The research employed data obtained from the NHANES spanning 2007 to 2018, which constitutes a nationally representative study administered collaboratively by the Centers for Disease Control and Prevention (CDC) and National Center for Health Statistics (NCHS). NHANES implements a complex sampling methodology involving multiple stages and stratification, gathering comprehensive health information from both adult and pediatric populations across the United States biennially. All investigative procedures received approval from the NCHS Research Ethics Review Board and strictly complied with Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting standards.

Participants aged 30 to 79 years were considered eligible for inclusion, consistent with the applicability of the PREVENT equations for cardiovascular risk estimation. Participants were excluded from the study if they did not have adequate data to determine their CKM syndrome stage, possessed incomplete dietary records necessary for computing DI-GM scores, or were diagnosed with CKM syndrome stage 4. Following these exclusion criteria, the final study cohort comprised 7884 individuals (Fig. 1). The NCHS Research Ethics Review Board granted approval for all NHANES study protocols, with each participant providing written consent prior to enrollment. This investigation adhered to the ethical principles outlined in the Declaration of Helsinki. Given that NHANES datasets are anonymized and accessible to the public, no further institutional review board authorization was deemed necessary.

Fig. 1.

Flowchart of study population inclusion and exclusion. NHANES, National Health and Nutrition Examination Survey; CKM, cardiovascular-kidney-metabolic.

CKM syndrome stages (0–4) were defined based on criteria from Aggarwal et al. [3] and adapted for NHANES data using the classification system from Tang et al. [19] (detailed information listed in Supplementary Table 1) [20]. In brief, CKM syndrome stages were categorized as follows:

Stage 0: No health risk factors related to CKM syndrome.

Stage 1: Dysfunctional adiposity.

Stage 2: With metabolic risk factors (hypertension, diabetes, and dyslipidemia), or chronic kidney disease (CKD).

Stage 3: Subclinical CVD on top of stage 2 criteria (10-year CVD risk $\ge$20%).

Stage 4: Established CVD.

The assessment of CKM syndrome progression stages employed Predicting Risk of Events via Estimated Cardiac Trajectories (PREVENT) equations [21], a validated tool designed for estimating 10-year cardiovascular disease risk in the U.S. adult population aged 30–79 years. Chronic kidney disease staging was determined using the race-neutral creatinine-based equation developed by the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI, 2021).

The calculation of DI-GM scores utilized 24-hour dietary recall information obtained from NHANES, incorporating 14 distinct dietary elements according to the scoring methodology established by Kase et al. [12] (refer to Supplementary Table 2). Each study subject provided dietary data through two separate 24-hour recall sessions conducted on non-consecutive days: the initial assessment occurred during face-to-face interviews at the Mobile Examination Center, followed by a subsequent telephone-based recall conducted 3–10 days afterward. To minimize individual variability and obtain more representative dietary patterns, the present analysis employed averaged consumption values derived from both recall sessions for DI-GM computation. This dietary index comprises 14 components classified as either advantageous or detrimental to gut microbiota health. Positive dietary elements encompassed avocado, broccoli, chickpeas, coffee, cranberries, fermented dairy products, dietary fiber, soy products, and whole grains (though green tea consumption data were unavailable in NHANES). Negative components consisted of processed meats, red meats, refined grain products, and high-fat dietary patterns (defined as $\ge$40% of total energy intake).

Scoring methodology assigned 1 point when beneficial food consumption exceeded gender-specific median values or when unfavorable food intake fell below median levels. The comprehensive DI-GM scale spanned from 0 to 13 points, with beneficial components contributing 0–9 points and unfavorable components accounting for 0–4 points. Study participants were subsequently stratified into quartile groups according to their total scores and beneficial component scores.

The selection of covariates was guided by existing research evidence, clinical importance, and statistical associations. These variables encompassed demographic characteristics (age, gender, racial background), socioeconomic indicators (educational attainment, marital situation, poverty-to-income ratio), lifestyle factors (tobacco use, alcohol intake), anthropometric measurements (body mass index), CKM syndrome severity, and biochemical parameters (serum uric acid levels and leukocyte counts). The poverty income ratio was calculated as household earnings relative to federal poverty guidelines and stratified into three categories: below poverty level (1), moderate income (1–3), and higher income ($\ge$5). Body mass index was derived from the formula of body weight in kilograms divided by the square of height in meters. The systemic inflammation index was determined through the equation: (platelet concentration $\times$ neutrophil count)/lymphocyte concentration [22]. Ethnic classification followed NHANES protocols, including Mexican American, other Hispanic populations, non-Hispanic Caucasian, non-Hispanic African American, and other racial groups. Educational background was dichotomized into incomplete secondary education versus secondary education completion or higher. Marital status was classified as either partnered (married/cohabiting) or unpartnered. Comorbid conditions were identified using established diagnostic criteria. Hypertension was diagnosed based on elevated blood pressure readings (systolic $\ge$140 mmHg or diastolic $\ge$90 mmHg), clinician-confirmed diagnosis, or antihypertensive medication use. Diabetes mellitus was defined by fasting glucose levels $\ge$126 mg/dL, glycated hemoglobin $\ge$6.5%, physician-documented diagnosis, or hypoglycemic drug therapy. Dyslipidemia criteria included total cholesterol $\ge$240 mg/dL, triglycerides $\ge$200 mg/dL, physician diagnosis, or lipid-modifying treatment. Chronic kidney disease was identified when the estimated glomerular filtration rate fell below 60 mL/min/1.73 m² or the urinary albumin-to-creatinine ratio exceeded 30 mg/g, calculated using the race-neutral CKD-EPI (2021) formula.

The study’s principal endpoint encompassed mortality from any cause, while cardiovascular-related deaths constituted the secondary endpoint. Vital status and specific causes of death were determined by cross-referencing NHANES participants with publicly available mortality records from the National Death Index (NDI), with follow-up data extending until December 31, 2019. Deaths attributed to cardiovascular causes were classified according to International Classification of Diseases (ICD)-10 codes I00–I09, I11, I13, and I20–I51. Follow-up duration for each participant was calculated from their initial examination date until either their death or the study’s termination date, whichever occurred earlier.

The statistical analyses incorporated NHANES’s sophisticated multistage probability sampling framework through the application of proper sampling weights, stratification variables, and primary sampling units, executed in R software (V4.4.2, R Foundation for Statistical Computing, Vienna, Austria) utilizing the survey package. In compliance with NHANES analytical protocols, descriptive statistics were presented as mean values with SE for normally distributed parameters, median values with interquartile range (IQR) for non-normally distributed variables, and frequency counts with weighted proportions for categorical measures. Comparative analyses between groups employed one-way ANOVA for normally distributed continuous data, Kruskal-Wallis tests for skewed continuous variables, and chi-square tests for categorical data comparisons.

The analysis of survival outcomes utilized Kaplan-Meier plots to evaluate variations in mortality across different DI-GM quartiles. Survey-adjusted Cox proportional hazards regression models were employed to accommodate NHANES’ intricate sampling methodology, generating hazard ratios (HRs) and corresponding 95% confidence intervals (CIs) for both all-cause and cardiovascular-related deaths. Covariates were chosen for model adjustment based on their statistical importance, clinical significance, and existing research evidence. To prevent excessive model adjustment, clinical parameters closely associated with CKM staging (including lipid profiles, blood pressure measurements, and glucose levels) were excluded, consistent with previous investigations [19, 20]. Due to significant data gaps in NHANES, factors such as exercise levels, caloric consumption, and pharmaceutical treatments (including antihypertensive, antidiabetic, and cholesterol-lowering medications) were omitted from the multivariate analyses. Three distinct analytical models were implemented:

Model 1: Unadjusted (crude).

Model 2: Adjusted for age, sex, race, education level, marital status, and poverty income ratio (PIR).

Model 3: Further adjusted for smoking status, drinking status, CKM syndrome stage, body mass index (BMI), white blood cell count, and uric acid levels.

To examine potential linear and non-linear relationships between DI-GM scores (including total, beneficial, and unfavorable components) and mortality outcomes, restricted cubic spline (RCS) modeling was implemented with four predetermined knot positions. These knots were strategically positioned at the 5th, 35th, 65th, and 95th percentile values of the DI-GM score distribution [23]. The mediating effect of SII on the association between DI-GM scores and all-cause mortality was investigated through mediation analysis performed with the R software’s “mediation” package. Additional stratified multivariate regression models were employed to conduct subgroup analyses. To ensure robustness of findings, sensitivity analyses were carried out by excluding individuals who experienced mortality events during the initial 24-month follow-up period.

Statistical processing was performed utilizing R software (version 4.4.2), with a predefined threshold of p 0.05 for determining statistical significance.

The research involved 7884 American participants aged between 30 and 79 years, showing a median age of 50 (interquartile range: 40–61), with females constituting 52.97% (n = 4182) of the sample (Table 1). The ethnic distribution comprised 7.67% Mexican Americans, 5.49% other Hispanic individuals, 68.87% non-Hispanic Caucasians, 10.61% non-Hispanic African Americans, and 7.36% representing other ethnicities. When analyzing participants based on quartiles of the DI-GM composite score (Table 1), individuals with elevated scores tended to be male, of non-Hispanic White descent, more educated, and with greater household earnings. No notable age variations were found among the groups (p = 0.282). Elevated DI-GM scores also correlated with increased alcohol intake, reduced incidence of diabetes, and chronic kidney disease. Nevertheless, no meaningful relationship emerged between DI-GM scores and CKM syndrome progression (p = 0.181). Conversely, when examining DI-GM beneficial score quartiles (Supplementary Table 3), higher values were connected with decreased occurrence of advanced CKM syndrome stages (Stage 2 and 3, p = 0.004). Furthermore, greater DI-GM total scores corresponded with higher body mass index, whereas increased beneficial scores were associated with reduced leukocyte counts and serum uric acid concentrations.

During a median observation period of 77 months, 469 deaths from any cause (4.56%) and 105 cardiovascular-related fatalities (1.02%) were documented. Survival analysis using Kaplan-Meier curves revealed that individuals in the bottom quartiles for both total DI-GM (Q1: scores 0–2) and beneficial component scores (Q1: scores 0–1) exhibited the greatest likelihood of all-cause mortality (log-rank p 0.001, Fig. 2). The association between adverse scores and overall mortality risk was not statistically significant (log-rank p = 0.062). Regarding cardiovascular mortality, only the beneficial component demonstrated a meaningful correlation (log-rank p = 0.017), with participants in the top quartile (Q4: scores $\ge$4) displaying the most favorable outcomes.

Fig. 2.

Kaplan-Meier survival curves for the DI-GM scores and mortality outcomes. Participants were categorized into quartiles (Q1–Q4) based on the DI-GM total, beneficial, and unfavorable scores, with Q1 representing the lowest and Q4 the highest dietary index scores. Survival probabilities over time are shown according to: (A) DI-GM total score, (B) DI-GM beneficial score, and (C) DI-GM unfavorable score for all-cause mortality; and (D) DI-GM total score, (E) DI-GM beneficial score, and (F) DI-GM unfavorable score for cardiovascular mortality. Log-rank tests were performed to assess differences in survival across quartile groups.

The results from Cox regression analysis (Fig. 3) demonstrated that every unit increment in the DI-GM total score correlated with a 13% decrease in overall mortality risk (HR = 0.87, 95% CI: 0.80–0.93), while no significant relationship was observed with cardiovascular mortality. Similarly, a one-point rise in the beneficial score corresponded to a 17% reduction in all-cause mortality (HR = 0.83, 95% CI: 0.75–0.92) and a 20% decrease in cardiovascular mortality risk (HR = 0.80, 95% CI: 0.72–0.89). When analyzed by quartiles, participants in the highest quartile of both DI-GM total and beneficial scores exhibited substantially lower all-cause mortality rates compared to the lowest quartile (Q4 vs Q1 HR = 0.48 [95% CI: 0.35–0.67] and 0.45 [95% CI: 0.31–0.65], respectively). Conversely, elevated unfavorable scores were linked to greater all-cause mortality risk (Q4 vs Q1 HR = 1.58, 95% CI: 1.12–2.24), but showed no significant association with cardiovascular mortality.

Fig. 3.

Cox regression analysis of DI-GM scores and mortality risks. Associations between DI-GM total, beneficial, and unfavorable scores and mortality outcomes. HRs and 95% CIs are presented for all-cause (A) and cardiovascular (B) mortality. DI-GM scores were analyzed both as categorical variables (quartiles, with Q1 as the reference group) and as continuous variables (per 1-point increase in score).

The findings from the multivariable Cox regression analysis (Model 3) demonstrated that both the overall DI-GM score and its beneficial component exhibited independent correlations with decreased mortality from all causes. Notably, every unit increment in the total score corresponded to a 10% decline in all-cause mortality risk (HR = 0.90, 95% CI: 0.82–0.98), whereas the beneficial component showed an even stronger association with a 12% risk reduction (HR = 0.88, 95% CI: 0.79–0.98; Table 2). In quartile-based comparisons, individuals positioned in the top quartile (Q4) for both the total and beneficial DI-GM scores displayed substantially lower all-cause mortality rates relative to those in the bottom quartile (Q1), with HRs of 0.57 (95% CI: 0.40–0.83) and 0.60 (95% CI: 0.41–0.87), respectively. Regarding cardiovascular mortality outcomes, only the beneficial score maintained statistical significance following comprehensive adjustment (HR = 0.87, 95% CI: 0.77–0.99; Table 3), while the relationship with the total score weakened and lost significance. Importantly, the unfavorable DI-GM score component failed to show any meaningful association with either all-cause or cardiovascular mortality across all adjusted models. When examining quartile distributions, no statistically significant variations in cardiovascular mortality risk were detected between the highest (Q4) and lowest (Q1) quartiles for any DI-GM score components after complete covariate adjustment.

RCS (Fig. 4) revealed a U-shaped relationship between DI-GM total score and all-cause mortality (p for non-linearity = 0.008), a linear association for beneficial scores (p for non-linearity = 0.076), and no significant pattern for unfavorable scores. For cardiovascular mortality (Supplementary Fig. 1), the DI-GM beneficial score showed an inverted U-shaped association (p for non-linearity = 0.031), while total and unfavorable scores showed no significant trends.

Fig. 4.

Restricted cubic spline (RCS) analysis results for the association between DI-GM scores and all-cause mortality. The associations between DI-GM scores and all-cause mortality outcomes were analyzed using RCS models, adjusted for covariates in Model 3. The relationships between DI-GM total score (A), beneficial score (B), unfavorable score (C) and all-cause mortality were evaluated. The figure displays the HR (solid lines) with 95% CI (shaded areas).

The overall and favorable DI-GM scores showed inverse correlations with SII, demonstrating $\beta$ values of –7.35 (95% CI: –12.46 to –2.24) and –12.86 (95% CI: –19.64 to –6.07), respectively (Table 4). Mediation assessments indicated that SII accounted for 5.29% (95% CI: 1.44%–10%, p = 0.024) of the total DI-GM score’s impact and 8.45% (95% CI: 2.12%–18%, p = 0.016) of the beneficial DI-GM score’s influence on mortality from all causes (Fig. 5). These findings imply that diets promoting gut microbiota health might lower mortality risk in part by decreasing systemic inflammatory responses.

Fig. 5.

Mediation analysis examining the role of SII in the relationship between DI-GM scores and mortality from all causes. The study performed mediation analyses to assess whether systemic inflammation index (SII) mediated the link between DI-GM composite score (A), positive component score (B) and overall mortality. These analyses incorporated adjustments for the same covariates used in Model 3, with the exception of leukocyte count.

Analyses conducted across various demographic subgroups, including age, gender, ethnicity, body mass index, and stages of CKM syndrome, demonstrated stable relationships between overall DI-GM scores or favorable dietary indices and mortality risks (all interaction p-values exceeded 0.05; refer to Table 5 and Supplementary Table 4). When participants who died within the initial 24 months of follow-up were excluded from consideration, the beneficial DI-GM score maintained its significant correlation with cardiovascular-related deaths (multivariable-adjusted HR = 0.87, 95% confidence interval: 0.77–0.99; see Supplementary Table 5), while its association with overall mortality showed reduced significance (fully adjusted HR = 0.92, 95% CI: 0.81–1.03; Supplementary Table 5).

The research underscores the significant influence of dietary habits on mortality outcomes, particularly for patients in CKM syndrome stages 0–3, a population group where dietary intervention studies remain scarce. Utilizing comprehensive NHANES data enhances the validity and applicability of these findings across diverse demographic groups in the United States. Furthermore, through distinct evaluation of both positive and negative elements within the DI-GM scoring system, our analysis indicates that consuming gut microbiota-friendly foods may serve as a key factor in lowering mortality rates, potentially mediated by their impact on inflammatory responses throughout the body.

Several important limitations warrant consideration in this study. Firstly, the cross-sectional nature of the research design restricts our capacity to establish causal relationships between DI-GM scores and mortality results. Future investigations employing longitudinal approaches would be essential to verify these observed connections and assess the sustained impacts of dietary habits. Secondly, the nutritional data collection relied on participant-reported 24-hour dietary recalls, a methodology potentially vulnerable to memory biases and incomplete reporting, which might compromise the precision of DI-GM computations. Thirdly, despite comprehensive adjustments for numerous sociodemographic characteristics, lifestyle variables, and clinical parameters, the possibility of unmeasured confounding factors cannot be entirely ruled out. In particular, physical activity, total energy intake, and medication use (antihypertensive, antidiabetic, and lipid-lowering drugs) were not incorporated into the models due to a large proportion of missing data, which may have limited our ability to fully account for these influences. Fourth, although NHANES data are nationally representative, findings may not be directly applicable to populations outside the US due to differences in dietary behaviors, cultural practices, and genetic backgrounds. Finally, as with all survival analyses, censoring could have affected the precision of the survival estimates. However, since censoring patterns were comparable across quartiles, the likelihood of systematic bias is minimized.