Consumption of coffee and tea with all-cause and cause-specific mortality: a prospective cohort study

The UK Biobank is a large-scale prospective study that recruited 502,507 participants aged 37–73 years from the general population between 2006 and 2010 [15]. Participants attended 1 of 22 dedicated assessment centers nationally across England, Wales, and Scotland where they provided information on health-related aspects through touch-screen questionnaires and physical measurements [16]. The details of the study design and methods have been described in previous studies [17]. In the present study, we excluded participants who were lost to follow-up (n = 1346) or with missing information on coffee or tea consumption (n = 3003) at baseline, leaving 498,158 participants for the primary analysis (Additional file 1: Fig. S1).

The touchscreen questionnaire included part of a dietary assessment of a series of common food and beverage items. Participants were asked about their average intake of coffee in the last year “How many cups of coffee do you drink each day (including decaffeinated coffee)?” and “How many cups of tea do you drink each day (including black and green tea)?” Participants either selected the number of cups, “Less than 1,” “Do not know,” or “Prefer not to answer.” If coffee and tea consumption exceeded 10 and 20 cups/day, respectively, then participants were asked to confirm their answers.

To guide covariates selection, we constructed a directed acyclic graph (DAG) based on sociodemographic characteristics and a prior knowledge of potential confounding factors associated with all-cause mortality [10, 18, 19]. Additional file 1: Fig. S2 shows the DAG depicted causal relationships between measured variables in the current analysis. The program of DAGitty was used to identify the minimally sufficient adjustment set [20]. We used the baseline touch-screen questionnaire to collect sociodemographic, behavioral, and other factors. Sociodemographic factors were documented including sex, age, ethnicity (White, Asian or Asian British, Black or Black British, and others), and education levels (college or university degree, upper secondary, lower secondary, vocational, and others). Behavioral factors included smoking status (never, previous, and current), alcohol intake frequency (never, special occasions only, one to three times a month, once or twice a week, three or four times a week, daily or almost daily), physical activity (low, middle, and high, measured using the International Physical Activity Questionnaire [IPAQ]), and dietary pattern (healthy and unhealthy, healthy diet was based on consumption of at least 4 of 7 dietary components: fruits: ≥ 3 servings/day, vegetables: ≥ 3 servings/day, fish: ≥ 2 servings/week, processed meats: ≤ 1 serving/week, unprocessed red meats: ≤ 1.5 servings/week, whole grains: ≥ 3 servings/day, refined grains: ≤ 1.5 servings/day) (Additional file 1: Table S1) [21, 22]. Body mass index (BMI) (< 25, 25 to < 30, ≥ 30 kg/m²) was derived from physical measurement and calculated by dividing weight (kg) over height (m) squared. General health status was categorized as excellent, good, fair, and poor. Information on chronic diseases (e.g., hypertension, diabetes, and depression) was collected from touchscreen questionnaires, medical examinations, and hospital inpatient records.

Mortality information was obtained from death certificates, which were provided by the NHS Information Centre (England and Wales) and the NHS Central Register Scotland (Scotland) for the date of death. International Classification of Diseases (ICD-10) codes were used to classify deaths from CVD (ICD 10 codes I00-I79), respiratory disease (ICD 10 codes J09-J18 and J40-J47), digestive disease (ICD 10 codes K20-K93), and other causes.

We summarized baseline characteristics according to coffee and tea consumption categories as percentages for categorical variables, while means with standard deviations (SDs) for normal continuous variables and median and interquartile range (IQR) for non-normal variables. Normality test was applied by Shapiro-Wilk normality test. Multiple imputations using the chained equations (MICE) method were performed to handle missing covariates. Five imputed datasets were constructed and Rubin’s rule was used to combine the results [23].

To assess the dose-response associations of separate coffee and tea consumption with all-cause mortality and cause-specific mortality, we used restricted cubic splines models with 4 knots at the 25th, 50th, 75th, and 95th centiles. Tests for linearity or nonlinearity used the Wald test to calculate P-values, which was performed to test the null hypothesis that the coefficient of the second spline is equal to 0 [24]. In our analysis, the null hypothesis was rejected (P < 0.05) and concluded that there was a nonlinear relationship between separate coffee and tea consumption with all-cause mortality and cause-specific mortality. In the spline models, we adjusted for potential confounders including sex, age, ethnicity, education levels, BMI, smoking status, alcohol intake frequency, physical activity, dietary pattern, general health status, hypertension, diabetes, and depression. Coffee and tea consumption were mutually adjusted. Then, we divided coffee and tea consumption into four groups using prior validated thresholds based on the restricted cubic spline of association between separate consumption of coffee and tea with mortality. Participants who drank ≥ 5 cups/day of coffee or tea were defined as excess consumption based on previous studies [25]. Finally, we defined the categories as follows: coffee: none, < 1–2, 3–4, and ≥ 5 cups/day; tea: none, < 1–1, 2–4, and ≥ 5 cups/day. Cox proportional hazards models were used to estimate the hazard ratios (HRs) and 95% confidence intervals (CIs) of separate coffee and tea consumption groups with all-cause mortality and cause-specific mortality. Proportional hazard assumptions were verified using the Schoenfeld residuals method, and no significant deviations were observed. Follow-up time was calculated from the date of questionnaire completion in which the baseline coffee and tea consumption were available, lost to follow-up, death, or end of follow-up (23 March 2021), whichever came first. The multi-adjusted models were adjusted with the same covariates as the restricted cubic spline. Furthermore, to quantify the magnitude of combined consumption of coffee and tea with mortality, participants were categorized into 16 groups according to coffee and tea consumption categories, with participants who had neither coffee nor tea consumption comprising the reference group. Coffee consumption of < 1–2 cups/day and tea consumption of 2–4 cups/day were combined into one category because these participants had the highest proportion and the lowest mortality rate.

In addition, we performed subgroup analyses to assess potential modification effects and determine whether there was any population heterogeneity according to age, sex, BMI, physical activity, smoking status, alcohol intake frequency, dietary pattern, hypertension, diabetes, and depression. The interactions between baseline characteristics and combined coffee and tea consumption (combined < 1–2 cups/day of coffee and 2–4 cups/day of tea vs. neither coffee nor tea consumption) were examined using the likelihood ratio test (LRT).

Additional analyses were further conducted. First, we repeated the main analyses by excluding mortality cases that occurred in the first 3 years of follow-up. Second, we conducted the main analyses using available data before multiple imputations for missing covariates. Third, because severe diseases could confound results, we defined a population that excluded participants with prevalent CVD and cancer at baseline. Fourth, smoking may have potential effect modification because there may be unmeasured confounders between previous and current smokers. Therefore, we repeated the main analyses adjusted for pack-years categories of cigarette smoking (nonsmokers: having smoked zero pack-years; light smokers: fewer than 20 pack-years; and heavy smokers: 20 or more pack-years). Pack-years of cigarette smoking were calculated as the number of cigarettes smoked per day divided by 20 and then multiplied by the number of years of smoking [26]. Fifth, the information on both coffee and tea consumption and potential confounders were collected at the baseline; it is very difficult to assess the role of depression in these associations. We performed main analyses without adjusting for baseline depression. Furthermore, we also investigated coffee and tea consumption interaction on mortality in the above sensitivity analyses. All analyses were performed using R (version 3.6.3, R Foundation for Statistical Computing) and STATA 15 statistical software (StataCorp). The two-sided P < 0.05 was considered statistically significant.

In total, 498,158 participants (median (IQR) age: 58 [50-63]; 54% female) were included in this analysis. The proportion of < 1–2 cups/day of coffee and 2–4 cups/day of tea were 46.1% (n = 229,498) and 43.7% (n = 217,454), respectively. Additional file 1: Fig. S3 shows the distribution for the combination of coffee and tea consumption. Drinking both < 1–2 cups/day of coffee and 2–4 cups/day of tea was more prevalent accounting for the largest proportion with 23.0% (n = 114,629). Table 1 presents the baseline characteristics of participants according to coffee and tea consumption. Compared with non-coffee consumers, those who consumed < 1–2 cups/day of coffee were more likely to be older, male, white, non-drinker, more likely to have a university education level, high physical activity, healthy diet, and more likely to report “excellent” health; but they were less likely to be obese, non-smoker, and less likely to have diabetes, hypertension, and depression. Likewise, as compared to non-tea consumers, those who consumed 2–4 cups/day of tea were more likely to be older, male, white, non-drinker, more likely to have high physical activity, healthy diet, hypertension, and more likely to report “excellent” health, but they were less likely to be obese, non-smoker, and less likely to have a university education level, diabetes, and depression. Over 15 years of follow-up (median [IQR] length of follow-up, 12.1 [11.4–12.8] years; total person-years, 5,900,033), we documented 34,699 (7.0%) deaths. Among them, 6663 (1.3%) participants died from CVD, 6018 (1.2%) participants died from respiratory disease, and 2864 (0.6%) participants died from digestive disease.

In Fig. 1, the dose-response relationship of separate coffee and tea consumption with all-cause, CVD, respiratory disease, and digestive disease mortality were significantly nonlinear (all P-nonlinear < 0.001). A multivariable-adjusted model showed J-shaped associations between separate coffee consumption with all-cause, CVD, and respiratory disease mortality and separate tea consumption with respiratory disease mortality. The reverse J-shaped associations were observed for separate tea consumption with all-cause and CVD mortality. The threshold consumption for the lowest mortality risk was about 1 cup/day of coffee and 3 cups/day of tea. Particularly, reverse J-shaped associations were observed between separate coffee and tea consumption with digestive disease mortality, and participants who consumed about 5 cups/day of coffee and 6 cups/day of tea showed the lowest risk.

To quantify the magnitude of relationships between coffee and tea consumption with mortality, we categorized coffee (none, < 1–2, 3–4, and ≥ 5 cups/day) and tea (none, < 1–1, 2–4, and ≥ 5 cups/day) consumption groups according to prior validated results of restricted cubic spline analyses in the present study. After adjusting for potential confounders, < 1–2 cups/day of coffee was inversely associated with a lower risk of health outcomes: 0.91 (0.88–0.93) for all-cause, 0.94 (0.88–1.00) for CVD, and 0.84 (0.78–0.89) for respiratory disease mortality (Table 2). Participants who drank 2–4 cups/day of tea had a lower risk for all-cause (HR, 0.86; 95% CI: 0.83–0.88), CVD (HR, 0.88; 95% CI: 0.81–0.94), and respiratory disease mortality (HR, 0.80; 95% CI: 0.74–0.87). Additionally, we found that ≥ 5 cups/day of coffee (HR, 0.67; 95% CI: 0.59–0.77) and ≥ 5 cups/day of tea (HR, 0.65; 95% CI: 0.58–0.73) showed the lowest risk for digestive disease mortality, which was similar to the results of restricted cubic splines.

We further explored the relationship of the combined associations of coffee and tea consumption with mortality. We found significant interactions between coffee and tea consumption with all-cause and digestive disease mortality. (P for interaction < 0.01). As shown in Fig. 2 and Additional file 1: Table S2, for individuals with low coffee consumption, the increasing tea consumption was almost linearly associated with all-cause mortality, whereas for those with high coffee consumption, the relationship was U-shaped. However, the increasing coffee consumption showed a U-shaped relationship with all-cause mortality for those with low or high tea consumption. Specifically, compared with those who did not drink coffee and tea, participants who drank both < 1–2 cups of coffee and 2–4 cups of tea per day were associated with a 22% lower risk for all-cause (HR, 0.78; 95% CI: 0.73–0.85), 24% lower risk for CVD (HR, 0.76; 95% CI: 0.64–0.91), and 31% lower risk for respiratory disease mortality (HR, 0.69; 95% CI: 0.57–0.83). Nevertheless, the lowest HR (95% CI) of drinking both < 1–2 cup/day of coffee and ≥ 5 cups/day of tea for digestive disease mortality was 0.42 (0.34–0.53).

Associations of separate coffee and tea consumption with all-cause and cause-specific mortality did not differ by sex, age, BMI, physical activity, smoking status, alcohol intake frequency, dietary pattern, hypertension, diabetes, and depression (Additional file 1: Tables S3-S12). Figure 3 shows associations between the combined consumption of coffee and tea with mortality by subgroup analyses. Multivariable hazard ratios were calculated for total and cause-specific mortality among study participants who consumed < 1–2 cups/day of coffee and 2–4 cups/day of tea compared to those who drank neither coffee nor tea. In the analyses stratified by potential confounding factors for death, the inverse associations between combined consumption of coffee and tea with mortality were presented in all subgroups. In terms of strata of smoking status, the association was stronger among current smoking participants than among never and previous smoking participants (P for interaction = 0.01). Similarly, a multiplicative interaction was also found in the stratification of alcohol intake frequency (P for interaction = 0.002), and the HR of the group with low alcohol intake frequency was lower than that of those with high alcohol intake frequency. The details of the combined effects of coffee and tea consumption for each group were shown in Additional file 1: Tables S3-S12, and the results were not much altered with stratification by these factors.

The results were not apparently altered in the following sensitivity analyses. After excluding those who died within 3 years, the association between separate and combined coffee and tea consumption and all-cause and cause-specific mortality remained significant (Additional file 1: Table S13). When the analyses were performed following the exclusion of all participants with CVD and cancer at baseline, the results were not altered (Additional file 1: Table S14). When the analyses were conducted after adjusting for pack-years categories of cigarette smoking, the results were basically unchanged (Additional file 1: Table S15). The results also did not change materially when the main analyses were conducted without adjusting for baseline depression (Additional file 1: Table S16). Moreover, the significant interaction between coffee and tea consumption on all-cause and digestive disease mortality remained in the above analyses (P for interaction < 0.05). Furthermore, the findings were similar when we repeated the analysis using data that excluded the missing values of covariates rather than multiple imputations (Additional file 1: Table S17).