This study included participants from the nationally representative consecutive National Health and Nutrition Examination Survey (NHANES) 1998–2018 linked to the National Death Index through December 31, 2019. NHANES was approved by the National Center for Health Statistics (NCHS) ethics review committee, and all participants submitted written informed consent. All procedures for this study were conducted following relevant guidelines and regulations.¹ Of the 101,316 subjects in the NHANES 1998–2018, individuals were excluded if (1) they were under 18 years of age (n = 42,112) (2), they had missing values for dietary live microbial intake (n = 6,717) (3), they had any missing values for the NHANES weight (n = 19) (4), they had missing follow-up data. (n = 85). Ultimately, a total of 52,383 subjects were included in this research (Supplementary Figure S1). This study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines (11).

The 24-h dietary data was linked to the USDA Food Survey Nutrition Database by the National Center for Health Statistics (NHANES) to evaluate nutrient and energy intakes (12–14). In Marco’ study, four experts assessed the number of live microbes in 48 food codes contained in 9388 subgroups of the NHANES database by drawing on references, authoritative reviews and the reported influence of food processing (e.g., pasteurization) on microbial viability, as well as consulting externally with USDA Agricultural Research Service microbiologist Fred Breidt. Foods with different viable microorganisms were classified as low (<10^{^4}CFU/g), medium (10^{^4}–10^{^7}CFU/g), or high (>10^{^7} CFU/g). To quantitatively measure an individual’s approximate total dietary intake of live microorganisms (CFU), the microbial intakes of the low, medium, and high groups were assumed to be 10^{^3} CFU/g, 10^{^6} CFU/g, and 10^{^9} CFU/g, respectively, and then multiplied by the corresponding grams of food in each group and summed. In addition, as the order of magnitude of individual dietary intake of live microorganisms was too large, a logarithm with a base of 10 live microbial intakes was taken for subsequent analysis. Finally, the total population was divided into 3 groups of low, medium, and high according to individual total dietary live microorganisms <10^{^7} CFU,10^{^7}–10^{^10} CFU, and > 10^{^10} CFU.

Information on mortality status (from baseline to 31 December 2019) was acquired from NCHS using a unique identification number for each participant. All-cause mortality is defined as death from any cause. Cardiovascular mortality was identified by codes I00-I09, I11, I13 and I20-I51 according to the International Classification of Diseases, Tenth Revision (ICD-10).

Demographic characteristics included age (18–39, 40–59, ≥60), sex (women, men), race (Mexican American, Non-Hispanic Black, Non-Hispanic White, Other Hispanic, other race-including multiracial), marital status (divorced/separated/widowed, married/living with a partner, never married, no record), education (<high school, high school, >high school, no record), and poverty-to-income ratio (PIR) (<1.3, 1.3–3.5, >3.5, no record), smoking status (never, former, now, no record), alcohol status (never, former, now, no record), and physical activity(no, yes, no record). BMI is calculated from measured height and weight and is classified as normal (<25kg/m²), overweight (25-30kg/m²), obese (≥30kg/m²), and no record. The estimated glomerular filtration rate (eGFR, mL/min/1.73m²) was categorized into <60 and ≥ 60 groups, which was calculated from serum creatinine measurements using the 2009 Chronic Kidney Disease Epidemiology Collaboration formula (15). Total diet quality was estimated using the 2015 version of the Healthy Eating Index (HEI) and total energy intake (16). Comorbidities included hypertension, diabetes, CVD, and cancer, of which the first two are diagnosed through index measurements, medication use, and self-reporting, while CVD and cancer are identified through self-reporting.

Individual sample weights were created using WTDR4YR weights in 1999–2002 and WTDRD1 weights in 2003–2018, given the complex sampling design of NHANES. During the baseline characteristics analysis, continuous variables were presented using weighted means (standard errors), and categorical variables were presented using sample numbers (weighted percentages). One-way ANOVA and chi-square tests were performed to compare differences according to baseline dietary intake of live microorganisms (<10^{^7} CFU,10^{^7}–10^{^10} CFU, and > 10^{^10} CFU).

Weighted multivariable Cox regression analysis was used to estimate corrected risk ratios (HR) for mortality and their 95%CI based on dietary live microbial intake in the low, medium, and high groups at baseline. The crude model did not adjust for any potential confounders. Model 1 adjusted for age, sex, race, education, marital status, and PIR. Model 2 was further adjusted for HEI, total energy intake, smoking status, alcohol status, physical activity, and BMI. Model 3 additionally adjusted for hypertension, diabetes, CVD, cancer, and eGFR. We carried out survival analyses with the use of standardized Kaplan–Meier curves and Log ranking tests. The non-linear relationship between mortality and log(dietary live microbial intake) was explored by multivariable-adjusted Cox restricted cubic spline (RCS) regression models. RCS regression models were used to explore the potential non-linear relationship between TDIIM and mortality outcomes, allowing for a more flexible assessment of the association by fitting a smooth curve to the data. Subgroup analyses were undertaken by sex (female, male), age (18–39, 40–59, ≥60), BMI (normal, overweight, obese, no record), smoking status (never, former, now, no record), alcohol status (never, former, now, no record), physical activity(no, yes, no record), hypertension (no, yes), diabetes (no, boundary, yes), CVD (no, yes), and eGFR (<60,≥60, no record). Multiplicative interactions were used to estimate potential interactions between multiple subgroup elements and log(dietary live microbial intake).

Sensitivity analyses were conducted to verify the robustness of the study results. Firstly, participants who died within 2 years of follow-up were excluded to eliminate potential reverse causality. Secondly, participants with chronic diseases (including hypertension, cardiovascular disease, diabetes mellitus, cancer, and eGFR<60) were excluded, as these individuals were more likely to die during the follow-up period. Thirdly, to test the sensitivity of dietary live microbial intake to the findings, we re-quantified the total intake of live microbes (CFU) in individuals’ diets. Microbial intakes were assumed to be 10^{^3.5} CFU/g, 10^{^7} CFU/g, and 10^{^10} CFU/g for the low, medium, and high groups, respectively, then multiplied by the number of grams of food in each group and summed, and later re-classified into low, medium, and high groups by <10^{^8} CFU, 10^{^8}–10^{^11} CFU, and > 10^{^11} CFU for analysis. In addition, the newly calculated logarithms of dietary live microbial intake were analyzed with different mortality rates. Finally, the association with the dietary intake of live microorganisms was assessed using mortality due to accidents and injuries as the dependent variable.

All statistical analyses were performed using R version 4.2.1 (R Foundation for Statistical Computing, Vienna, Austria; http://www.r-project.org), and statistical significance was ascertained by a two-sided p value <0.05.

A total of 52,383 subjects, 51.62% of whom were female, with a mean age of 46.10 ± 0.20 years, were ultimately enrolled in our study. Over a median follow-up period of 118.75 months, there were 7,711 deaths and an all-cause mortality rate of 14.72%, including 1985 CVD deaths and a CVD mortality rate of 3.79%. The mean individual estimated dietary live microbial intake was 20,623.09 ± 536.25 (*10^{^6} CFU) and the mean of its logarithm was 8.14 ± 0.02. Of these, the low, medium, and high dietary live microbial intake groups had 19,917, 22,981, and 9,485 participants, respectively. According to Table 1, participants with higher levels of dietary intake of live microorganisms were more likely to be female, Non-Hispanic White, married/living with a partner, never smokers, current drinkers, physical exercises, had higher levels of education, wealth, BMI, eGFR, HEI, and energy intake, and had no comorbid hypertension, diabetes, cardiovascular disease, or cancer (all p < 0.0001). Finally, the higher the dietary intake of live microorganisms, the fewer deaths there were in the group (p < 0.0001).

The results of the COX regression analysis were displayed in Table 2. A dose–response relationship was found between dietary live microbial intake and all-cause mortality and CVD mortality. In the all-cause mortality analysis, compared to the low dietary live microbial intake group, the multivariate-adjusted HRs were smaller in the high group for model 1 (HR = 0.81, 95% CI = 0.73–0.90, P for trend<0.0001), model 2 (HR = 0.89, 95% CI = 0.80–0.99, P for trend = 0.01) and model 3 (HR = 0.91. 95% CI = 0.82–1.00, P for trend = 0.01). In addition, after multivariate adjustment, log (dietary intake of live microbes) was significantly associated with a reduced risk of all-cause mortality (HR = 0.97, 95% CI = 0.95–0.99). For the CVD mortality analysis, after adjustment for all confounding variables, the HR for microbial intake was smaller in the high group (HR = 0.77, 95% CI 0.63–0.95, P for trend = 0.005) and the log (dietary intake of live microbes) was negatively associated with CVD mortality (HR = 0.93, 95% CI = 0.89–0.97).

In restricted cubic spline regression models fully adjusted for covariates, the relationship between log dietary live microbial intake and mortality was non-linear (Figure 1A, p for nonlinear = 0.0029) and the relationship with CVD mortality (Figure 1B) was linear (p for nonlinear = 0.1113), with an overall decreasing trend.

As revealed by the Kaplan–Meier survival curves (Figure 2), higher dietary live microbial intake was accompanied by significantly lower all-cause mortality (Figure 2A) and CVD mortality (Figure 2B) in the following life (log-ranked p < 0.0001).

As illustrated in Figure 3, subgroup analyses showed the link between log(dietary live microbial intake) and reduced overall mortality was significant in females, those aged 40–59, overweight individuals, never drinkers, people with hypertension, those with borderline diabetes, and those without CVD (all p < 0.05). This link was also tied to lower CVD mortality in females, individuals aged ≥60, overweight people, never drinkers, and those without hypertension (all p < 0.05). Factors like age, alcohol status, and physical activity modified the association with all-cause mortality decline (P for interaction <0.05). Age, BMI, drinking status, physical activity, and hypertension altered the link with reduced CVD mortality (P for interaction <0.05). No significant interactions were found for other subgroups. These findings suggest that higher dietary intake of live microbes may have significant public health implications for reducing mortality risk, particularly in special populations.

In sensitivity analyses, dietary live microbial intake remained negatively associated with all-cause mortality and CVD mortality after excluding subjects dying within the first 2 years of follow-up (Supplementary Table S1). After excluding the baseline chronic disease population, dietary live microbial intake was negatively related to all-cause mortality and CVD mortality, but the relationship with all-cause mortality was not statistically significant (Supplementary Table S2), so we performed an RCS analysis to reveal that the negative association continued to exist (Supplementary Figure S2). Furthermore, these results remained statistically significant (all p < 0.05) after recalculation of dietary live microbial intake (Supplementary Table S3). No statistically significant associations were obtained in sensitivity analyses examining the association between dietary live microbial intake and mortality from accidents and injuries (Supplementary Table S4S1).