NHANES is a nationwide study by the National Center for Health Statistics, which assesses the health and nutrition status of the US noninstitutionalized civilian population with a complex, stratified, and multistage probability sample design. It collects interview information from the respondents at their homes, and conducts physical examinations and collects dietary information at a mobile examination center. Elaborate insights into the study framework and research methodologies are available in previous publication [12]. The study protocol was approved by the Institutional Review Board of the National Center for Health Statistics. All participants offered written informed consent.

This study included individuals aged 20 to 85, across nine survey cycles spanning from 2001 to 2002 to 2017–2018, comprising a cohort of 50,068 individuals with documented mortality data. After excluding those with incomplete tea consumption information (n = 5,647) and pregnant women at the time of examination (n = 1,145, because they might be restricting their tea intake based on advice from a healthcare professional.), the eligible study population consisted of 43,276 participants (Supplementary Figure S1).

Dietary data were acquired through 24-h dietary recalls facilitated by adept interviewers using USDA (US Department of Agriculture) automated multi-pass methods during two distinct time periods [13, 14]. From 2001 to 2002, a singular dietary recall was conducted in person at a mobile examination center. Subsequently, from 2003 to 2018, an additional dietary recall, administered via phone, was incorporated approximately 3–10 days after the initial recall. Dietary data from the participants who underwent at least one dietary recall interview were incorporated into the current analysis. Specifically, data from the single-day participants were used as reported, whereas data from the two-day participants were averaged. We determined the tea consumption of each participant based on the USDA food codes and their descriptions. During this process, we excluded unconventional teas such as herbal tea, hibiscus tea, corn tea, chamomile tea, and chai tea. We evaluated the stability of the tea consumption pattern using two-day tea consumption data from individuals, employing weighted Pearson correlation (r = 0.76; 95% confidence interval [CI]: 0.74 to 0.78). The tea consumption was categorized into five distinct subgroups: non-drinking, < 1 cup/day, 1 to < 3 cups/day, 3 to < 5 cups/day, and ≥ 5 cups/day, where one cup is approximately equal to 8 US fluid ounces (236.59 milliliter). Based on the description information provided for each tea, we determined the type of tea (black, green, or unreported) and whether sugar was added to the tea (with sugar, without sugar). Detailed tea codes and description information can be found in Supplementary Table S1. Due to the limited sample size of green tea drinkers (n = 1,116, 7.9%) among tea drinkers, we combined them with individuals who did not report the type of tea consumed (n = 4,681, 33.13%) as ‘Other tea’, compared to ‘Black tea’. The specific amount of sugar added to tea was determined using the nutritional information from the Food and Nutrient Database for Dietary Studies (https://www.ars.usda.gov/northeast-area/beltsville-md-bhnrc/beltsville-human-nutrition-research-center/food-surveys-research-group/docs/fndds).

Mortality information was obtained from the NHANES Linked Mortality File up to December 31, 2019. The causes of death attributed to diseases were defined using the International Classification of Diseases, Tenth Revision (ICD-10). The follow-up period was computed from the date of the interview or examination until the date of death or until the study ended on December 31, 2019, whichever came first. Data on all-cause mortality rates for 2019 by single-year age from 50 to100 years were extracted from the National Center for Health Statistics (by sex or ethnicity, where applicable) [15]. We downloaded the specific cause-of-death (cancer, and cardiovascular and cerebrovascular diseases [CVD]) mortality data for the US population in 2019, which included mortality data for individuals aged 20 to 84 years by single-year age groups, provided by the Wide-Ranging Online Data for Epidemiologic Research database of the Centers for Disease Control and Prevention. Due to the absence of specific cause-of-death mortality data for individuals aged 85 and older, we estimated the missing data using a Poisson regression model (Supplementary Figure S2).

To attenuate potential confounding effects, we incorporated covariates into our analytical framework when constructing Cox regression models and estimating life expectancy. Data on demographic and socioeconomic information were collected from questionnaires, including sex (male and female), age, race and ethnicity (White, Black, Hispanic, and others), educational attainment (< high school, high school, and > high school), family income (poor, moderate, and rich), marital status (married, unmarried, and others), and medical insurance (covered and not covered). Moreover, lifestyle factors, such as smoking status (never, former, or now) and dietary intake, were included. The dietary intake variables included alcohol consumption, fruit and vegetable consumption, grain consumption, dairy intake, protein food consumption, and total energy intake. Detailed insights regarding the data collection and definitions of these variables can be found in eMethod 1. Additionally, in eMethod 1, we described the definition of one grouping factor in the subgroup analysis—physical activity (insufficient; moderate or sufficient) [16].

Complex sampling design and sampling weights were considered to strengthen the reliability of our analysis with nationally representative findings. To handle missing data pertaining to covariates, we employed multiple imputation with 20 datasets and chained Eqs. [17, 18]. The association between tea consumption and mortality risk was estimated using Cox proportional hazards models, adjusting for covariates described previously. The proportional hazards assumption was appropriate, as assessed by the Kaplan-Meier and Schoenfeld residuals methods [19], with no violations detected (all p > 0.05). Furthermore, we employed the restricted cubic spline (RCS) method to investigate whether there was a non-linear relationship between tea consumption and all-cause mortality.

The estimated life expectancies of participants across various levels of tea consumption were calculated by life table method [20,21,22]. We utilized three estimates to compute the cumulative survival rates from the age of 50 years to 100 years: (1) age-specific mortality of all causes in 2019; (2) adjusted hazard ratios; and (3) age-specific distribution of tea consumption. The expected life expectancy gain from tea consumption is calculated as the cumulative sum of the survival curve difference compared to the non-tea-drinking group. To calculate the confidence intervals (CIs) for life expectancy estimates, we employed a bootstrapping simulation with 1000 iterations, reporting the 95% CI as the 25th and 975th ordered draws of the distribution. We used the all-cause mortality rate for the entire population to compute life expectancy. Further details regarding the methods employed to estimate these differences in life expectancy have been described in eMethod 2 and in a prior study [15, 22,23,24]. Due to the absence of mortality data for individuals of ‘other’ race, their life expectancy was not estimated in this study. Additionally, we employed the Arriaga decomposition method to calculate life expectancy gains attributed to different contributions from three causes of death, including cancer, CVD, and others. This methodology encompasses two essential steps: (1) decomposing the disparity in life expectancy across various age groups, and (2) within distinct age group, breaking down the difference in life expectancy by specific causes [25, 26].

We performed five sensitivity analyses to assess the robustness of our primary findings. First, we excluded participants without tea consumption to detect whether the results would be influenced by a high proportion of non-drinkers. Second, individuals lacking complete variable information were excluded from the analysis to detect the robustness of results. Third, we excluded participants who died within the first 2 years of follow-up to mitigate the potential influence of reverse causation [27]. Fourth, we included several health-related factors that may influence the relationship between tea consumption and life expectancy in our model, such as BMI, the presence of cancer, diabetes, and hypertension (described in eMethod 3) [16]. Fifth, we classified unconventional teas as “Other types of tea” to assess the stability of the model. All statistical analyses were performed using R version 4.3.1. All statistical tests were two-sided, and a significance level of P < 0.05 was utilized to determine statistical significance.