Lower CALLY index linked to higher overall and heart-related death risk in chronic kidney disease, partly due to faster biological aging

The National Health and Nutrition Examination Survey (NHANES) is a nationally representative study that collects health and nutrition data on the U.S. household population. Data were collected through a multistage probability sampling design, including structured household interviews, physical examinations at ambulatory centers, and laboratory tests. The NHANES protocol was approved by the Institutional Review Board (IRB) of the National Center for Health Statistics (NCHS), and all participants provided written informed consent. A total of 7,564 individuals with CKD were identified from the NHANES 1999–2010 dataset. We excluded 2,206 individuals due to missing albumin, lymphocyte, or C-reactive protein values required for the CALLY formula. Additionally, 802 individuals were excluded due to missing survival status. Finally, to ensure result reliability, 41 pregnant women were excluded. In total, 4,515 participants were included in the final analysis. The data screening process is illustrated in Figure 1. For the mediator BioAgeAccel, 175 participants had missing values. These individuals were excluded only from the mediation analysis, as their data were otherwise complete for the main analyses. This approach ensured that exclusion due to mediator missingness did not introduce unnecessary selection bias into the primary analyses. Missing values for other covariates were imputed using the random forest method. All data used in this study are publicly available at: https://www.cdc.gov/nchs/nhanes/?CDC_AAref_Val=https://www.cdc.gov/nchs/nhanes/index.htm↗.

CKD was defined by the diagnostic criteria of the Kidney Disease: Improving Global Outcomes (KDIGO) guidelines. The estimated glomerular filtration rate (eGFR) was calculated using the updated Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation, incorporating serum creatinine, age, sex, and race. The albumin-to-creatinine ratio (ACR), expressed in mg/g, was obtained from spot morning urine samples and used to indicate microalbuminuria. Participants were considered to have CKD if they met either eGFR < 60 mL/min/1.73 m² or ACR > 30 mg/g [17].

Blood count analysis, including neutrophils, platelets, monocytes, and lymphocytes, was performed using a Kurt HMX Hematology Analyzer, with results reported in standardized units (×10³ cells/μl). This study focuses on the CALLY Index, a novel metric derived from albumin levels, lymphocyte counts, and C-reactive protein (CRP) values. The CALLY Index is calculated as [(albumin, g/L) × (lymphocyte count, 10³ cells/μl)]/(CRP, mg/dL) and offers a comprehensive assessment of inflammatory, nutritional, and immune status.

All-cause mortality was determined using National Death Index (NDI) records linked to the NHANES dataset, with follow-up through December 31, 2019. Cause-specific mortality was classified based on ICD-10 codes. Cardiovascular disease (CVD)-related deaths were defined by ICD-10 codes I00-I09, I11, I13, and I20-I51.The dataset is available at:https://ftp.cdc.gov/pub/Health_Statistics/NCHS/datalinkage/linked_mortality/↗.

Biological age, as proposed by Klemera and Doubal [18], is derived from eight biomarkers: Ln-C reactive protein (CRP), serum creatinine, glycosylated hemoglobin, serum albumin, total cholesterol, blood urea nitrogen, alkaline phosphatase, and systolic blood pressure. The formula is given in Supplementary documents. BioAgeAccel is the value obtained by subtracting the actual age from the biological age.

Covariates included age, sex, race, education, marital status, poverty-to-income ratio (PIR), body mass index (BMI), smoking status, alcohol consumption, moderate activity, vigorous activity, diabetes, hypertension, hyperlipidemia, cardiovascular disease (CVD), estimated glomerular filtration rate (eGFR), alanine aminotransferase (ALT), aspartate aminotransferase (AST) and uric acid. Physical activity was classified as vigorous or moderate (yes/no). Participants who had smoked at least 100 cigarettes in their lifetime and were still smoking at the time of the survey were classified as current smokers. Ex-smokers were those who had smoked at least 100 cigarettes in their lifetime but had quit before the study. Individuals who had smoked fewer than 100 cigarettes in their lifetime were classified as nonsmokers. Nondrinkers were participants who had never consumed at least 12 alcoholic beverages in their lifetime or any single year. Participants who had consumed at least 12 alcoholic drinks in their lifetime or any single year but had abstained in the past 12 months were classified as ever drinkers. Current drinkers were those who had consumed at least 12 alcoholic beverages in their lifetime or any single year and had consumed alcohol in the past 12 months. Diabetes was defined as physician-diagnosed diabetes, a fasting blood glucose level ≥7.0 mmol/L, glycosylated hemoglobin ≥6.5%, a 75 g oral glucose tolerance test result ≥200 mg/dL, or glucose-lowering medication or insulin. Hypertension was defined as a history of hypertension, use of antihypertensive medication, or systolic blood pressure ≥130 mm Hg or diastolic blood pressure ≥80 mm Hg. Hyperlipidemia was defined as triglycerides ≥150 mg/dL, total cholesterol ≥200 mg/dL, low - density lipoprotein cholesterol (LDL-C) ≥ 130 mg/dL, high-density lipoprotein cholesterol (HDL-C) ≤ 40 mg/dL in men or ≤50 mg/dL in women, or use of lipid-lowering therapy. CVD was defined as a history of congestive heart failure, coronary artery disease, angina, stroke, or myocardial infarction.

Missing covariate data were handled using a random‑forest–based imputation procedure (missForest), which accommodates mixed variable types and nonlinear relations [19]. The proportion of missingness for covariates was low (uric acid 0.02%, AST 0.42%, ALT 0.42%, BMI 2.79%, CVD 3.61%, smoking 3.70%, and PIR 8.82%); accordingly, single imputation with missForest was used for the main analyses [20,21]. This choice is supported by prior work showing strong performance of random‑forest imputation relative to common alternatives in biomedical datasets [22]. uACR was not imputed in the primary models because it is part of the KDIGO definition of CKD, and imputing it could lead to misclassification of disease status. Therefore, uACR was used only in sensitivity analyses among participants with available measurements. This strategy is consistent with current practice and acknowledges the risk of bias when imputing disease-defining variables [23,24]. Baseline characteristics were summarized as mean ± standard deviation (SD) for normally distributed continuous variables, median with interquartile range (IQR) for skewed continuous variables, and proportions for categorical variables. Between-group comparisons were conducted across tertiles of Ln-CALLY, using one-way analysis of variance (ANOVA) for normally distributed continuous variables, the Kruskal–Wallis test for skewed continuous variables, and the Chi-square test for categorical variables. Because the CALLY index shows marked right-skewness, especially driven by the CRP component, it was transformed using the natural logarithm (Ln-CALLY) prior to all regression analyses. This transformation reduces influence of extreme CRP values, stabilizes variance, and improves model fit and plausibility of log-linear assumptions in Cox regression, consistent with methods employed in prior studies using CALLY and similar composite biomarker indices [25–27]. Multivariate Cox regression assessed the association between CALLY and all-cause and CVD mortality. Three models were constructed: Model 1 (unadjusted); Model 2, adjusted for age, sex, and race; and Model 3, further adjusted for education, marital status, PIR, BMI, smoking status, alcohol consumption, moderate activity, vigorous activity, diabetes, hypertension, hyperlipidemia, CVD, eGFR, ALT, AST and uric acid. Subgroup analyses assessed the association between CALLY and all-cause mortality across different subgroups. We performed a Kaplan-Meier survival analysis to map the association of different Ln-CALLY groups with all-cause mortality and CVD mortality. Smoothed curve fitting and threshold effect analyses were performed to explore the nonlinear relationship between CALLY and all-cause or CVD mortality. Mediation analyses were conducted to examine the potential role of BioAgeAccel in mediating the association between Ln-CALLY and all-cause or CVD mortality, adjusting for covariates in Model 3, as per VanderWeele's framework [28]. Previous studies have suggested that the atherogenic index of plasma (AIP) and the systemic immune-inflammation index (SII) are potential predictors of mortality in patients with CKD [29,30]. To further evaluate the predictive utility of CALLY, we compared its discrimination ability with AIP and SII for all-cause and cardiovascular mortality. Time-dependent ROC curves were constructed at 5-, 10-, and 15-year follow-up using the R package timeROC. The area under the curve (AUC) with 95% confidence intervals was calculated to quantify predictive performance. For the primary analyses, missing covariates were imputed using a random forest approach. Considering that uACR is a key variable in the definition of CKD and thus not suitable for imputation, we included it only in sensitivity analyses. Specifically, we repeated the Cox regression models with additional adjustment for uACR to assess the robustness of the findings. To reduce the potential influence of reverse causation, we excluded participants who died within the first two years of follow-up and reassessed associations using Cox models. In addition, to minimize the potential influence of extreme values, we conducted a sensitivity analysis by excluding participants in the lowest and highest 1% of the CALLY distribution. All statistical analyses were performed using R version 4.2.0 (The R Foundation for Statistical Computing, Vienna, Austria; http://www.R-project.org↗), and a two-tailed P value < 0.05 was considered statistically significant.

The manuscript was edited for readability, style, and grammatical accuracy using Grammarly, an artificial intelligence–based tool for non-generative copyediting.

Table 1 presents the baseline characteristics of participants stratified by Ln-CALLY tertiles. No significant differences were found among the three groups in sex, marital status, and albuminuria (p > 0.05). However, significant differences were observed in BMI, diabetes mellitus, hypertension, race, smoking, and alcohol consumption (p < 0.05). The lower CALLY group had a higher proportion of individuals with eGFR < 60 mL/min/1.73 m², BMI ≥ 30, and a greater prevalence of diabetes, hypertension, and CVD.

As shown in Table 2, higher Ln-CALLY was significantly associated with a reduced risk of all-cause mortality in Model 1 (HR = 0.818, 95% CI: 0.793–0.844). The association remained consistent after multivariable adjustment in Model 2 (HR = 0.827, 95% CI: 0.801–0.855) and Model 3 (HR = 0.845, 95% CI: 0.817–0.874). Similarly, for cardiovascular mortality, a significant inverse association was observed in Model 1 (HR = 0.825, 95% CI: 0.780–0.872), Model 2 (HR = 0.839, 95% CI: 0.790–0.890), and Model 3 (HR = 0.860, 95% CI: 0.809–0.915). When participants were stratified into tertiles of Ln-CALLY, those in the third tertile had a 37.4% lower risk of all-cause mortality and a 34.0% lower risk of cardiovascular mortality compared with the first tertile. A significant linear trend was consistently observed across tertiles (P for trend < 0.001).

The Kaplan-Meier curve (Figure 2) supports these findings. The curve illustrates that CKD patients with lower Ln-CALLY levels had a greater decline in survival over follow-up compared to those with higher Ln-CALLY levels.

As shown in Table 3 and Figure 3, nonlinear associations were observed between Ln-CALLY and both all-cause and cardiovascular mortality among individuals with CKD. Using a standard linear model, higher Ln-CALLY was inversely associated with all-cause mortality (HR = 0.845, 95% CI: 0.817–0.874, p < 0.001) and cardiovascular mortality (HR = 0.860, 95% CI: 0.809–0.915, p < 0.001). A two-piecewise linear regression model further identified an inflection point at Ln-CALLY = 4.96 (CALLY = 142.59). Below this threshold, Ln-CALLY was strongly associated with decreased risk of all-cause mortality (HR = 0.750, 95% CI: 0.698–0.806, p < 0.001) and cardiovascular mortality (HR = 0.742, 95% CI: 0.650–0.848, p < 0.001). In contrast, above the threshold, the associations were weaker for all-cause mortality (HR = 0.912, 95% CI: 0.864–0.962, p < 0.001) and became non-significant for cardiovascular mortality (HR = 0.942, 95% CI: 0.856–1.036, p = 0.219). Log-likelihood ratio tests supported the nonlinear relationship for both outcomes (p < 0.001 for all-cause mortality; p = 0.019 for cardiovascular mortality).

Adjusted for age, sex, race, education, marital status, PIR, body mass index, smoking, drinking, moderate activity, vigorous activity, diabetes, hypertension, hyperlipidemia, cardiovascular disease, eGFR, ALT, AST and uric acid.

To evaluate the robustness of the findings, subgroup analyses were conducted across multiple strata. As shown in Table 4, the inverse associations of Ln-CALLY with all-cause and cardiovascular mortality were generally consistent across sex, marital status, PIR, BMI, smoking, drinking, diabetes, hypertension, hyperlipidemia, CVD, physical activity, and eGFR. Although not all subgroup estimates reached statistical significance, no significant interactions were detected (P for interaction > 0.05).

As shown in Supplementary Table 1↗, Ln-CALLY was inversely associated with BioAgeAccel across all three models. In the fully adjusted model (Model 3), the regression coefficient was β = −3.364 (95% CI: −3.750, −2.979, p < 0.001). Supplementary Table 2↗ further demonstrated that BioAgeAccel was positively asso­ciated with both all-cause mortality (HR = 1.010, 95% CI: 1.008–1.012, p < 0.001) and CVD mortality (HR = 1.014, 95% CI: 1.010–1.018, p < 0.001).

Mediation analyses (Figure 4 and Table 5) revealed that BioAgeAccel partially mediated the associations of Ln-CALLY with mortality outcomes. Specifically, BioAgeAccel explained 14.618% (95% CI: 0.091–0.219, p < 0.001) of the association between Ln-CALLY and all-cause mortality, and 27.442% (95% CI: 0.148–0.505, p < 0.001) of the association with CVD mortality. Both the mediation and direct effects remained statistically significant after adjustment for potential confounders.

As shown in Figure 5, CALLY consistently demonstrated superior predictive performance for both all-cause and cardiovascular mortality in CKD compared with AIP and SII. For all-cause mortality (Figure 5(A)), the AUCs of CALLY were 0.686 (95% CI: 0.539–0.834), 0.687 (95% CI: 0.583–0.792), and 0.667 (95% CI: 0.593–0.741) at 5, 10, and 15 years, respectively, which were higher than those of AIP and SII. Similarly, for cardiovascular mortality (Figure 5(B)), CALLY achieved AUCs of 0.620 (95% CI: 0.293–0.947), 0.606 (95% CI: 0.383–0.828), and 0.568 (95% CI: 0.432–0.739) at 5, 10, and 15 years, again outperforming AIP and SII. These findings suggest that CALLY may provide better long-term prognostic value for mortality risk stratification in CKD patients.

In the sensitivity analysis, after additional adjustment for the urine albumin-to-creatinine ratio (uACR) (Supplementary Table 3↗), the inverse association between log-transformed CALLY and all-cause and cardiovascular mortality remained essentially unchanged. Furthermore, after excluding participants who died within the first 2 years of follow-up, we conducted an additional Cox regression analysis (Supplementary Table 4↗). The inverse association remained consistent. In addition, when extreme values of CALLY were truncated in sensitivity analyses (Supplementary Table 5↗), the associations were materially unchanged, further supporting the robustness of our findings.

Not applicable.