Researchers publishing in JAMA Internal Medicine
Prof Kevin McConway, Emeritus Professor of Applied Statistics at The Open University, said:
“Several previous studies have observed that drinking coffee seems to reduce death rates, at least for moderate levels of coffee consumption. Why do we need another one? Really it’s a response to a difficulty in interpreting the results from such studies. The studies, including this new one, are observational. That is, they don’t involve experiments that make people drink different amounts of coffee and see what happens. Instead they find a group of people, record how much coffee each one drinks, and follow them up (for many years, usually) and record deaths.
“If death rates are lower in coffee drinkers, it might be because of something in the coffee, or it might be because coffee drinkers differ from coffee-abstainers in other ways than coffee drinking, and these other ways are what lead to the differences in death rates and not the coffee at all. This can be allowed for statistically up to a point, but only for characteristics on which the researchers have data. The new study couldn’t avoid this problem, because it’s also observational. But another issue in observational studies it can be difficult to investigate more subtle aspects, such as whether particular characteristics of the participants or their coffee drinking have particular effects on death rates, because the relevant aspects might be hard to measure or because there are not sufficient participants with the relevant characteristics.
“This new study is good, in my view. It’s carefully performed and reported. It included over half a million people and was able to use reasonably detailed information on their coffee drinking, on various potentially relevant lifestyle factors and characteristics, and, interestingly, on certain genes that are known to affect the way caffeine is processed in the body. So it does help, to some extent, with investigating what it is about coffee drinking that might affect death rate.
“The results are interesting in several ways – for example, the genetic factors don’t seem to be relevant after all to the relationship between coffee consumption and death rates, and (in contrast to some other studies) the effects on death rates are still there for people consuming quite a lot of coffee, 8 or more cups a day. But the study is still observational, and that’s why the authors explicitly (and rightly) say that their results “should be interpreted with caution”. They can’t do more than suggest what might or might not be important and relevant.
“Taking that on board, though, there are still some interesting findings. Some of the new results suggest that other components of coffee rather than caffeine may be important in any relationship between coffee drinking and death rates. For instance, those who reported that they usually drink decaffeinated coffee also had rather lower death rates than those who drank no coffee at all, though the effect was not as strong as for ground coffee drinkers. However, the observational nature of the study is relevant here too as well – the authors point out that the difference may be due to some differences in personal characteristics, between people who drink different types of coffee, that they could not allow for statistically.”
A couple of explanatory points:
“The study did include over 500,000 participants, but the key limitation in such studies is not the total number of people, but the number of deaths. The participants were not particularly old – the oldest was 73 – and there were only 11,000 deaths. That’s definitely enough to support their conclusions, but not big enough to look at possible effects on detailed specific causes of death.
“The study used data from UK Biobank. Many Mendelian randomisation studies also use data from that source, because it contains lots of genetic and genomic data. Mendelian randomisation is a way of studying whether associations are causal, even if they are based on observational data. But this new study does not involve Mendelian randomisation or any other kind of causal inference, so all the usual provisos about observational data do definitely apply to it.
“Of course, drinking coffee doesn’t reduce anyone’s eventual chance of death – it comes to us all eventually. The results of the study are in terms of hazard ratios, that is, in reductions (or increases) in the risk of death over a given period of time.”
* ‘Association of Coffee Drinking With Mortality by Genetic Variation in Caffeine Metabolism Findings From the UK Biobank’ by Erikka Loftfield et al published in JAMA Internal Medicine on Monday 2 July 2018.
Prof Kevin McConway is a member of the Science Media Centre’s Advisory Committee.