NHANES is a national survey which utilizes a complex, multistage probability sampling design to analyze the nutritional and health conditions of civilians in the United States every two years. Home-interviews were conducted and participants were invited to the mobile examination center (MEC) for various physiologic examinations and blood tests. The NHANES study was authorized by the National Center for Health Statistics Research Ethics Review Board (Hyattsville, MD, USA). Informed consent was signed by all participants. Detailed descriptions of NHANES and guidance on analytical approaches are available at https://www.cdc.gov/nchs/nhanes/index.htm.

Data from NHANES during 1999–2004 (N = 31,126) were analyzed in this cohort study. We excluded pregnant participants (N = 968), those without CVD morbidity or mortality outcome information (N = 16,034), or those without cystatin C or creatinine data (N = 12,985). Since serum creatinine and cystatin C levels are influenced by renal function [18], we also excluded participants with estimated glomerular filtration rate (eGFR) 30 mL/min/1.73 m${}^2$ or kidney failure (N = 510). Overall, 19,976 participants were excluded from our study. The analysis consisted of 11,150 participants, as shown in Fig. 1.

Fig. 1.

Flow chart of the study. NHANES, National Health and Nutrition Examination Survey; eGFR, estimated glomerular filtration rate; CVD, cardiovascular disease; N, number.

Blood samples were taken, centrifuged, and then stored at –20 °C prior to analysis. Serum creatinine was measured using the kinetic Jaffe method. Regression models have been developed based on the National Center for Health Statistics to adjust for serum creatinine values in NHANES 1999–2000 [19]. Surveys conducted in 2001–2002 and 2003–2004 did not need any corrections. A Cystatin C immunoassay (Siemens Healthcare Diagnostics) was used to measure Cystatin C in serum.

CVDs were determined by self-reported physician diagnosis through a home interview using a standardized questionnaire. “Has a doctor or other health expert ever told you that you have heart failure (HF)/coronary heart disease (CHD)/angina /myocardial infarction (MI)/a stroke?” the researchers asked the participants. Participants were considered to have CVD if they answered “yes” to one of the questions. A probabilistic match was used to derive death outcomes for each participant by linking to the National Death Index (NDI) until December 31, 2019. The cause of death was determined using code of the International Classification of Diseases (ICD-10). Cardiovascular mortality included death due to cardiac disease (I00–I09, I11, I13, and I20–I51) and cerebrovascular disease (I60–I69). The follow-up time was counted from the date of home interview to the date of death or December 31, 2019, whichever came first. Further details can be found at https://www.cdc.gov/nchs/data-linkage/mortality-public.htm.

Demographic data (age, gender, and ethnicity) were acquired during home interviews. Alcohol consumption and smoking were obtained via self-reporting. Diagnosis of hypertension included self-reported medical diagnosis, the use of blood pressure medications, or measured blood pressure of $\ge$140/90 mmHg. The definition of diabetes included self-reported medical diagnosis, any application of antidiabetic treatment, or a glycated hemoglobin A1c (HbA1c) level of more than 6.5%. Hyperlipidemia was defined as a history of hypercholesterolemia, the use of lipid-lowering agents or or low-density lipoprotein cholesterol (LDL-C) $\ge$4.14 mmol/L. Body mass index (BMI) was calculated as the weight in kilograms divided by the square of the height in metres. Appendicular lean mass (ALM) was refered to the combined fat-free mass of all limbers. ALM and ALM divided by BMI were representative of muscle mass on the basis of the Foundation for the National Institutes of Health [20].

Laboratory results, including baseline biochemical tests, lipid profiles, and HbA1c, were measured during the MEC examination in line with standard protocols. The LDL-C was calculated using the Friedewald equation [21]. The equation of Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) was applied to calculate the estimated glomerular filtration rate (eGFR) [22].

Given the complex survey design, proper sample weights were employed during all analyses according to NHANES analytic guidelines. Continuous variables with normal distribution were expressed as mean $\pm$ standard deviation (SD) while those with a skewed distribution were shown as median (interquartile range, IQR). Categorical variables were manifested as numbers (percentage, %). Chi-square test was utilized to analyze categorical variables as well as Kruskal-Wallis test for continuous variables. In addition, we performed univariable linear regression analysis to display the relationship between the Cr/CysC ratio and ALM or ALM divided by BMI.

The relationship between the Cr/CysC ratios and total CVD or individual CVD types was investigated via univariable and multivariable logistic regression models. We adjusted the multivariable model for age, gender, ethnicity, eGFR, as well as traditional cardiovascular risk factors (smoking, BMI, systolic blood pressure (SBP), LDL-C, and HbA1c).

Cumulative incidence was estimated with Kaplan–Meier analyses and log-rank tests. The Cox proportional hazard model for CVD mortality was adjusted for sex, age, ethnicity, SBP, LDL-C, BMI, HbA1c, smoking, eGFR, as well as HF, angina pectoris, MI, CHD, and stroke at baseline. The Cr/CysC ratio was modeled as a continuous variable or categorical groups.

We performed subgroup analysis according to age, gender, race, smoking, CVD, BMI, hypertension, diabetes, eGFR, and hyperlipidemia. We used the same variables for adjustment as in the Cox proportional hazard models. Potential interactions were also tested.

The “survey” package (version 4.1-1) for R statistical software (version 4.1.1, R Foundation for Statistical Computing, Vienna, Austria) was applied for all statistical analysis. All statistical tests were two-tailed. A p-value less than 0.05 was regarded as statistically significant.

In Table 1, we present the characteristics of the participants. Among all participants, 49% were men, and the average age was 44. The CVD morbidity in participants with higher Cr/CysC ratios was lower than that in participants with lower Cr/CysC ratios [3.3% in quartile 4 (Q4, $\ge$1.26) vs. 14% in quartile 1 (Q1, $\le$0.89)]. Significant differences in uric acid, cholesterol, BMI, HbA1c, ethnicity, diabetes, hypertension, hyperlipidemia, BMI, and smoking were found among participants with different quartiles of Cr/CysC ratios (all p 0.05).

In addition, the Cr/CysC ratio was positively related with muscle mass index, including ALM and ALM to BMI (Table 2, p 0.001).

The association between the Cr/CysC ratio and the incidence of CVD was exhibited in Table 3. In the univariable model, the odds ratio (OR) of the Cr/CysC ratio for total CVDs was 0.50 (95% CI: 0.45–0.56, p 0.001). After controlling for demographic variables, traditional cardiovascular risk factors and eGFR, the correlation remained significant (OR: 0.65, 95% CI: 0.52–0.81, p 0.001). We further analyzed the relationship between the Cr/CysC ratio and CVD morbidity by gender, age, malignancy and inflammation, and obtained the similar results (Supplementary Tables 1,2,3,4). According to the univariable model, significant associations were found between the Cr/CysC ratios and individual types of CVDs, including HF, CHD, angina, MI, and stroke (all p 0.01). However, after adjusting for all variables, a significant correlation was only found between the Cr/CysC ratio and HF, angina, MI, and stroke (all p 0.05).

During a median follow-up of 16.9 years (173,344 person-years), 997 records (8.94%) of CVD death were documented. As is shown in the Kaplan-Meier curves, participants with the Cr/CysC ratio $\ge$1.26 suffered from lower CVD mortality (Fig. 2). The trends were the similar for participants of different ages and genders (Supplementary Fig. 1 and Supplementary Fig. 2). Compared with participants whose Cr/CysC ratio $\le$0.89, the multivariable-adjusted hazard ratio (HR) (95% CI) for those whose Cr/CysC ratio $\ge$1.26 was 0.37 (95% CI: 0.26–0.54) for CVD mortality. A higher Cr/CysC ratio (per standard deviation) was associated with decreasing risk of CVD mortality (adjusted HR: 0.54, 95% CI: 0.46–0.65, p 0.001, per SD increase) (Table 4). Whether suffering from inflammation or malignancy, the relationship between the Cr/CysC ratio and CVD mortality was significant (Supplementary Table 5).

Fig. 2.

Cumulative incidence of CVD mortality. Kaplan-Meier curves were depicted according to Cr/CysC ratio quartiles. Log-rank p 0.001. CVD, cardiovascular disease; Cr, creatinine; CysC, cystatin C.

Fig. 3 shows a stratified analysis of the Cr/CysC ratio and CVD mortality risks. Results of stratified analyses were almost consistent with those from the overall analysis, except for Mexican Americans and other ethnicities. Significant interactions were found in diabetes, gender, smoking, and baseline CVD. The Cr/CysC ratio had a greater impact on cardiovascular mortality among females, diabetics, non-smokers, non-CVDs and non-Hispanic black participants.

Fig. 3.

Subgroup analysis of the association of Cr/CysC ratio (per 1 SD) with CVD mortality. N, number; HR, hazard ratio; CVD, cardiovascular disease; BMI, body mass index; eGFR, estimated glomerular filtration rate; Cr, creatinine; CysC, cystatin C.