The Difference Between Cystatin C- and Creatinine-Based Estimated Glomerular Filtration Rate and Risk of Diabetic Microvascular Complications Among Adults With Diabetes: A Population-Based Cohort Study

Diabetes is a global public health issue, with >529 million adults affected worldwide, and this number is projected to rise to 1.31 billion by 2050 (1). Diabetic microvascular complications (DMCs), including diabetic retinopathy (DR), diabetic kidney disease (DKD), and diabetic neuropathy (DN), occur in >50% of people with diabetes (2). These complications result in several adverse events, such as loss of vision, end-stage renal disease, pain, and paresthesia, leading to a deterioration in overall quality of life and imposing a substantial burden on society (3–5). With the continuous rise in diabetes prevalence globally and the improvement of life expectancy, DMCs have become the major component of disease burden (6,7). Therefore, cost-effective clinical markers to target the population at high-risk of developing DMCs at an early stage hold paramount significance in public health.

The estimated glomerular filtration rate (eGFR), usually measured by serum creatinine or cystatin C, is a widely used marker in medical practice to evaluate renal function and predict adverse prognosis in the diabetes population (8). The large intraindividual difference between cystatin C-based eGFR (eGFRcys) and creatinine-based eGFR (eGFRcr) has gradually been recognized in recent years and is concerning, and it is reported to be associated with a range of adverse events, including falls, hospital admission, kidney failure, cardiovascular events, and mortality (9–13). To our best knowledge, no empirical data are currently available regarding the associations between the difference in eGFRcys and eGFRcr (eGFRdiff) and DMCs.

To fill these knowledge gaps, we prospectively evaluated the association of eGFRdiff with risk of overall DMCs and its subtypes in patients with diabetes from the UK Biobank.

We used data from the UK Biobank, which is a population-based cohort study recruiting >0.5 million participants aged 40 to 69 years from 22 sites across England, Scotland, and Wales between March 2006 and October 2010. The study design and methods have been described in detail previously (14). The study was approved by the North West Multi-Centre Research Ethics Committee in Haydock, the National Information Governance Board for Health and Social Care in England and Wales, and the Community Health Index Advisory Group in Scotland. All participants provided written informed consent at baseline.

We included 31,642 patients with diabetes at recruitment. Diabetes was defined as those who had self-reported or doctor-diagnosed diabetes, were taking antihyperglycemic medications or using insulin, or glycosylated hemoglobin A_1c (HbA_1c) was >6.5% (48 mmol/mol) (15). Finally, 25,825 participants with diabetes were included in the main analysis after excluding participants with missing values for serum creatinine or cystatin C (n = 2,284), or who had prevalent DMCs at baseline (n = 2,069), or a baseline eGFR based on both serum creatinine and cystatin C (eGFRcr-cys) of <60 mL/min/1.73 m² (n = 1,464). The flowchart of patients included in the current study is presented in Supplementary Fig. 1.

Touch screen questionnaires were used to obtain information on sociodemographic status, lifestyle, history of medical conditions, and medication at the baseline assessment. Anthropometric measurements, including height, weight, and blood pressure (BP), were collected following standardized protocols. Hypertension was defined as a self-reported history of hypertension, or mean systolic BP (SBP) ≥140 mmHg and/or diastolic BP (DBP) ≥90 mmHg, or use of antihypertensive medication. BMI was determined as weight in kilograms divided by the square of height in meters. Muscle and fat masses were determined using bioimpedance analysis. Appendicular skeletal muscle mass was calculated as the sum of lean soft tissue in all four limbs and standardized by BMI. Biochemical markers in biological samples were all measured in a central laboratory. Serum and urine creatinine were measured by enzymatic analysis on a Beckman Coulter AU5800 and a Beckman Coulter AU5400. Serum cystatin C was measured by latex enhanced immunoturbidimetric analysis on a Siemens ADVIA 1800. Urine microalbumin was measured by immunoturbidimetric analysis on a Beckman Coulter AU5400 and standardized to the urine albumin-to-creatinine ratio (uACR). Serum lipids profile, C reactive protein (CRP), urate, and glucose were quantified using standard procedures through the Beckman Coulter AU5800. HbA_1c levels were measured by high-performance liquid chromatography analysis on a Bio Rad Laboratories VARIANT II TURBO. Smoking status was categorized as ever or never smokers. Further details can be found at the UK Biobank website (https://biobank.ctsu.ox.ac.uk/showcase).

The 2012 Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) cystatin C equation and 2021 race-free CKD-EPI equations were used to calculate eGFRcys, eGFRcr, and eGFRcr-cys, respectively (8,16). Our exposure of interest, eGFRdiff, included the absolute difference between eGFR (eGFRabdiff), defined as eGFRcys minus eGFRcr, and relative difference between eGFR (eGFRrediff), defined as eGFRcys/eGFRcr. The eGFRabdiff was further stratified by three categories: 1) < −15 mL/min/1.73 m² (negative eGFRabdiff), 2) −15 to 15 mL/min/1.73 m² (midrange eGFRabdiff), and 3) ≥15 mL/min/1.73 m² (positive eGFRabdiff), while the eGFRrediff was categorized into two groups: <0.6, and ≥0.6 (9,17).

The primary outcome of interest was incident DMCs, a composite indicator of the first occurrence of DR, DKD, or DN. Secondary outcomes included the incidence of three DMC subtypes, including DR, DKD, and DN. We defined incident DMCs according to the ICD-10 codes, which were obtained by using linkage with death registers, primary care, and hospital inpatient records. The dates and causes of hospital admissions were identified by record linkage to Hospital Episode Statistics (England and Wales) and the Scottish Morbidity Record (Scotland). The data on dates and causes of death were obtained from the death registries of the National Health Service Information Centre (England and Wales) and the National Health Service Central Register (Scotland). Detailed information regarding the linkage procedure is available online (https://biobank.ctsu.ox.ac.uk/crystal/refer.cgi?id=149480). DR (ICD-10: E103, E113, E123, E133, E143, H280, H360), DKD (ICD-10: E102, E112, E122, E132, E142, N180, N181, N182, N183, N184, N185, N188, N189), and DN (ICD-10: E104, E114, E124, E134, E144, G590, G629, G632, G990) cases were extracted from the “first occurrence of health outcomes defined by a three-character ICD-10th Revision code” (category ID in UK Biobank 1712). We calculated the follow-up time as the duration between the date of the baseline assessment and censored at the date of DMCs incidence, date of death, or 31 December 2022, whichever occurred first.

Baseline characteristics are reported as mean ± SD for normally distributed continuous variables, medians (interquartile ranges) for nonnormally distributed continuous variables, and numbers (percentages) for categorical variables. The difference between groups was compared using the t test, the ANOVA, or Kruskal-Wallis test for continuous variables, depending on data distribution, and the χ² test for categorical variables.

The eGFRdiff was analyzed separately as a continuous predictor and categorical predictor. Kaplan-Meier curves and log-rank tests were used to compare event rates among groups. Cox proportional hazard regression models were used to estimate the hazard ratios (HRs) and 95% CIs of outcomes associated with eGFRdiff. Proportional hazards assumptions were assessed using Schoenfeld residuals, and we found no significant deviation from the assumption. We fitted two statistical models. Model 1 was adjusted for age, sex, race, HbA_1c, diabetes duration, and eGFRcr. Model 2 was further adjusted for SBP, DBP, total cholesterol (TC), triglyceride (TG), LDL-cholesterol (LDL-C), HDL-cholesterol (HDL-C), BMI, CRP, urate, uACR, smoking status, hypertension, use of antidiabetes medication, use of antihypertensive medication, and use of lipid-lowering medication. To investigate dose-response associations between eGFRdiff and outcomes, we used a restricted cubic spline regression model with three knots at the 10th, 50th, and 90th percentiles of the distributions of eGFRdiff. Tests for nonlinearity were performed by using the likelihood ratio test.

All baseline covariables had <9% missing data in the UK Biobank. Therefore, we performed multiple imputations using SAS Proc MI. The fully conditional specification methods for data with an arbitrary missing pattern were used to impute missing values in both continuous variables and class variables with 20 imputations.

To test the robustness and potential variations in different subgroups, we repeated all analyses stratified by sex (men or women), race (White or other), age (<60 years or ≥60 years), hypertension (yes or no), and cardiovascular disease (CVD) history (yes or no).

We performed several sensitivity analyses. First, we repeated the analysis without using multiple imputation to fill missing data of covariates. Second, we excluded events that occurred within the first 2 years of follow-up to reduce the potential reverse causation. Third, we repeated our analyses of incident outcomes using Fine-Gray models instead, which accounted for the competing risk of death. Fourth, we repeated our analyses after excluding participants who had CVD history. Fifth, we repeated our analyses across different baseline eGFRcr stages (≥90, 60–89, 30–59). Sixth, we repeated our analyses further adjusting for eGFRcys and eGFRcr-cys instead of eGFRcr. Finally, we repeated our analyses further adjusted for muscle mass.

All statistical analysis was conducted using SAS 9.4 software (SAS Institute), and R 4.3.1 software (R Foundation for Statistical Computing). All tests were two-tailed with a statistical significance level of P < 0.05.

Data from the UK Biobank (ukbiobank.ac.uk/) are available to researchers on application to the UK Biobank following the steps outlined here: https://www.ukbiobank.ac.uk/enable-your-research.

Table 1 shows baseline characteristics of participants. Among 25,825 participants with diabetes from UK Biobank (mean age, 59.1 years; 60.4% men), 10,045 (38.9%) were of negative eGFRabdiff, 15,028 (58.2%) of midrange eGFRabdiff, and 752 (2.9%) of positive eGFRabdiff; 25,247 (97.8%) of eGFRrediff ≥0.6 and 578 (2.2%) of eGFRrediff <0.6 (Supplementary Fig. 2). The mean eGFRabdiff and mean eGFRrediff were −11.1 ± 14.1 mL/min/1.73 m² and 0.9 ± 0.2 at baseline. Participants with large negative eGFRdiff tended to be older, more were women, and had higher prevalence of ever smokers, CVD history, and hypertension. In addition, participants with negative eGFRabdiff had the highest SBP, TG, BMI, CRP, urate, and uACR among the three eGFRabdiff groups.

During a median follow-up of 13.59 (interquartile range 12.77, 14.43) years, DMCs developed in 5,753 participants, including 2,752 cases of incident DR, 3,203 incident DKD, and 1,149 incident DN. Kaplan-Meier curves showed significant differences in risk of all outcomes among all groups (Supplementary Fig. 3).

In comparison with participants with midrange eGFRabdiff, participants with negative eGFRabdiff had a multivariable adjusted HR of 1.29 (95% CI 1.22, 1.36) for overall DMCs, 1.13 (1.04, 1.23) for DR, 1.63 (1.50, 1.76) for DKD, and 1.40 (1.23, 1.58) for DN; while the corresponding HRs for participants with positive eGFRabdiff were 0.55 (0.46, 0.66), 0.76 (0.58, 0.98), 0.39 (0.31, 0.50), and 0.88 (0.57, 1.37). Linear dose-response relationships of eGFRabdiff with outcomes were demonstrated (all P for nonlinearity >0.05) except for DR (Fig. 1). Each SD decrease of eGFRabdiff was associated with a 28% higher risk of overall DMCs, 14% higher risk of DR, 56% higher risk of DKD, and 29% higher risk of DN (Table 2).

Compared with participants with eGFRrediff ≥0.6, participants with eGFRrediff <0.6 showed a higher risk of all outcomes, with HRs of 1.75 (95% CI 1.51, 2.04) for overall DMCs, 1.49 (1.17, 1.90) for DR, 2.32 (1.94, 2.79) for DKD, and 1.64 (1.19, 2.27) for DN. For each 10% increment in eGFRrediff, the corresponding HRs were 1.16 (1.14, 1.18) for overall DMCs, 1.08 (1.05, 1.11) for DR, 1.29 (1.26, 1.33) for DKD, and 1.17 (1.12, 1.22) for DN. Although nonlinear dose-response relationships of eGFRrediff with outcomes were demonstrated (all P for nonlinearity <0.05), similar dose-response association patterns were found between eGFRrediff with all outcomes.

When analyses were stratified by sex (men or women), race (White or other), age (<60 years or ≥60 years), hypertension (yes or no), and CVD history (yes or no), the associations between eGFRdiff and DMCs remained consistent (Fig. 2). Substantial interactions were identified from subgroup analyses. For example, age significantly modified the relationships of eGFRdiff with DR and DN, while significant interactions between eGFRdiff and CVD history were identified for overall DMCs and DKD.

In all sensitivity analyses, the magnitude of associations was not materially altered after using data without imputation, excluding events that occurred within the first 2 years of follow-up, accounting for competing risk by death, excluding participants with CVD history, across different baseline eGFRcr stages and further adjusting for muscle mass (Supplementary Tables 1–6). When adjusting further for eGFRcys and eGFRcr-cys instead of eGFRcr, the associations of eGFRdiff with DMCs were attenuated but still significant, except for the association between eGFRdiff and DR, which lost statistical significance when adjusting for eGFRcys (Supplementary Table 7).

In this large population-based prospective cohort of adults with diabetes, we found that baseline eGFRdiff (both eGFRabdiff and eGFRrediff) was inversely associated with risk of incident DR, DKD, DN, and the composite DMCs. Additionally, analogical dose-response relationship patterns were demonstrated between baseline eGFRdiff and future risks of DMCs. With each 1-SD decrease in the eGFRabdiff, the risk of overall and subtypes of incident DMCs elevated by 14% to 56%. Similarly, with each 10% decrease in the eGFRrediff, the risk of overall and subtypes of incident DMCs increased by 8% to 29%. Our study indicated that eGFRdiff represents an early indicator for DMCs and suggested that monitoring eGFRdiff might facilitate the risk stratification and precise management of patients with diabetes at the population level.

To date, a few studies have revealed significant association between eGFRdiff and kidney outcomes among diverse specific populations. For example, in 13,197 individuals in the Atherosclerosis Risk in Communities Study and in 158,601 adults in the Stockholm Creatinine Measurements project, large negative eGFRdiff was associated with higher risk of end-stage kidney disease and acute kidney injury (12,18). Similarly, large negative eGFRdiff has also been found to be associated with increased risk of worsening kidney function in patients with heart failure (13). To our knowledge, there are no published data exploring eGFRdiff and prognosis focusing on diabetes. Our study firstly reported that baseline eGFRdiff (both eGFRabdiff and eGFRrediff) was inversely correlated to overall DMCs and its subtypes. The associations remained robust after adjustment for common risk factors and confounders, which suggested that large negative eGFRdiff might be a unique predictor for incident DMCs. The specific pathophysiological state reflected by a large negative eGFRdiff remains incompletely understood.

Although the association between large negative eGFRdiff and poor outcomes has been reported in a recent series of studies, it is still controversial whether eGFRdiff should be applied as a routine measure due to its cost-effectiveness of clinical and public health value. In previous analyses in populations with hypertension, advanced age, and chronic kidney disease, the prevalence of eGFRabdiff lower than −15 mL/min/1.73 m² varied from 8% to 16% (9,10,19,20). Unlike other specific populations, the proportion of large negative GFRdiff was obviously higher in the diabetes population, with 38.9% of the participants exhibiting a large negative eGFRabdiff. This is of critical importance since this particular subtype was considered as an group at risk of developing adverse outcomes. Monitoring eGFRdiff among diabetes populations might further provide important prognostic information for risk stratification and early intervention. Notably, screening for cystatin C routinely has been highly recommended by the U.S. National Kidney Foundation and the American Society of Nephrology (21). Consistent with this recommendation, two recent studies demonstrated that the combined creatinine-cystatin C eGFR equation performed better than equations based on either of the two markers and reinforced the need for more frequent cystatin C measurements in clinical settings (22,23). In our study, the associations between eGFRdiff and DMCs were still robust when adjusting for eGFRcr-cys/eGFRcys, which suggests that the associations are likely to be explained by non-GFR determinants rather than more accurate measures of kidney function. Our results, in conjunction with previous information that discordant eGFRcr and eGFRcys are common, suggest eGFRdiff screening may yield additional valuable prognostic information and public health benefits.

The mechanisms underpinning the connection between eGFRdiff and risk of DMCs remain inadequately understood. In our results, participants with large negative eGFRabdiff demonstrated higher levels of inflammatory marker and BMI, more ever smokers, and higher prevalence of hypertension and cardiovascular history, consistent with previous studies (18,24). These factors are all well-documented risk factors of DMCs. In nearly 3,000 participants in the Health, Aging, And Body Composition Study, lower eGFRdiff was strongly associated with lower muscle quantity and muscle strength and poor functional status, both of which would increase the risk of DMCs (25–27). Additionally, a hypothesis proposed by Grubb et al. (17,28), termed shrunken pore syndrome, describes a pathophysiologic state of eGFRcys/eGFRcr <0.6 or <0.7 and reflects selective impairment of glomerular sieving of cystatin C and other similar molecular weight molecules (5–30 kDa). Two recent proteomic studies have revealed that certain serum proteins, such as interleukin 6, MCP-3, and osteoprotegerin, would accumulate in patients with shrunken pore syndrome (29,30). Those proteins were presumed to link to atherosclerosis, inflammation, and vascular endothelial injury, which are all important causative factors for DMCs (3–5). Further research is needed to explore the possible pathophysiology mechanisms that link eGFRdiff with the risk of DMCs.

This study is the first to investigate the association between baseline eGFRdiff and the future incident DMCs. The strengths of this study include the large sample size, the prospective study design, and a series of sensitivity analyses to confirm the robustness of the findings.

Nevertheless, we also acknowledge potential limitations. First, causal inference cannot be made because of the nature of observational studies. Second, although we adjusted for potential confounders, residual confounding cannot be entirely excluded. Third, our findings may not be directly generalizable to other populations, for most participants in our study were White British people. Fourth, our exposure relied on single baseline measures of serum creatinine and cystatin C, leading to potential misclassification bias. Finally, ICD-10 codes may be insufficient for detecting cases in an early stage or for classifying DMCs.

In this large population-based prospective study of adults with diabetes, we found that eGFRdiff was inversely associated with risk of overall DMCs and its subtypes. A considerable proportion of the diabetes population exhibits large eGFRdiff; thus, monitoring eGFRdiff in this specific population could provide public health benefits for risk stratification and clinical decisions. Future research is needed to explore the modifiable factors that contribute to large GFRdiff and to further evaluate the impact of interventions.

Acknowledgments. The authors thank all of the survey teams and all of the study participants in the UK Biobank for their contributions and dedication to this study.

Funding. This study was supported by grants from the National Key Research and Development Program of China (2022YFF1203001), National Natural Science Foundation of China (72125009, 81771938, 81900665, 82003529, 82090021), National Key R&D Program of the Ministry of Science and Technology of China (2019YFC2005000), Chinese Scientific and Technical Innovation Project 2030 (2018AAA0102100), Young Elite Scientists Sponsorship Program by China Association for Science and Technology (CAST) (2022QNRC001), Chinese Academy of Medical Sciences (CAMS) Innovation Fund for Medical Sciences (2019-I2M-5-046), and P Peking University (PKU)-Baidu Fund (2020BD004, 2020BD005).

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

Author Contributions. D.H. analyzed the data. D.H. and B.G. designed the study, interpreted the results, and drafted the manuscript. D.H., J.W., and C.Y. were involved in data collection and data cleaning. B.G., M.-H.Z., and L.Z. made critical revisions to the manuscript for important intellectual content. B.G., M.-H.Z., and L.Z. obtained funding. B.G. and L.Z. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Handling Editors. The journal editors responsible for overseeing the review of the manuscript were Elizabeth Selvin and Csaba P. Kovesdy.