Association of Coffee Drinking with Total and Cause-Specific Mortality

Image credit: Mighty Optical Illusions (moillusions.com)
I'd normally put something like this in the weekend links roundup, but it seemed particularly salient for FE readers.
Association of Coffee Drinking with Total and Cause-Specific Mortality
Freedman et al., NEJM 2012
In this large, prospective U.S. cohort study, we observed a dose-dependent inverse association between coffee drinking and total mortality, after adjusting for potential confounders (smoking status in particular). As compared with men who did not drink coffee, men who drank 6 or more cups of coffee per day had a 10% lower risk of death, whereas women in this category of consumption had a 15% lower risk. Similar associations were observed whether participants drank predominantly caffeinated or decaffeinated coffee. Inverse associations persisted among many subgroups, including participants who had never smoked and those who were former smokers and participants with a normal BMI and those with a high BMI. Associations were also similar for deaths that occurred in the categories of follow-up time examined (0 to <4 years, 4 to <9 years, and 9 to 14 years). 
Our study was larger than prior studies, and the number of deaths (>52,000) was more than twice that in the largest previous study. Whereas the results of previous small studies have been inconsistent, our results are similar to those of several larger, more recent studies, including the Health Professionals Follow-up Study and the Nurses' Health Study.
Given the observational nature of our study, it is not possible to conclude that the inverse relationship between coffee consumption and mortality reflects cause and effect. However, we can speculate about plausible mechanisms by which coffee consumption might have health benefits. Coffee contains more than 1000 compounds that might affect the risk of death. The most well-studied compound is caffeine, although similar associations for caffeinated and decaffeinated coffee in the current study and a previous study suggest that, if the relationship between coffee consumption and mortality were causal, other compounds in coffee (e.g., antioxidants, including polyphenols) might be important.
In summary, this large prospective cohort study showed significant inverse associations of coffee consumption with deaths from all causes and specifically with deaths due to heart disease, respiratory disease, stroke, injuries and accidents, diabetes, and infections. Our results provide reassurance with respect to the concern that coffee drinking might adversely affect health.
Please feel free to discuss feasible instruments for coffee consumption in the comments (or not). (via bb)


  1. IV: rain in coffee-growing countries, just like usual, right?

  2. I think you'd want something that instruments for picking up the habit of drinking coffee, rather simple variation in dosing. E.g., your college roommate was randomly assigned, and he / she drank coffee so you started doing the same. No?

  3. This study participants and nature of counting variables makes conclusions very difficult. The study participants were retired people aged 50-71 years old who had been exposed to many potential harmful exposures throughout one's lifetime. Coffee drinking in the participants also has strong correlations to unhealthy dietary habits such as increased alcohol binging, tobacco use and consumption of red meat. It therefore makes it very difficult to isolate the effects of coffee on the cause-specific mortality.
    If this isn't enough, the study is purely observational and so it does not reflect causality and is possible an association.
    The only positive outcome from this study is that there was limited to no evidence of a harmful effect of coffee intake and so this give assurance that coffee isn't killing us!


  4. The paper definitely has identification issues, but I think it's fairly open about it; hence "Association", right? The issue is I think this is one of those classic problems where clean identification really isn't easy, since we're talking about the cumulative effect of a lifetime habit. So it might be the case that at least at the population level this is as good as we're going to get with current data.

    Or not. Someone might think of a decent identification strategy (see the roommate thing above).