Coffee Drinking and Mortality in 10 European Countries: A Multinational Cohort Study

To the Editor,
We have read the recent studies by Gunter MJ et al. [1] and Park SY et al. [2] with great interest. Nevertheless, we believe there are two potential threats to the validity of both studies that are worth discussing.
Daily coffee intake usually begins around age 20, in the context of adult student life or the incorporation to the job market. Nevertheless, median age at baseline in both studies was above 50 years. Thus, individuals from the source populations had had the opportunity to be exposed to coffee for a median of 30 years before study entry. This “prevalent user” design may have allowed the depletion of susceptible individuals from the coffee intake groups (who may have had events before qualifying for study entry), resulting in a “healthy survivor” exposed study population. “Incident user” –i.e., exposure-naïve– designs are considered the gold standard in observational pharmacoepidemiology, as the selection bias derived from inclusion of prevalent users is minimized and the associations are more consistent with those from randomized trials [3]. Interestingly, in an analysis of the Precursors study by Klag and colleagues [4] in which baseline age was less than 30 (i.e., closer to an incident user design), there was a strong association between coffee use and cardiovascular events in multivariable analyses.
The risk of residual confounding is also worth mentioning. In particular, socially conditioned behaviors such as coffee intake are challenging and subject to confounding by socioeconomic status (SES). As with studies of moderate alcohol consumption, these behaviors are strongly prevalent in higher SES groups, which in itself is a protective factor for health outcomes [5]. While the authors did adjust for education, SES is a complex latent construct for which proxies (such as education, income, or occupation) are often measured with error, when at all measured, and may fail to fully capture the effect of SES. Importantly, in their sensitivity analysis assessing robustness to residual confounding, Gunter et al. [1] found that for a risk factor with a HR of 1.50 (within the realm of the feasible effect sizes of poverty [5]), there needs only be a 20% difference in the confounder between groups.
In view of these limitations, we believe the reports of these studies in the general press may have exaggerated their causal significance. Future research on coffee intake and health should aim at including younger individuals and use more comprehensive measurements of SES.

REFERENCES
1. Gunter MJ, Murphy N, Cross AJ, et al. Coffee Drinking and Mortality in 10 European Countries: A Multinational Cohort Study. Ann Intern Med. 2017 Jul 11. doi: 10.7326/M16-2945.
2. Park SY, Freedman ND, Haiman CA, et al. Association of Coffee Consumption With Total and Cause-Specific Mortality Among Nonwhite Populations. Ann Intern Med. 2017 Jul 11. doi: 10.7326/M16-2472.
3. Danaei G, Tavakkoli M, Hernán MA. Bias in observational studies of prevalent users: lessons for comparative effectiveness research from a meta-analysis of statins. Am J Epidemiol. 2012;175(4):250-62.
4. Klag MJ, Mead LA, LaCroix AZ, et al. Coffee intake and coronary heart disease. Ann Epidemiol. 1994;4(6):425-33.
5. Galea S, Tracy M, Hoggatt KJ, et al. Estimated deaths attributable to social factors in the United States. Am J Public Health. 2011;101(8):1456-65.

Gunter and colleagues report that people in the highest quartile of coffee consumption have a 7%-10% higher risk of death compared to non-drinkers in a study of over 450,000 adults from 10 European countries.1 However, this association was reversed after adjustment for covariates. In the multivariable adjusted model, high coffee consumption is related to a 12% lower mortality risk in men and a 7% lower risk in women. The authors suggest that coffee may confer health benefits.

The instability of the results in relation to covariate adjustments raised our concern about the validity of this conclusion. Here we elaborate on the possibility that the findings may be attributable to bias and confounding by unmeasured or imprecisely measured risk factors. For example, some diseases and symptoms related to hypersensitivity or intolerance to coffee consumption might also be linked to shorter life expectancy contributing to a spurious association between coffee consumption and mortality (NICE Evidence Search Caffeine. https://www.evidence.nhs.uk). Socioeconomic health determinants, such as income, occupational position and neighbourhood characteristics, are related to coffee consumption2 and mortality.2-4 Such factors may have affected the results but were not controlled for in the analyses. This interpretation is consistent with the result that the coffee consumption-mortality relation was more marked in men than women, as was the association between coffee consumption and education, another socioeconomic health resource.1

According to calculations of the e-value,5 a single confounder, related to coffee consumption and mortality risk, and with a hazard ratio of 1.5 could entirely explain the 12% lower mortality in men with high coffee consumption. In women, the corresponding hazard ratio would be 1.4. Thus, even minor uncontrolled risk and protective factors, including those listed above, have, in combination, the potential to bias the findings by Gunter and colleagues.1 Indeed, Mendelian randomisation analyses, which are more protected from bias and confounding than the observational findings of Gunter and others, do not support protective effects for high coffee consumption in relation to cardiovascular disease or all-cause mortality.3

The US NIH, with support from the alcohol industry, is starting a $100 million clinical trial to obtain a definitive answer to the question of whether moderate alcohol consumption prevents heart attacks. As coffee consumption is more widespread and possibly more amenable to change than alcohol consumption, a large-scale trial of coffee consumption is equally justified to obtain definite conclusions about its health effects.

Professor Mika Kivimaki and Dr Jane E. Ferrie, University College London, UK
Email: [email protected]

Conflict of interest: None declared.

REFERENCES

1. Gunter MJ, Murphy N, Cross AJ, Dossus L, Dartois L, Fagherazzi G et al. Coffee drinking and mortality in 10 European countries: A multinational cohort study. Ann Intern Med 2017 [ePub ahead of the print] doi:10.7326/M16-2945

2. Nordestgaard AT and Nordestgaard BG. Coffee intake, cardiovascular disease and all-cause mortality: observational and Mendelian randomization analyses in 95 000-223 000 individuals. Int J Epidemiol. 2016;45:1938-1952.

3. Stringhini S, Carmeli C, Jokela M, Avendaño M, Muennig P, Guida F et al. Socioeconomic status and the 25 x 25 risk factors as determinants of premature mortality: a multicohort study of 1.7 million men and women. Lancet 2017;389:1229-37.

4. Ludwig J, Sanbonmatsu L, Gennetian L, et al. Neighborhoods, obesity, and diabetes: a randomized social experiment. N Engl J Med 2011;365:1509–19.

5. vander Weele T, Ding P. Sensitivity analysis in observational research: Introducing the E-value. Ann Intern Med 2017; [ePub ahead of the print] doi:10.7326/M16-2607.

We thank the authors of the letters addressing our article and welcome further discussion on the study design and its results. We fully agree that the results from our analysis do not provide evidence for a causal relationship between coffee drinking and reduced risk of all-cause and cause-specific mortality. Unfortunately we are not able to control hyperbole generated by the lay press, though in our interactions with journalists we made every effort to stress the limitations of observational analyses. As suggested, a clinical trial could provide more definitive evidence on whether coffee drinking directly impacts health, and we would welcome such an initiative
Both letters raised the prospect that residual confounding by socioeconomic status (SES) may have contributed to the observed inverse relationship between coffee consumption and mortality. As indicated, we used education level as a proxy for SES and while we agree that education is an imperfect measure of SES, it is a widely used indicator of latent SES and is strongly associated with mortality rates in the EPIC cohort (1). Using education as a surrogate marker of SES allowed the classification of all participants (including those not working or retired), is stable over the lifecourse, and can be accurately recorded allowing comparisons across the EPIC study. The inverse associations between coffee and mortality observed in our analysis were stable after we adjusted for education, and, importantly, coffee drinking across the 10 European countries was not related to education status and was not more common among those with healthier lifestyle habits. Further, while coffee drinking may have become more prevalent in countries such as the United Kingdom in recent years and could track with SES, such changes have not been observed in most other countries included in the study.
We cannot exclude the possibility that survivor bias may have influenced the observed coffee consumption and mortality relationships. However, if survivor bias did significantly impact our results; we would expect the coffee drinking and mortality relationships to differ substantially by age. However, we did not detect any effect modification by age at recruitment categories (<45, 45-<60, and ≥60 years; P-interaction=0.94). Whilst this observation does not rule out an impact of survivor bias, it is an indication that any such impact is unlikely to be large.
While we fully acknowledge the limitations of our analysis and urge caution in the interpretation of the results, we also wish to emphasize the striking consistency between our findings and those of several other large-scale studies conducted in the U.S. (2-4) and Asia (5). The impact of coffee drinking on health deserves further investigation.

Yours Sincerely,
Marc J. Gunter, PhD; Neil Murphy, PhD; David C Muller, PhD; Elio Riboli, MD, ScM


1. Gallo V, Mackenbach JP, Ezzati M, Menvielle G, Kunst AE, Rohrmann S, et al. Social Inequalities and Mortality in Europe – Results from a Large Multi-National Cohort. PLOS ONE. 2012;7(7):e39013.
2. Freedman ND, Park Y, Abnet CC, Hollenbeck AR, Sinha R. Association of Coffee Drinking with Total and Cause-Specific Mortality. New England Journal of Medicine: Massachusetts Medical Society; 2012:1891-904.
3. Ding M, Satija A, Bhupathiraju SN, Hu Y, Sun Q, Han J, et al. Association of Coffee Consumption with Total and Cause-Specific Mortality in Three Large Prospective Cohorts. Circulation. 2015.
4. Park S, Freedman ND, Haiman CA, Le Marchand L, Wilkens LR, Setiawan V. ASsociation of coffee consumption with total and cause-specific mortality among nonwhite populations. Annals of Internal Medicine. 2017;167(4):228-35.
5. Saito E, Inoue M, Sawada N, Shimazu T, Yamaji T, Iwasaki M, et al. Association of coffee intake with total and cause-specific mortality in a Japanese population: the Japan Public Health Center–based Prospective Study. The American Journal of Clinical Nutrition. 2015;101(5):1029-37.