UK Biobank is a large-scale multicenter prospective cohort study with more than 500,000 participants aged 37–73 years who were recruited between 2006 and 2010. The participants were primarily drawn from the general populations of 22 dedicated assessment centers in England, Wales, and Scotland. Participants provided information on health-related aspects through touchscreen questionnaires, verbal interviews, and physical measurements (13). The details of the study design and methods have been described in previous studies (14). Written informed consent was obtained from all study participants in accordance with the Declaration of Helsinki. Ethical approval for this study was granted by the National Health Service (NHS) North West Multicenter Research Ethics Committee. The study was conducted under application No.81680.

In this study, we focused on participants who completed the online 24-h recall questionnaire at least once between April 2009 and June 2012. Participants were invited to complete the dietary recall on up to five separate occasions. The average dietary intake across all available recalls was used as an estimate of habitual intake. In total, 210,950 individuals had completed at least one dietary questionnaire. We excluded participants who were pregnant (n = 171), those with extreme energy intake (estimated energy requirement in the highest or lowest 1% of the distribution) (n = 4,216), individuals who were lost to follow-up (n = 562), or those with missing or overlapping information on tea consumption (n = 7,692). Additionally, we also excluded participants with more than two missing covariates (approximately 10%) (n = 2,948), leaving 195,361 participants for the primary analysis (Figure 1 and Supplementary Table S1).

During the study, participants were asked about their tea consumption through a touchscreen questionnaire, “How many drinks of tea yesterday?” If participants reported drinking tea, they were asked to specify the type (standard tea, rooibos tea, green tea, herbal tea, or other tea) and to indicate whether they added sugar or sweeteners. Participants were categorized into four groups based on their tea consumption habits: non-consumers, unsweetened tea consumers, sugar-sweetened tea consumers, and artificially sweetened tea consumers. A tea consumer was defined as a participant who reported drinking tea at least once during the dietary recalls, based on responses to up to five 24-h dietary assessments. If participants consumed different types of tea (unsweetened, sugar-sweetened, or artificially sweetened tea) across the five recalls, they were classified as overlapping consumers and were subsequently excluded from the analysis (n = 5,342) (Supplementary Table S1). To minimize confusion regarding serving size (e.g., mug vs. cup), a standard drink size was defined as approximately 250 mL or 8.5 ounces. Tea consumption was categorized into five levels: 0–1.5, 1.5–2.5, 2.5–3.5, 3.5–4.5, and 4.5 or more drinks per day. Participants who reported drinking more than 10 cups of tea per day were asked to confirm their responses.

Data on potential confounding variables were obtained from baseline touchscreen questionnaires and health records, including sociodemographic characteristics, behavioral factors, and health conditions. Sociodemographic variables included age, sex, ethnicity (White or other), education level (college/university degree or other), and the Townsend Deprivation Index (TDI), a continuous measure of socioeconomic deprivation based on postcode, incorporating indicators such as unemployment, non-car ownership, non-home ownership, and household overcrowding. Behavioral factors included smoking status (never, former, current), pack-years of smoking (calculated as: [cigarettes per day ÷ 20] × years of smoking), alcohol consumption, and physical activity level (low, moderate, or high), assessed using the International Physical Activity Questionnaire (IPAQ) and categorized according to standard scoring protocols based on frequency, intensity, and duration of activity. Basal metabolic rate (BMR), reflecting resting energy expenditure, was calculated from anthropometric data and treated as a continuous variable. Dietary factors included intakes of total energy, total sugar, fruits, vegetables, red, and processed meats, coffee, milk, sugar-sweetened beverages (SSBs), artificially sweetened beverages (ASBs), and naturally sweet juices (NSJs). Beverage categories were defined as follows: sugar-sweetened drinks (e.g., carbonated soft drinks or fruit-flavored beverages), artificially sweetened beverages (e.g., squash, cordial, or fizzy drinks), and natural fruit and vegetable juices.

Health-related factors included body mass index (BMI, calculated as weight in kilograms divided by height in meters squared), self-rated overall health (poor, fair, good, or excellent), self-reported cardiovascular disease (myocardial infarction, angina, or stroke), self-reported cancer, family history of CVD or cancer, hypertension, diabetes, depression, long-standing illness, current use of cholesterol-lowering or antihypertensive medications, and use of vitamin or mineral supplements (including vitamins A, B, C, D, E, multivitamins, calcium, iron, zinc, selenium, and fish oil). Disease status was defined based on self-report, medication use, or hospital linkage using UK Biobank clinical classification codes.

Mortality data, including the date and cause of death, were obtained from the NHS Information Centre (for England and Wales) and the NHS Central Register (for Scotland). The underlying cause of death was classified using the International Statistical Classification of Diseases, 10th Revision (ICD-10). We examined all-cause mortality, cancer mortality (ICD-10 codes C00–D48), and cardiovascular disease (CVD) mortality (ICD-10 codes I00–I99). Participants were followed from the date of their first completed 24-h dietary recall until the date of death or the end of the follow-up period (March 11, 2023), whichever occurred first.

The UK Biobank genotyping was conducted using two genome-wide arrays: the UK BiLEVE array (for 50,000 participants) and the UK Biobank Axiom array (for 450,000 participants). These arrays were highly similar and collectively covered approximately 800,000 single-nucleotide polymorphisms (SNPs) and insertion–deletion markers (15). After excluding sample outliers based on heterozygosity and missingness, and applying kinship-based exclusions, genetic data on caffeine metabolism were available for 403,816 individuals in the analytic cohort (16, 17). We used four SNPs—rs2472297 (near CYP1A2), rs6968554 (near AHR), rs17685 (in POR), and rs56113850 (near CYP2A6)—to construct a genetic caffeine metabolism score (CMS_G4), which has been shown to correlate with blood caffeine metabolite levels. These variants are located in or near genes involved in caffeine metabolism. A weighted CMS_G4 was calculated by multiplying the number of risk alleles by their corresponding β-coefficients. The wCMS_G4 was then divided into four categories: 0–2, >2–3, >3–4, and >4, with higher scores indicating faster caffeine metabolism.

Baseline characteristics related to tea consumption were presented as means with standard deviations (SDs) for continuous variables and as frequencies with percentages for categorical variables. We performed multiple imputations using chained equations to generate five datasets to address missing covariate data. Details of the missing data are summarized in Supplementary Table S2. The average amount of sugar or sweetener added to tea across multiple 24-h dietary recall questionnaires is presented in Supplementary Table S3. In addition, Spearman correlation coefficients were used to assess the consistency of tea consumption across multiple recalls (Supplementary Table S4).

Cox proportional hazard regression models were used to examine the association between tea consumption and the risk of all-cause and cause-specific mortality, with non-consumers serving as the reference group. Two models were constructed: a basic model and a multivariable-adjusted model. Model 1 was adjusted for baseline age (continuous) and sex (female or male). Model 2 was further adjusted for the following covariates: ethnicity (White or other), Townsend Deprivation Index (continuous), education level (degree or no degree), smoking status (current, former, or never), pack-years of smoking (continuous), overall health (poor, fair, good, or excellent), basal metabolic rate (continuous), physical activity level (low, moderate, or high), body mass index (continuous), hypertension (yes or no), diabetes (yes or no), depression (yes or no), family history of CVD disease (yes or no), family history of cancer (yes or no), long-standing illness (yes or no), use of cholesterol-lowering drug use or blood pressure drug use (yes or no), vitamin and mineral supplement use (yes or no), and dietary intakes of energy, total sugar, fresh fruit, vegetables, red and processed meats, alcohol, coffee, milk, naturally sweet juices, sugar-sweetened beverages, and artificially sweetened beverages. The proportional hazard assumption was tested using Schoenfeld residuals, and no significant violations were detected.

Restricted cubic spline (RCS) regression, a method for modeling non-linear relationships where the strength or direction of association may vary across the range of an independent variable, was used to assess potential non-linear associations between tea consumption and mortality outcomes. RCS models were fitted with four knots placed at the 5th, 35th, 65th, and 95th percentiles to evaluate non-linear relationships between the amounts of sugar or artificial sweeteners added to tea and risks of all-cause, CVD, and cancer mortality. Based on previous studies, individuals consuming five or more cups of tea per day were considered excessive drinkers. Tea consumption was classified into five categories: 0 to <1.5, 1.5 to <2.5, 2.5 to <3.5, 3.5 to <4.5, and ≥4.5 drinks per day. We conducted subgroup analyses to evaluate whether the associations between tea consumption and mortality differed by age (<60 or ≥60 years), sex (female or male), ethnicity (white or other), education (with or without degree), TDI (mean), BMI (<30 or ≥30 kg/m²), smoking status (former/current or never), alcohol consumption, coffee (mean), physical activity (low, moderate, or high), hypertension (yes or no), diabetes (yes or no), family history of CVD (yes or no), and family history of cancer (yes or no). To separately examine the effects of unsweetened, sugar-sweetened, and artificially sweetened tea, we performed stratified multivariable-adjusted models by wCMS_G4 categories and by ethnicity (White vs. non-White). Given that standard black tea is the most commonly consumed type in the UK Biobank cohort, we also conducted a subgroup analysis restricted to standard tea consumers to assess consistency of associations (Supplementary Table S5).

In sensitivity analyses, we first excluded participants with missing covariates to assess the robustness of our findings. We then excluded deaths occurring within the first 2 years of follow-up to minimize potential reverse causality. Additionally, we excluded participants for whom added sugar in tea was excluded from total sugar intake calculations, and those who reported tea consumption in the past year but did not consume tea the day before the dietary recall. To evaluate potential confounding by overall diet quality, we further adjusted for a modified Alternative Healthy Eating Index (AHEI) score based on 11 components (Supplementary Table S6). As part of the E-value analysis, we estimated the strength of unmeasured confounding that would be required to explain away the observed associations between tea consumption and mortality. E-values and their lower confidence limits (LCLs) were calculated to support the sensitivity analysis. All analyses were performed using R software (version 4.1.3), and two-sided p-values <0.05 were considered statistically significant.

A total of 195,361 individuals were included in the analysis of all-cause mortality. For cause-specific mortality outcomes, participants without corresponding events were excluded, resulting in 186,934 participants for cardiovascular disease mortality and 178,057 for cancer mortality. Baseline characteristics of participants stratified by type of tea consumption are presented in Table 1. Among all participants, 162,516 (83.2%) were tea consumers. The majority consumed unsweetened tea (81.6%), followed by sugar-sweetened tea (12.2%) and artificially sweetened tea (6.2%). On average, 1.1 teaspoons of sugar and 1.4 teaspoons of artificial sweeteners were added per drink of tea. Compared to tea consumers, non-consumers tended to have a higher BMI and were more likely to prefer alcohol and coffee. Consumers of unsweetened tea were more likely to have higher educational attainment, lower BMI, lower basal metabolic rate, and generally exhibited healthier lifestyle behaviors. Consumers of sugar-sweetened tea were more likely to be male, with higher total energy intake and lower fruit consumption. Those who drank artificially sweetened tea were generally older, more likely to be current or former smokers, had higher BMI, were more frequently overweight, reported fair or poor self-rated health, and were more likely to have hypertension, diabetes, CVD, long-standing illness, and to use cholesterol-lowering or antihypertensive medications.

Over a median follow-up of 13.6 years (IQR 13.0–14.4 years), 11,718 deaths from all causes were recorded, including 6,415 cancer deaths and 2,202 CVD deaths. Cox proportional hazards models revealed a significant inverse dose-dependent association between tea consumption and all-cause mortality (p < 0.001) after adjustment for multiple covariates (Table 2). Compared with non-consumers, participants who consumed unsweetened tea had progressively lower risks of all-cause mortality: 0–1.5 drinks (HR = 0.92; 95% CI: 0.85, 0.99), 1.5–2.5 drinks (HR = 0.83; 95% CI: 0.78, 0.89), 2.5–3.5 drinks (HR = 0.84; 95% CI: 0.78, 0.90), 3.5–4.5 drinks (HR = 0.80; 95% CI: 0.75, 0.86), and >4.5 drinks (HR = 0.83; 95% CI: 0.77, 0.89), respectively. For sugar-sweetened tea, hazard ratios for all-cause mortality across increasing consumption categories were 0.81 (95% CI, 0.70–0.94), 0.95 (95% CI, 0.84–1.07), 0.91 (95% CI, 0.80–1.03), 0.85 (95% CI, 0.74–0.98), and 0.87 (95% CI, 0.76–0.99), respectively (Table 2). The inverse association between unsweetened tea consumption and all-cause mortality was confirmed in RCS models, showing a significant U-shaped non-linear relationship (P for overall association <0.001; P for non-linearity <0.001) (Figure 2). In contrast, a monotonically decreasing, linear relationship was observed for sugar-sweetened tea (P for non-linearity >0.05). However, no significant association was observed between artificially sweetened tea and all-cause mortality in either the multivariable-adjusted Cox model or the RCS analysis.

In both multivariable Cox regression and nonlinear RCS models, we observed inverse associations between unsweetened tea consumption and mortality from cancer and CVD. For cancer-specific mortality, the HRs associated with daily unsweetened tea consumption were: 0–1.5 drinks (HR = 0.90; 95% CI: 0.80, 1.01), 1.5–2.5 drinks (HR = 0.87; 95% CI: 0.78, 0.96), 2.5–3.5 drinks (HR = 0.81; 95% CI: 0.73, 0.90), 3.5–4.5 drinks (HR = 0.86; 95% CI: 0.77, 0.97), and >4.5 drinks (HR = 0.87; 95% CI: 0.78, 0.97). For CVD-specific mortality, the corresponding HRs were: 0 to 1.5 drinks (HR = 0.93; 95% CI: 0.76, 1.13), 1.5–2.5 drinks (HR = 0.78; 95% CI: 0.66, 0.93), 2.5–3.5 drinks (HR = 0.85; 95% CI: 0.72, 1.02), 3.5–4.5 drinks (HR = 0.73; 95% CI: 0.60, 0.89), and >4.5 drinks (HR = 0.75; 95% CI: 0.62, 0.91), respectively. A nonlinear U-shaped relationship between unsweetened tea and cancer mortality was observed (P for nonlinearity =0.008). In contrast, a linear inverse association was observed between unsweetened tea and CVD mortality, with no evidence of nonlinearity (P for nonlinearity >0.05). For both sugar-sweetened and artificially sweetened tea, the number of cancer- and CVD-related deaths was relatively small. While potential associations were observed, they were not statistically significant. To further illustrate survival probabilities, Kaplan–Meier survival curves were generated for all-cause, cancer, and CVD mortality (Supplementary Figure S1). Participants with moderate intake of unsweetened tea demonstrated significantly better survival outcomes compared to non-consumers. These findings are consistent with the Cox model results.

To examine whether the association between tea consumption and mortality was influenced by genetic variation in caffeine metabolism, we analyzed data stratified by weighted caffeine metabolism score (wCMS_G4) derived from four established SNPs. Participants drank >4.5 drinks/d unsweetened tea had similar inverse associations with all-cause mortality in the lower strata of wCMS_G4 0–2 [HR = 0.73 (95% CI, 0.58–0.93)] to those in the higher strata of wCMS_G4 >4 with >4.5 drinks/d [HR = 0.80 (95% CI, 0.71–0.90)], indicating no significant modification of the association between unsweetened tea and all-cause mortality by caffeine metabolism genotype (Supplementary Table S7). However, no statistically significant inverse association was observed for cancer mortality in any wCMS_G4 subgroup among those consuming >4.5 drinks/day of unsweetened tea. The effect size of the association between >4.5 drinks/d unsweetened tea and CVD mortality was much lower in the strata of wCMS_G4 3–4 [HR = 0.69 (95%CI 0.49–0.99)] than that in the strata of wCMS_G4 > 4 [HR = 0.71 (95%CI 0.53–0.97)] (Supplementary Tables S8, S9). For sugar-sweetened and artificially sweetened tea, no statistically significant associations with all-cause, cancer, or CVD mortality were observed in any genetic subgroup. Additionally, for white individuals consuming more than 4.5 drinks, we found similar inverse associations of unsweetened tea with all-cause mortality (HR = 0.77, 95% CI 0.60–0.98) in the lower strata of wCMS_G4 0–2 and in the higher strata of wCMS_G4 > 4 (HR = 0.80, 95% CI 0.71–0.90). No evidence of effect modification by wCMS_G4 was found for the associations between unsweetened tea and all-cause or cancer mortality in White individuals (Supplementary Tables S10–S12).

In stratified analyses, the associations between tea consumption and all-cause or cause-specific mortality did not differ significantly by age group, sex, ethnicity, education, Townsend Deprivation Index (TDI), BMI, smoking status, alcohol intake, coffee intake, physical activity (IPAQ), hypertension, diabetes, family history of cancer, or family history of CVD. After adjustment for potential confounders, inverse associations between unsweetened tea consumption and all-cause mortality were observed consistently across most subgroups, except among participants of non-White ethnicity, those with low physical activity levels, and those with diabetes. A significant interaction was found between unsweetened tea consumption and coffee intake on all-cause mortality (P for interaction =0.03) (Supplementary Figure S2), with a stronger inverse association observed among participants with low coffee consumption compared to those with high coffee consumption. Inverse associations between sugar-sweetened tea and all-cause mortality were significant only in the following subgroups: Age ≥ 60 years, white ethnicity, BMI < 30 kg/m², Alcohol intake ≥ average level, Presence of hypertension, No diabetes, and No family history of CVD. For cancer mortality, inverse associations with unsweetened tea were significant only among individuals with: TDI above the average level, BMI < 30 kg/m², former or current smoking status, and No diabetes (Supplementary Figure S3). A significant interaction was observed between sugar-sweetened tea and hypertension on cancer mortality (P for interaction =0.02), with a stronger inverse association among individuals with hypertension than those without. For CVD mortality, we observed a multiplicative interaction between ethnicity and artificially sweetened tea (P for interaction =0.01), where the hazard ratio among White participants was much lower than that among non-White individuals. A similar interaction was identified between artificially sweetened tea and hypertension (P for interaction =0.04), with a stronger inverse association in participants without hypertension compared to those with hypertension (Supplementary Figure S4). To determine whether these associations were primarily driven by black tea, we conducted a subgroup analysis limited to participants who reported consuming standard tea. The results for all-cause, cancer, and CVD mortality were consistent with the overall findings, indicating that the observed inverse associations were largely attributable to black tea consumption (Supplementary Table S13).

Sensitivity analyses demonstrated that the observed associations were robust under multiple scenarios: Excluding participants with missing covariates did not materially alter the associations between tea consumption and mortality (Supplementary Table S14). Excluding deaths within the first 2 years of follow-up (Supplementary Table S15). Excluding added sugar in tea from the total sugar intake did not change the results (Supplementary Table S16). Removing participants who reported drinking tea in the past year but not on the recall day did not significantly alter the estimates (Supplementary Table S17). Additional adjustment for a modified AHEI score also yielded consistent results, supporting the robustness of our findings (Supplementary Table 18). Further adjustment for the total number of chronic conditions at baseline did not materially change the results, indicating robustness to baseline multimorbidity (Supplementary Table 19). Moreover, E-values for the association of unsweetened, sugar-sweetened, and artificially sweetened tea with all-cause mortality ranged from 1.29 (lower confidence limit [LCL], 1.00) to 1.81 (LCL, 1.60), suggesting a moderate level of unmeasured confounding would be required to fully explain these associations (Supplementary Table S20).