The NHANES study is carried out every two years by the National Center for Health Statistics (NCHS) to estimate the nutritional and health conditions of the entire population of the US employing intricate and multi-level probability sampling design methods.

CKD is characterized by an estimated glomerular filtration rate (eGFR) < 60 mL/min/1.73 m2 and/or urinary albumin to creatinine ratio (UACR) ≥ 30 mg/g, as per the KDIGO 2021 guideline [13]. The eGFR was computed through the CKD-Epidemiology Collaboration equation, which includes serum creatinine, age, sex, and race/ethnicity [14]. The calculation for eGFR was as follows: 141 × min (SCr/κ, 1)α × max (SCr/κ, 1)−1.209 × 0.993Age × 1.018 if female × 1.159 if Black, where SCr represents serum creatinine, κ is 0.7 for females and 0.9 for males, α is − 0.329 for females and − 0.411 for males, min denotes the lesser value between SCr/κ or 1, and max signifies the greater value between SCr/κ or 1. In the present analysis, we initially identified CKD participants aged 20 and above from a series of 10 consecutive cycles spanning from 1999 to 2018. Afterward, we removed pregnant females and individuals who did not have 24-h dietary recalls and follow-up information, which led to a total of 8725 participants in the final sample size (Supplementary Fig. 1).

Given possible variables that could affect the results, this study incorporated several covariates to clarify any uncertainty in the correlation between the consumption of dietary live microbes and mortality. Demographic parameters, such as age, gender, race/ethnicity (including Black, White, Mexican, and others), body mass index (BMI), educational attainment (ranging from below high school to above high school), and poverty income ratio (classified as < 1.3, 1.3–3.5, and ≥ 3.5), were collected through questionnaires. BMI was determined through anthropometric measurements. The participants self-reported their habits and health conditions, encompassing smoking habits (never, former, current), alcohol intake (none, moderate, heavy, and binge), levels of physical activity (none, moderate, and vigorous), hypertension, diabetes, hyperlipidemia, and CVD. The diagnosis of diabetes met one of four different criteria: self-reported diagnosis, the use of diabetes medications, a hemoglobin A1c ≥ 6.5%, or a fasting plasma glucose ≥ 126 mg/dL. The definition of hyperlipidemia was based on three distinct criteria: (1) the utilization of lipid-lowering drugs; (2) triglyceride levels equal to or greater than 150 mg/dL; (3) total cholesterol levels equal to or greater than 200 mg/dL, with high-density lipoprotein cholesterol below 40 mg/dL for males or below 50 mg/dl for females, or low-density lipoprotein cholesterol levels equal to or greater than 130 mg/dL. Hypertension was determined through the use of antihypertensive drugs or systolic/diastolic blood pressure ≥ 140/90 mmHg. CVD was ascertained by self-reported presence of one of five subtypes from among congestive heart failure, coronary heart disease, angina, heart attack, or stroke. Additional covariate indicators included Healthy Eating Index-2015 (HEI-2015) and laboratory data, including UACR and serum creatinine. The HEI-2015 score ranges between 0 and 100 points. A higher score reflects healthier eating.

The Public Use Linked Mortality File was used to determine the outcomes of the participants including data on their survival status from the National Death Index until December 31st, 2019. The primary cause of death was identified following ICD-10 codes, encompassing deaths caused by CVD (I00-I09, I11, I13, I20-I51, and I60-I69) and other causes.

Dietary data are collected using an in-person 24-h dietary recall. NHANES dietary data were obtained from the What We Eat in America (WWEIA) program conducted by the United States Department of Agriculture (USDA), which aims to provide the nutrient values for foods and beverages reported in WWEIA (https://www.ars.usda.gov/northeast-area/beltsville-md-bhnrc/beltsville-human-nutrition-research-center/food-surveys-research-group/). The method for determining dietary live microbes was performed as before [10], and has been employed in multiple studies [15, 16]. In short, to estimate the number of live microbes per gram of food corresponding to the 9388 food codes from 48 subgroups in the NHANES database, four experts (Maria L. Marco, Mary E. Sanders, Robert Hutkins, and Colin Hill) conducted evaluations of each food item. Their assessments were based on a comprehensive review of existing literature, authoritative reviews, and the known impacts of food processing techniques, such as pasteurization, on microbial viability. The differences in each food classification were established through internal team consultations and validated by external consultation with Fred Breidt, a microbiologist at the USDA Agricultural Research Service. Based on the expected number of live microbes, the food was classified as low (< 104 CFU/g), medium (104–107 CFU/g), or high (> 107 CFU/g). Foods with low microbial content primarily consisted of pasteurized products, foods with moderate microbial content mainly included unpeeled fresh fruits and vegetables, and foods with high microbial content included unpasteurized fermented foods. MedHi refers to food classified as having moderate or high microbial content, and we calculated the grams of food classified as MedHi for each individual. The classification of each food item was documented in the prior study [10].

Following the NHANES analysis guidelines, we employed sampling weights (WTDRD1) and masked variance in R 4.2.2 to account for the intricate study design of NHANES. We examined the characteristics of individuals with and without all-cause mortality using Student’s t-tests for continuous factors and chi-square tests for categorical factors. In regression analysis, our primary focus was to investigate the correlation between exposure to MedHi dietary intake and the rate of mortality. Thus, we conducted an examination of MedHi as both a continuous and categorical variable. The categorization involved dividing the participants into three groups based on tertile distribution: Tertile 1 referred to individuals who did not consume any live microbe foods classified as Med and Hi, the Tertile 2 group included individuals whose consumption of MedHi foods was in the range of 0–110 g/d, and the Tertile 3 group consisted of individuals with consumption exceeding 110 g/d. Weighted multivariate Cox regression analysis was adopted to examine the relationship between MedHi dietary live microbe consumption and mortality across three different models. In Model I, adjustments were made for age, gender, and race/ethnicity. Model II incorporated age, gender, race/ethnicity, educational attainment, poverty income ratio, BMI, smoking habits, alcohol consumption, and physical activity levels, while Model III further incorporated health conditions (hypertension, diabetes, hyperlipidemia, and CVD), serum creatinine, UACR and HEI-2015. Restricted cubic splines (RCS) with four knots based on multivariate regression analysis were implemented to visually represent the linear or nonlinear relationship between MedHi dietary active microbe intake and mortality rate. Stratified analyses were implemented to verify the robustness of our findings based on gender, age, hyperlipidemia, hypertension, diabetes, and CVD. Additionally, we implemented several sensitivity analyses. Firstly, since prebiotic/probiotic supplements were not included in the dietary live microbes, we further adjusted for Prebiotic/Probiotic supplements (only information concerning 4659 participants was available). Secondly, given that unpasteurized foods contain a large amount of vitamins, we further adjusted for dietary factors including vitamin A, vitamin C, vitamin E, and carotenoids. Thirdly, to reduce the risk of reverse causality bias, the analysis excluded individuals who died within a 2-year time frame. Fourthly, survival times were censored at the latest at 15 years to ensure consistency. Finally, individuals with implausible energy intake levels falling below 500 kcal or exceeding 4000 kcal per day were excluded from the analysis. A p < 0.05 showed statistical significance.