The NHANES is a set of cross-sectional surveys with a complicated sampling strategy intended to assess the health and dietary status of the non-institutionalized US population across all age groups (30). Each participant was monitored until death. This cross-sectional cohort investigation included outcomes from the 1999–2018 NHANES cycles.

The initial sample included 101,316 participants from the NHANES fasting subsample (1999–2018). After excluding 40,235 individuals aged under 20 years, 55,081 participants remained. Additional exclusions were made for those with insufficient data for diagnosing CKM syndrome (n = 29,858), those diagnosed with CKM syndrome stages 0–2 (n = 20,547), and participants missing data on ALB, neutrophil, or lymphocyte counts (n = 251). This left 4,425 participants for analysis. Additional exclusions were implemented due to absent covariate (n = 1,158) and mortality data (n = 1). The final cohort for analysis comprised 3,266 participants (Figure 1).

The NHANES protocol received approval from the Institutional Review Board for the National Center for Health Statistics (currently known as the Ethics Review Board), which permitted the release of the data files for public use. A written informed consent was provided from each participant prior to data collection. This study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) instructions for cohort investigations.

Data on ALB, neutrophil, and lymphocyte counts were collected, and the ANLR was computed using the formula previously outlined (29):

The ANLR were computed using values obtained from liver function tests and routine blood tests. Currently, no established cutoffs exist to differentiate high from low values for ANLR. To address this, we employed the maximally selected rank statistics method to identify the optimal threshold points for it (31). This statistical technique, which is based on the log-rank test, identifies the most significant cutoff in continuous variables where the disparity in survival rates is most pronounced. It is particularly well-suited for time-to-event data, such as survival outcomes. Using this data-driven approach, the identified cutoff values were 1.04 for ANLR (Figure 2).

To mitigate the risk of overfitting, we applied a Bonferroni-adjusted significance level and further performed internal validation via bootstrap resampling (n = 1,000). The Bootstrap analysis indicates that the distribution of the cutoff values of ANLR is concentrated and close to 1.04, suggesting that the result is stable. This stability reflects the minimal fluctuation of the cutoff values during repeated sampling, further validating the reliability of the model estimation. Moreover, the 95% confidence interval (CI) is ranging from 0.942 to 2.497, and the original cutoff value of 1.04 falls within this interval. This indicates that the cutoff estimate is statistically reliable. Therefore, considering both the Bootstrap distribution and the CI, the original cutoff estimate can be regarded as stable and robust (Supplementary Figure S1). Figure 2 illustrates the distribution of ANLR values and their relationship with mortality outcomes, highlighting the association between ANLR values < 1.04 and increased mortality risk. Following this, participants were stratified into two distinct groups based on the threshold value.

CKM syndrome was classified into five stages (0–4) based on modified criteria adapted for NHANES data availability, referencing the AHA definitions (1) and prior large-scale analyses (30, 32). Due to the absence of certain diagnostic tools (e.g., coronary artery calcium [CAC] scores, NT-proBNP, echocardiography, Ankle-Brachial Index [ABI]), we used proxy indicators as detailed below.

Stage 0: Participants with normal body mass index (BMI) (<25 kg/m²), waist circumference (WC) (< 88 cm for women, < 102 cm for men), and these participants will not be classified into a higher stage category (normoglycemia, normotension, normal lipid profile, no evidence of CKD or clinical/subclinical CVD).

Stage 1: Participants with elevated BMI (≥ 25 kg/m²), increased waist circumference (≥ 88 cm for women, ≥ 102 cm for men), or prediabetes (hemoglobin A1c [HbA1c]: 5.7–6.4% or fasting blood glucose [FBG]: 100–125 mg/dL), without the presence of other metabolic risk factors or CKD.

Stage 2: Participants with metabolic risk factors or moderate-to-high-risk CKD per KDIGO guidelines (33). Metabolic risk factors included elevated triglycerides (TG) (≥ 135 mg/dL), hypertension, diabetes, or metabolic syndrome (≥ 3 of the following: elevated WC, low HDL [< 40 mg/dL for men, < 50 mg/dL for women], elevated TG [≥150 mg/dL], elevated blood pressure [systolic ≥130 mmHg, diastolic ≥80 mmHg], or prediabetes).

Stage 3: Participants with very high CKD risk was characterized by either Stage G4 or G5 CKD (eGFR < 30 mL/min/1.73 m²) or a classification of very high risk based on KDIGO guidelines, determined by GFR and UACR (eGFR < 30 mL/min/1.73m² or UACR ≥ 300 mg/g) (33), or a calculated 10-year atherosclerotic cardiovascular disease risk (ASCVD) ≥ 20% [https://professional.heart.org/en/guidelines-and-statements/prevent-calculator], based on pooled cohort equations.

Stage 4: Participants with a self-reported physician diagnosis of CHD, stroke, angina, CHF, or MI. Atrial fibrillation (AF) and peripheral artery disease (PAD) were not included due to data unavailability.

Advanced CKM syndrome was defined as encompassing stages 3 and 4 of CKM syndrome, in alignment with the classification by Zhu et al. (4). The specific NHANES variables used for each criterion are listed in Supplementary Table S1.

This study evaluated two mortality outcomes: all-cause mortality (ACM) and cardiovascular mortality (CVM). ACM refers to death from any cause, while CVM specifically pertains to death attributed to CVDs. Mortality outcomes were acquired from the National Death Index (NDI) and linked to NHANES data up to December 31, 2019.

Drawing from existing literature and biological plausibility, a comprehensive set of covariates known to influence outcomes in advanced CKM syndrome was considered. These included demographic factors (age, sex, and race), socioeconomic predictors (marital status, educational attainment, and the family poverty-income ratio [PIR]), lifestyle and health habits (BMI, smoking habits, alcohol intake), and comorbidities such as hypertension, diabetes, hyperlipidemia, liver condition, CHD, CHF, MI, angina, stroke, CKD, and cancer. Laboratory parameters, including neutrophil, lymphocyte, ALB, TG, high-density lipoprotein (HDL), and HbA1c, were also included. By adjusting for these factors, we sought to minimize confounding influences and enhance the robustness of our analysis on the relationship between ANLR and both CVM and ACM in individuals with advanced CKM syndrome.

Due to the retrospective design of this study, which is susceptible to selection bias and possible confounding variables, propensity score matching (PSM) was utilized to rectify these concerns and reduce their influence. A logistic regression model was utilized to produce propensity scores, and individuals were thereafter matched in a 1:1 ratio according to these scores. The covariates’ selection for the PSM procedure was informed by an extensive review of relevant literature. These covariates included demographic factors, socioeconomic indicators, lifestyle factors, clinical characteristics, and laboratory parameters. This rigorous selection process ensured the inclusion of scientifically relevant and contextually appropriate covariates.

To minimize discrepancies between matched pairs, the closest neighbor matching algorithm was applied with a caliper width of 0.1. The effectiveness of the PSM in balancing baseline characteristics between groups was evaluated by calculating the absolute standardized differences (ASDs). ASD values below 0.10 following matching indicated that the potential biases and confounding factors had been effectively addressed, confirming a fair comparison between the groups for subsequent analyses.

We applied methods to account for masked variance and used a weighting approach (34). Continuous variables were expressed as means with interquartile ranges (IQR) and compared via the t-test or Mann–Whitney U-test based on normality evaluated by the Shapiro–Wilk test. Categorical variables were reported as counts and proportions, with comparisons made using Fisher’s exact test or the Chi-square test. The ideal ANLR cutoff for predicting ACM and CVM was determined using maximally selected rank statistics, set at 1.04 to optimize the risk ratio.

To validate the proportional hazards (PH) assumption, we utilized both graphical and statistical methods. The statistical assessment was performed using Schoenfeld residuals and the Grambsch-Therneau test, while visual validation was carried out via Kaplan–Meier (KM) survival curves. The Cox PH model was used to account for censored data, considering individuals who had no experience of the event during the investigation as non-events. The analysis followed each participant’s timeline from the baseline assessment to either death or the study endpoint.

The Cox PH model identified significant predictors of mortality through both univariate and multivariate analyses. Outcomes are presented as hazard ratios (HRs) with 95% confidence intervals (CIs). Model 1 was unadjusted; Model 2 was adjusted for age, sex, and race; Model 3 involved additional adjustments for education level, family PIR, tobacco consumption, alcohol intake, and BMI. Disease variables that are integral to CKM syndrome diagnosis—such as CHD, stroke, CKD, diabetes, hypertension, and hyperlipidemia—were excluded to avoid overadjustment and collider bias. Subgroup analyses were carried out to explore ANLR influence on mortality outcomes across different participant groups and comorbid conditions, using a multivariate Cox PH model. Factors considered included age (categorized into 20–59 years and ≥ 60 years), gender, race/ethnicity, hypertension, diabetes, hyperlipidemia, stroke, and CKM stages. Additionally, ANLR was divided into tertiles to examine its relationship with ACM and CVM, with comparisons made between the highest tertile (reference) and other groups.

Restricted cubic splines (RCS) were employed for adaptable curve fitting to examine possible non-linear associations. Generalized additive models were utilized to evaluate the impact of ANLR on mortality. The placement and quantity of knots were established based on previous research and preliminary analyses, specifically in the 10th, 50th, and 90th percentiles of the ANLR distribution, which were corroborated through first-model diagnostics. Statistical significance was deemed for two-sided testing at p < 0.05. Data analysis was conducted utilizing SPSS Statistics 26 and R statistical software (version 4.2.2) to guarantee thorough assessment.

The baseline characteristics and findings of individuals prior to PSM are summarized in Table 1. The total sample size was 3,266, with participants classified into two groups according to the ideal ANLR cutoff: low ANLR (440 participants) and high ANLR (2,826 participants). Internal validation via 1,000 bootstrap iterations yielded a median optimal cutoff of 1.04, suggesting that the data-driven threshold is relatively stable. The frequency distribution of bootstrap-derived cutoffs is shown in Supplementary Figure S1. The low ANLR group had a significantly higher median age (76.5 vs. 72.0 years) and a lower proportion of males (37.05% vs. 41.26%) compared to the high ANLR group. Significant variations were noticed between the groups in age, ethnicity, marital status, tobacco consumption, hyperlipidemia, stroke, CKD, CKM stage, and cancer, while no significant differences were found for gender, education level, family PIR, alcohol consumption, hypertension, diabetes, MI, angina, liver conditions, and BMI. Physiological markers indicated a significant elevation in neutrophil concentrations in the low ANLR group, while the high ANLR group demonstrated higher levels of ALB, lymphocytes, and TG. No significant differences were noted in HDL or HbA1c levels. Clinically, the low ANLR group exhibited significantly higher rates of CVM (26.14% vs. 17.34%, p < 0.001) and ACM (68.41% vs. 47.17%, p < 0.001) compared to the high ANLR group.

Table 2 presents the outcomes of both univariate and multivariate Cox PH analyses for CVM and ACM before PSM across three models of adjustment. Regarding CVM, participants with an ANLR score < 1.04 exhibited a significantly raised risk compared to those with higher ANLR scores, with HRs of 2.23 (Model 1), 1.80 (Model 2), and 1.81 (Model 3), all with p-values < 0.001. Analysis by tertiles further demonstrated that the lowest tertile (T1) consistently showed a significantly higher risk than the highest tertile (T3), with HRs of 2.06, 1.66, and 1.65 across the three models (all p < 0.001). For ACM, an ANLR score < 1.04 was also associated with increased risk, with HRs of 2.14 (Model 1), 1.75 (Model 2), and 1.77 (Model 3), all p < 0.001. The T1 group exhibited HRs of 1.94, 1.58, and 1.58 across the three models, all with p-values < 0.001.

KM survival curves further validated the differences in mortality outcomes between the low and high ANLR score groups, demonstrating that the lower ANLR group experienced significantly higher CVM and ACM rates compared to those with higher ANLR scores. The survival percentages showed significant disparities: 26.14% vs. 17.34% for CVM (p < 0.001) and 68.41% vs. 47.17% for ACM (p < 0.001). These findings are visually represented in Figures 3A,C.

To explore potential non-linear associations, RCS was employed to fit smooth curves using generalized additive models. This analysis aimed to identify the threshold effect of ANLR on CVM and ACM before PSM, highlighting any inflection points. The results revealed an L-shaped non-linear correlation between ANLR and both CVM and ACM pre-PSM (all p-values for non-linearity < 0.001, Figure 4A).

The application of a 1:1 PSM method aimed at minimizing baseline disparities between the low and high ANLR groups resulted in the successful pairing of 404 patient pairs (Table 3). Figure 5 illustrates the efficacy of the PSM process, as assessed by the measurement of ASD in both pre- and post-PSM. Following the matching procedure, the median age across the two groups became comparable. The proportion of males also became more similar, with 37.05% in the low ANLR group and 42.27% in the high ANLR group. However, ethnic disparities remained to some degree, with a higher percentage of non-Hispanic White in the low ANLR group (74.55% vs. 57.05%) and a higher proportion of non-Hispanic Black in the high ANLR group (18.64% vs. 9.54%). The groups did not differ significantly in terms of marital status, family PIR, educational attainment, alcohol consumption, smoking status, or BMI following matching. Regarding comorbidities, only MI (p = 0.02) and cancer (p = 0.003) exhibited significant variation between the groups. Physiological measures showed a notable increase in neutrophil levels in the low ANLR group, while the high ANLR group exhibited higher ALB and lymphocyte counts. No significant differences were observed in TG, HDL, or HbA1c levels. Clinically, individuals in the low ANLR group still had significantly higher rates of CVM (26.14% vs. 16.14%, p < 0.001) and ACM (68.41% vs. 54.55%, p < 0.001).

Table 4 summarizes the outcomes of univariable and multivariable Cox PH regression analyses for CVM and ACM following PSM, using three levels of model adjustment. Regarding CVM, participants with an ANLR score below 1.04 were found to have a significantly higher risk of mortality compared to those with higher ANLR values. In the unadjusted model (Model 1), the HR was 2.07 (95% CI: 1.54–2.78, p < 0.001). This elevated risk persisted after adjustment for age, gender, and race/ethnicity in Model 2, with an HR of 1.99 (95% CI: 1.47–2.69, p < 0.001), and remained significant in Model 3, which was additionally adjusted for educational level, family PIR, tobacco consumption, alcohol use, hypertension, diabetes, hyperlipidemia, stroke, and BMI, yielding an HR of 2.02 (95% CI: 1.49–2.75, p < 0.001). Similarly, for ACM, an ANLR score below 1.04 was related to a significantly higher death risk. In Model 1, the HR was 1.60 (95% CI: 1.35–1.90, p < 0.001). This association remained robust subsequent to the adjustment of confounders in Models 2 and 3, with HRs of 1.52 (95% CI: 1.28–1.81, p < 0.001) and 1.52 (95% CI: 1.27–1.81, p < 0.001), respectively. KM survival curves further established the significantly lower survival rates among participants with an ANLR score below 1.04 compared to those with an ANLR score of 1.04 or higher (Figures 3B,D). RCS revealed an L-shaped non-linear association between ANLR and both CVM and ACM post-PSM (all p-values for non-linearity < 0.005, Figure 4B).

Subgroup analyses were carried out to examine the relationship between ANLR scores and CVM and ACM across different categories, including age, sex, race/ethnicity, hypertension, diabetes, hyperlipidemia, smoke, drinking, and CKM stages (Figure 6). The outcomes revealed a significant inverse correlation between ANLR scores and CVM among participants with advanced CKM syndrome in all subgroups, except for those in the 20–59 years age group (p = 0.50). Moreover, ANLR and the various subgroup factors did not interact significantly (all p > 0.05). Regarding ACM, a similar negative relationship was detected between ANLR scores and mortality in all subgroups of advanced CKM syndrome. However, significant interactions were noted between ANLR and age, gender, and CKM stages (p < 0.05).

An assessment of the predictive value of ANLR, NLR, and ALB for CVM and ACM in individuals with advanced CKM syndrome revealed that the area under the ROC curve (AUC) values for ANLR and NLR for ACM were 0.602 and 0.601, respectively, while the AUC for both ANLR and NLR for CVM was 0.565. These results indicate that although both indices provide some prognostic value, their diagnostic performance is moderate, and they cannot be considered highly effective indicators. ANLR and NLR exhibited comparable predictive capabilities for both ACM and CVM, outperforming ALB (Supplementary Figure S2). Moreover, the predictive performance of ANLR and NLR for ACM was slightly superior to that for CVM (0.602 vs. 0.565).