levels, were not considered. Further studies should incorporate comprehensive control of confounding factors to clarify the associations of GA and HbA1c with outcomes.
Fifth, we did not model a formal GA×HbA1c interaction. HbA1c reflects 2–3-month glycemia, whereas GA captures 2–3-week exposure and is sensitive to albumin turnover. The joint pattern may provide additional risk information beyond either alone.
Future studies should include both biomarkers with an interaction to determine whether joint modeling improves risk stratification. Finally, the sample size of the present study was relatively small, especially in some subgroups, such as Black and Mexican participants, underweight or normal weight subjects, and patients with CV events, which might limit the statistical power and reduce the strength of the conclusions of the present study. Large-sample trials investigating the relationship between GA and outcomes are needed to verify our findings.
Nevertheless, the strengths of a nationally representative sample, extended follow-up duration, detailed clinical data, and rigorous sensitivity analyses support the robustness of our results.
In conclusion, this study revealed that GA showed stronger associations with all-cause and CV mortality than HbA1c in patients with CKD, particularly in the early stages. However, this advantage was not evident in advanced CKD patients in the current sample size. Therefore, future prospective, multicenter studies with larger numbers of advanced CKD patients are warranted to determine whether GA provides better prognostic performance.
Supplementary Information
Supplementary Material 1