The National Health and Nutrition Examination Survey (NHANES) is a complex, multi-stage, stratified sampling health survey conducted in the United States. Data sources include structured interviews, telephone follow-ups, health screenings at mobile examination centers, and laboratory sample analysis. Prior to data collection, approval was obtained from the National Center for Health Statistics Institutional Review Board, and participants provided written informed consent upon enrollment. This study utilized data from adults aged 20 and older who participated in eight cycles of the NHANES survey (from 2003–2004 to 2017–2018) [12].

To ascertain whether participants have CVD, they are asked the following question: "Has a doctor or other health professional ever told you that you had coronary heart disease, congestive heart failure, heart attack, stroke or angina?" If a participant answers "yes" to this question at least once, they are considered to have CVD [13]. Among 80,312 participants, 5,205 had CVD disease. After excluding participants under 20 years old, pregnant women, and those with missing data on dietary niacin intake, follow-up time (‘permth_int’), Body mass index (BMI, kg/m2)., marital status, and education level, a final sample of 4,377 individuals was included in this study.

The dietary interview component, known as "What We Eat in America" (WWEIA), is conducted in collaboration between the United States Department of Agriculture (USDA) and the Department of Health and Human Services (DHHS). Data on dietary niacin intake are collected through two 24-h dietary recall interviews. The first recall is conducted in-person at mobile examination centers, followed by a second recall via telephone 3 to 10 days later. These recalls are facilitated using the Computer-Assisted Dietary Interview System managed by NHANES interviewers. The USDA’s Food and Nutrient Database for Dietary Studies (FNDDS) was utilized to calculate the nutrients and food components in all food items [14]. Dietary intake in this study represents either the average of the two recalls or the value from one recall (if only one recall was available for a participant). Among the 4,377 participants, 456 (10.4%) had data from only one dietary recall.

The outcome variables were mortality status including mortality of all-cause and CVD, which were determined by National Death Index (NDI) by 31 December 2019. The NDI is a highly reliable and widely used resource for death identification. The ICD-10 was used to determine disease-specific death [15]. Death due to CVD was defined as ICD-10 codes 100–109, 111, 113, 120–151 or 160–169. The total number of deaths was 1659, with 790 deaths attributed to CVD.

The following covariates, which were all baseline measurements, were included in this study: age(years), sex(men/women), race (non- Hispanic white or other), marital status (married/unmarried/divorced), education(less than high school/high school or equivalent/college/above), annual poverty-income ratio of household income to the poverty line), energy intake (kcal), smoking status (current/previous/none), drinking status (current/previous/none), regular exercise (yes/no), BMI, hyperlipidemia (defined as the presence of one or more of the following serum measures: total cholesterol > 200 mg/dL, triglycerides > 200 mg/dL, high density lipoproteins < 40 mg/dL in males and < 50 mg/dL in females, low density lipoproteins > 130 mg/dL or current use of cholesterol lowering medications [16]), hypertension (defined as having a systolic blood pressure greater than or equal to 140 mmHg or diastolic blood pressure greater than or equal to 90 mmHg, or currently taking medication to lower high blood pressure) [17], chronic kidney disease (CKD) (defined as glomerular filtration rate (eGFR) < 60 ml/min/1.73m2 and/or urinary albumin/creatinine ratio (ACR) > 30 mg/g) [18], and diabetes mellitus (DM) (defined by a self-reported diagnosis, medication for hyperglycemia, glycosylated hemoglobin (HbA1c) > 6.5%, or fasting blood glucose > 7.0 mmol/L, or random blood glucose/two-hour OGTT blood glucose > 11.1 mmol/L [19]). Meanwhile, we performed multicollinearity diagnostics for these covariates and found that multicollinearity was weak (all GVIF values were < 0.05) (Supplementary Table 1).

Given the complex sampling design of NHANES, all analyses in this study incorporate sample weights (wtmec2yr), strata, and primary sampling units. Baseline characteristics of sociodemographic information, lifestyle behaviours and disease status, were presented as mean ± SD (standard deviation) or numerical (percentage). General linear models and chi-square tests were used to compare the difference. Weighted Cox proportional hazards regression models were used to calculate hazard ratios (HRs) and 95% confidence intervals (Cis) to examine the associations between dietary niacin intake and the risk of all-cause mortality and CVD mortality. The survival time of participants is calculated from the NHANES interview date to the date of death or the end of follow-up (December 31, 2019), whichever occurs first. We adjusted for a series of confounding factors, including age, sex, race, marital status, smoking status, drinking status, regularly exercise, BMI, poverty income ratio, energy intake, DM, hyperlipidemia, hypertension and CKD. Additionally, we created continuous variables by taking the median of dietary niacin intake for each category and examined linear trends. We employed restricted cubic spline (RCS) analysis with four nodes (at the 25th, 50th, 75th and 95th percentiles) to investigate the non-linear relationship between dietary niacin intake and both all-cause mortality and CVD mortality.

Further subgroup analysis will be conducted based on age (< 65 or ≥ 65), sex (male or female), race/ethnicity (Mexican American, Other Hispanic, Non-Hispanic White, Non-Hispanic Black, or Other), smoking status (never, former, or current), alcohol consumption (never, former, or current), regular exercise (no or yes), BMI (< 25, 25–29.9, or ≥ 30), CKD (no or yes), and DM (no or yes) to observe their interaction with dietary niacin. We also conducted a series of sensitivity analyses: (1) To minimize potential bias from reverse causality, participants with follow-up times less than two years were excluded (n = 3811). (2) Participants with only one dietary recall were excluded (n = 3921). (3) Repeat analyses were performed by excluding CVD participants who died within two years of follow-up and had only one dietary recall (n = 3431). (4)Sensitivity analyses were conducted by altering the number of nodes in the restricted cubic spline, using three nodes (at the 25th, 50th, and 75th percentiles) and five nodes (at the 20th, 40th, 50th, 60th and 80th percentiles), to further explore the relationship.

All statistical analyses were conducted by R 4.3.1, and p-values < 0.05 were considered statistically significant.