Association of Coffee Consumption With Total and Cause-Specific Mortality Among Nonwhite Populations

Follow-up time for each participant was calculated from the date of cohort entry to the date of death or the end of follow-up (31 December 2012). Hazard ratios (HRs) and 95% CIs were calculated for mortality associated with coffee intake, using Cox proportional hazards models with age as the underlying time metric. Cox models were first adjusted for the strata variables of age at cohort entry (<50, 50 to 54, 55 to 59, 60 to 64, 65 to 69, 70 to 74, or ≥75 years), sex, and race/ethnicity. They were further adjusted for smoking using a detailed model to account for smoking cessation over time, which was common but varied by race/ethnicity (15). Smoking variables included ethnicity, smoking status, average number of cigarettes, squared average number of cigarettes, number of years smoking (time-dependent), number of years since quitting (time-dependent), and interactions between ethnicity and smoking status. The Cox models were then further adjusted for the following covariables in the log-linear component: BMI (<18.5, 18.5 to 22.9, 23.0 to 24.9, 25.0 to 29.9, 30.0 to 34.9, or ≥35.0 kg/m²), education (high school or less, vocational school or some college, or college graduate), vigorous physical activity (for men, <0.1, 0.1 to <0.25, 0.25 to <0.80, or ≥0.80 hour per day; for women, <0.1, 0.1 to <0.25, 0.25 to <0.5, or ≥0.5 hour per day), alcohol (ethanol) consumption (for men, 0, 1 to <5.2, 5.2 to <23, or ≥23 g/d; for women, 0, 1 to <2.5, 2.5 to <10, or ≥10 g/d), total energy intake (<1317, 1317 to 1717, 1718 to 2159, 2160 to 2835, or ≥2836 kcal/d), energy from fat (<23.7%, 23.7% to 28.0%, 28.1% to 31.8%, 31.9% to 35.9%, or ≥36.0%), and preexisting chronic disease at baseline (self-reported heart attack or angina, stroke, diabetes, or high blood pressure, and cancer that was self-reported or ascertained from tumor registries).

We performed tests for linear trend by entering the median value of each category of coffee consumption as a continuous variable in the models. A competing risk analysis was done to compare the association with coffee intake among causes of death, where each cause was simultaneously modeled as a different event (16). We used a Wald test to compare the parameter estimates between causes of death. The proportional hazards assumption over all ages was tested by modeling the interaction of age with coffee consumption, and no violation was found. Direct adjusted mortality curves by category of coffee consumption were computed over the age range from the Cox model as the average of the model-based curves at observed profiles for the following covariates (17, 18): age at cohort entry; sex; ethnicity; smoking status; average number of cigarettes; squared average number of cigarettes; number of years smoking; number of years since quitting; interactions between ethnicity and smoking status, average number of cigarettes, squared average number of cigarettes, and number of years smoking; BMI; education; physical activity; alcohol consumption; total energy intake; energy from fat; and preexisting illness.

We did subgroup analyses by race/ethnicity, sex, age at cohort entry, and smoking status. Heterogeneity across subgroups was tested using the Wald statistics for cross-product terms of trend variables and subgroup membership. To assess potential reverse causality, we performed sensitivity analysis by excluding deaths that occurred within the first 5 years of follow-up. We also did a formal sensitivity analysis, as described by Lin and colleagues (19), to assess the potential effect of unmeasured confounding on our results. For all analyses, we used SAS, version 9.4 (SAS Institute). All P values were 2-sided.

The MEC was supported by the National Cancer Institute, which had no role in the study design; recruitment; data collection and analysis; or preparation, approval, or publication of the manuscript.

About 16% of participants did not drink coffee, 7% drank 1 to 3 cups per month, 13% drank 1 to 6 cups per week, 31% drank 1 cup per day, 25% drank 2 to 3 cups per day, and 7% drank 4 or more cups per day (Table 1). Participants who drank more coffee were more likely to be younger, male, and white and to drink more alcohol. There was a strong correlation between higher coffee consumption and smoking status; among non–coffee drinkers, 58% were never-smokers, but among those who drank 4 or more cups per day, 26% were never-smokers. The proportion of preexisting illness at baseline was lower among participants who drank coffee more frequently.

During an average of 16.2 years of follow-up (3 195 484 person-years), we identified 58 397 deaths in the cohort. In the age-, sex-, and ethnicity-adjusted analysis, consuming 4 or more cups of total (caffeinated and decaffeinated) coffee per day was associated with a higher risk for all-cause death (Table 2). However, after adjustment for smoking and other covariates, we saw significant inverse associations between increasing coffee consumption and all-cause mortality. The HR for death (vs. no coffee) was 1.00 (95% CI, 0.95 to 1.05) for 1 to 3 cups of coffee per month, 0.97 (CI, 0.93 to 1.01) for 1 to 6 cups per week, 0.88 (CI, 0.85 to 0.91) for 1 cup per day, 0.82 (CI, 0.79 to 0.86) for 2 to 3 cups per day, and 0.82 (CI, 0.78 to 0.87) for 4 or more cups per day. Exclusive consumption of either caffeinated coffee (P for trend < 0.001) or decaffeinated coffee (P for trend = 0.008) was inversely associated with total mortality. Associations were similar in men and women (Appendix Table 1).

The adjusted mortality rates by different levels of coffee consumption are presented in the Figure. The inverse association applies to consumption of at least 1 to 6 cups of caffeinated coffee per week and seems to be monotonic with increased consumption.

The results of the sensitivity analysis indicated that the associations between coffee consumption and mortality were robust to unmeasured confounders, except in the case of a strong unmeasured confounder, on the order of smoking or diabetes, that was substantially associated with coffee drinking (Appendix Table 2).

We examined the association of coffee consumption and total mortality across the 5 racial/ethnic groups in the MEC (Table 3). We found no indication that the associations varied by race/ethnicity (P for heterogeneity = 0.166). Coffee consumption was significantly associated with lower risk for death in African Americans (P for trend = 0.001), Japanese Americans (P for trend < 0.001), Latinos (P for trend = 0.002), and whites (P for trend = 0.003). Although similar patterns were seen for Native Hawaiians, these associations did not reach statistical significance (P for trend = 0.141).

The 10 leading causes of death in the United States accounted for 81% of deaths in the MEC, with cardiovascular disease (36%) and cancer (31%) accounting for the largest proportions. We investigated the associations between coffee consumption and each of the 10 leading causes of death (Table 4). Coffee consumption was inversely associated with risk for death due to heart disease (P for trend < 0.001), cancer (P for trend = 0.023), chronic lower respiratory disease (P for trend = 0.015), stroke (P for trend < 0.001), diabetes (P for trend = 0.009), and kidney disease (P for trend < 0.001). There was no significant association between coffee consumption and deaths from influenza and pneumonia, Alzheimer disease, accidents, and intentional self-harm. In a competing risk analysis, the association of mortality with coffee consumption did not statistically differ among causes of death (P = 0.92).

Smoking is a known strong confounder of the coffee–mortality association. We therefore examined the associations between coffee consumption and mortality stratified by smoking status (Table 5). In never-smokers, former smokers, and current smokers, coffee consumption was inversely associated with total mortality (P for heterogeneity = 0.120).

We conducted a stratified analysis by baseline health status (Appendix Table 3). We found significant inverse associations of coffee consumption with total mortality among participants with self-reported heart disease and cancer at baseline. Among participants without self-reported heart disease and cancer at baseline who were never-smokers, we saw similar inverse associations between coffee consumption and total mortality. We conducted stratified analysis by age at cohort entry (<55, 55 to 70, and >70 years) and education level (high school or less, vocational school or some college, and college graduate) (Appendix Table 4). Inverse associations were found in all subgroups (P for trend ≤ 0.014 for each), with a stronger association in younger participants (P for heterogeneity = 0.019). In sensitivity analyses excluding deaths within the first 5 years of follow-up, the inverse associations between total coffee consumption and total mortality remained statistically significant.

In this large multiethnic population, we found an inverse association between coffee consumption and total mortality after adjustment for smoking and other potential confounders. Similar trends were observed for caffeinated and decaffeinated coffee. Among the 10 leading causes of death in the United States, mortality was inversely related to coffee consumption for heart disease, cancer, respiratory disease, stroke, diabetes, and kidney disease. Inverse associations were seen in never-smokers, former smokers, and current smokers; 4 of 5 racial/ethnic groups; those with and without preexisting heart disease and cancer; those with different education levels; and all age groups.

Overall, our finding of decreased risk for all-cause death with coffee consumption is consistent with meta-analyses (8, 20–22) and subsequent cohort studies done in the United States (9, 10), Europe (23, 24), and Japan (25). Risk reduction for all-cause death in our study was 18% (CI, 14% to 21%) for 2 to 3 cups versus 0 cups of coffee per day, which was similar to previous reports from other large U.S. cohorts. The Prostate, Lung, Colorectal and Ovarian Cancer Screening Trial (10) saw a risk reduction of 18% (CI, 12% to 23%), and the National Institutes of Health–AARP Diet and Health Study (26) found risk reductions of 10% (CI, 7% to 14%) in men and 13% (CI, 8% to 17%) in women. In a pooled analysis of the Nurses’ Health Study, Nurses’ Health Study II, and Health Professionals Follow-up Study, drinking more than 1 to 3 cups of coffee per day was associated with 9% (CI, 5% to 12%) lower risk for all-cause death than drinking no coffee (9).

However, until now, few data have been available on the association between coffee consumption and mortality in nonwhites in the United States and elsewhere. Such investigations are important because lifestyle patterns and disease risks can vary substantially across racial/ethnic backgrounds, and findings in one group may not necessarily apply to others. On the other hand, findings that persist across different racial/ethnic groups have greater biological plausibility. Because of the unique opportunity that the MEC provides, we were able to investigate the coffee–mortality associations in participants from 5 racial/ethnic backgrounds using the same method and as part of the same cohort. We saw similar inverse associations between coffee drinking and mortality in 4 of the 5 racial/ethnic groups; the association in Native Hawaiians did not reach statistical significance, likely owing to fewer participants in that group.

Our findings for specific causes of death were generally similar to those of previous cohorts (9, 10, 26). One difference was for cancer mortality. Several recent cohort studies (9, 10, 23, 25) and 2 meta-analyses (20, 21) of prospective studies reported no significant association between total cancer mortality and coffee consumption, although other studies have reported inverse associations (24). We saw a modest reduction in risk for cancer death (8% lower risk with 2 to 3 cups vs. 0 cups per day). One advantage of our study was our ability to comprehensively adjust for cigarette smoking, a strong cancer risk factor, which may have allowed us to observe an inverse association with cancer mortality. We note that our findings persisted across several strategies for controlling for smoking, including analyses by smoking subgroup and models that adjusted for smoking information collected only at baseline (smoking status, years smoking, and number of cigarettes).

We also examined whether associations varied by age at enrollment. Two previous studies have reported inconsistent associations between coffee drinking and mortality in younger people. One cohort reported increased risk for all-cause death among persons younger than 55 years who drank more than 4 cups per day (27), whereas a second study found no association among women younger than 50 years (23). We found inverse associations across categories of age at enrollment in our cohort, although the association seemed stronger in younger participants than in older ones. Hazard ratios among younger participants (<55 years) in the cohort were 0.85 (CI, 0.77 to 0.95) for 1 cup, 0.73 (CI, 0.66 to 0.82) for 2 to 3 cups, and 0.74 (CI, 0.64 to 0.85) for 4 or more cups per day. Therefore, our findings do not suggest a harmful effect of coffee consumption among younger adults.

Coffee contains many biologically active chemicals, including caffeine and phenolic compounds, that may impact health by means of various mechanisms, such as antioxidant and antimutagenic capacity, insulin sensitivity, liver function, and chronic inflammation (28–31). We observed similar associations for both caffeinated and decaffeinated coffee, as seen in most other studies, suggesting that these associations may be related not to caffeine but to the many other compounds found in coffee. However, future mechanistic work is needed to characterize the many compounds in coffee and elucidate possible physiologic effects.

Our study has several key strengths, including its prospective design, large sample size (>185 000 participants with >58 000 deaths), relatively long follow-up (an average of 16 years), and inclusion of substantial numbers of participants from various racial/ethnic backgrounds. To reduce the effect of reverse causality, we excluded deaths within the first 5 years of follow-up in sensitivity analyses and also restricted the analyses to participants who did not report preexisting disease. We had extensive information on potential confounders (for example, smoking, alcohol use, physical activity, and preexisting illness) because of our comprehensive baseline questionnaire.

Our study also has limitations. As with any observational study, we cannot exclude the possibility of residual or unmeasured confounding. However, we are reassured by similar findings within different racial/ethnic populations in our cohort and in previous studies in the United States, Europe, and Japan. Also, a formal sensitivity analysis showed that the coffee–mortality association was robust to all but very strong levels of unmeasured confounding, which seems unlikely. On the other hand, self-reported coffee consumption at baseline is subject to measurement error, and consumption might have changed throughout follow-up. Among the participants who responded to the repeated QFFQ in 2003 to 2007 (n = 84 170) (average 11.0 years between measurements), the intraclass correlation coefficient between the 2 QFFQs was 0.60, which reflects potential changes over time or misclassification of coffee consumption. In the subset of participants with follow-up data on coffee consumption, the analysis using a time-dependent model yielded similar results. Because the highest level of coffee consumption in the QFFQ was 4 or more cups per day, we were not able to examine associations with higher levels. A recent meta-analysis of 15 prospective studies found similar associations for 4 cups per day and 5 or more cups per day (21), although the numbers of participants drinking more than 5 cups per day in epidemiologic studies have typically been small. Future studies are needed to examine associations in the heaviest coffee drinkers to address potential toxicity at high doses. In addition, we lacked information on preparation, which may affect the composition and bioavailability of coffee compounds (32).

In summary, higher coffee consumption was associated with lower risk for all-cause death and death from heart disease, cancer, respiratory disease, stroke, diabetes, and kidney disease. Inverse associations were found in African Americans, Japanese Americans, Latinos, and whites; never-smokers, former smokers, and current smokers; those with preexisting heart disease or cancer; and healthy participants. Our findings support the recent dietary guidelines from the U.S. Department of Agriculture (33), which indicate that moderate coffee consumption can be integrated into a healthy diet and lifestyle, by confirming an inverse association with mortality and suggesting that association’s generalizability to different racial/ethnic groups.