Introduction
Last year I made a post on "The Health Benefits of Coffee".
As I did not have academic access to view full papers at the time this was based on a combination of abstracts, a Medscape presentation (by Dr Bret S. Stetka) and my own notes.
Shortly after my post a new meta-analysis on Coffee and Mortality by Giuseppe et. al was released but I had no means of legally accessing it.
Luckily I have found a subscription service called DeepDyve that allows access to selected journals for a monthly fee and they have the study!
It is called :
I would like to summarise and discuss the findings which I think will be of interest to those of us who are coffee lovers.
I will try to keep it brief!
Before We Start: Meta-Analysis - What is it and Why Should We Care?
A meta-analysis is a type of study that combines the results of multiple studies.
The basic theory is that by combining lots of smaller data sets you can derive a larger data set for statistical analysis, thereby improving the validity of your results.
Sample size is one of the frequent issues in certain types of scientific research. Meta-analyses allow you to simulate a larger sample size than you would otherwise have available to you.
There are some problems with this approach but the TLDR is that in theory a meta-analysis gives you more "bang for your buck" using existing data than those single studies on their own.
The Findings - A TLDR Summary
Drinking 4 cups of coffee per day REDUCES:
- All cause mortality (i.e. death from all causes) with a RR (Relative Risk see explanation below) of 0.84
- CVD (Cardiovascular Disease) mortality with a RR of 0.83
- CHD (Coronary Heart Disease) mortality with a RR of 0.84
- Stroke mortality with a RR of 0.72
NB: Figures are taken from the raw data in the author's tables (2 and 3) for the RR of drinking 4 cups of coffee, rather than the up to 4 cups of coffee measure in the abstract as I could not find the full comparative data for the latter measurement.
Cancer Mortality Results, Smoking vs Non Smoking
However 4 cups of coffee slightly increases Cancer Mortality with a RR of 1.01 - this is believed to be an artifact of the association between smoking and coffee drinking:
Excluding Smokers Seems to Restore the Pattern
Considering non-smokers ONLY reduction in mortality seems to be consistent with the other results, in that 4 cups per day reduces mortality due to Cancer with a RR of 0.9.
Quick Explanation of RR/Relative Risk
This is a measure that is used to give a probability for something occurring.
For example when we say that 4 cups of coffee reduce all cause mortality/death with a RR of 0.84 it means that compared to someone who drinks no coffee you have only 84% of the risk of death.
So 4 cups of coffee per day reduce your risk of death by 16%.
Summary of Methods
Studies were found using electronic databases i.e. PubMed, EMBASE.
I personally use PubMed as it is publicly available for free. I believe EMBASE requires a subscription.
The Inclusion Criteria for studies were:
- Prospective design - i.e. the study followed people over time, rather than looking back at historical data.
- Evaluated association between coffee drinking and mortality - obviously since this is what it is trying to study!
- Assessing and reporting of Hazard Ratios and 95% Confidence Intervals for mortality > 3 exposure categories - a minimum baseline needed to be set to allow proper comparison.
- Studies provided data on the number of coffee cups consumed.
The Exclusion Criteria for studies were:
- Insufficient Reported Statistics - rather obvious reasons!
- Assessed Composite Outcome - the meaning is not clear in the paper (to me at least). I think they are referring to a lack of a clear outcome in e.g. rather than specifying death by CVD, a study may cite an even like MI(heart attack).
- Where more than one study or result came from the same cohort (group) of people studied, only the study with the entire cohort and/or longest follow up was used.
Sampling Data:
- 470 studies screened but only 31 Studies were used, with smaller samples for specific causes of mortality (see table).
- It is common to exclude studies based on various criteria, see discussion later.
- 1.6 million participants.
- 183,991 total deaths (all causes).
- 34,574 CVD deaths.
- 40,991 Cancer deaths.
Discussion
Overall the inclusion and exclusion criteria make sense. Due to the nature of how meta-analyses work it is uncommon for any two studies to be exactly identical.
The inclusion and exclusion criteria enable a certain degree of uniformity between the studies that you use.
Further although 31 studies were used in total it should be noted that different numbers of studies were used for analysis of each form of death.
This is because not every study used looked at every form of death we are considering here.
I have constructed a quick table to summarise this:
The Number of Studies Used For Each Mortality Type
| Mortality | No. of Studies Used | No. of Deaths | Total Sample Size |
|---|---|---|---|
| All cause | 24 | 183,991 | ? |
| CVD | 23 | 34,574 | ? |
| CHD | 12 | ? | ? |
| Stroke | 9 | ? | ? |
| Cancer | 15 | 40,991 | ? |
Unfortunately I did not find a summary of the (death) sample sizes that this breakdown represented beyond the figures already stated for all cause, CVD and cancer.
Further we do not know the TOTAL sample sizes (people who lived and died) for each form of death.
For example the total sample size of 1.6 million participants represents all 31 studies used, but none of the mortality samples use all 31 studies (e.g. all cause only uses data from 24 studies).
The authors have published a huge multi-table analysis which includes every study but does not include separate stroke/CHD data.
I suspect the lack of data for Stroke and CHD is due to lack of space but there may be other reasons.
Having specific sample sizes may have provided some additional useful information.
Further the large discrepancy between the sample for all cause mortality and the other causes that are listed is likely down to some of the studies only noting mortality rather than specific causes.
Geographic Breakdown of the Studies
| Location | No. of Studies |
|---|---|
| Europe | 14 |
| USA | 11 |
| Asia | 7 |
As we can see the distribution of studies varied geographically in a non uniform manner i.e. the majority were in Europe and the USA.
This could cause distortion of results due to variation in environmental and genetic factors.
It would be useful to be able to see if there was a difference in RR based on geographical location.
The authors do briefly mention this in their discussion.
Smoking
The issue of smoking and not smoking seems to have a massive effect on the benefits of coffee drinking as it applies to cancer mortality risk.
It would seem that adding coffee drinking may actually increase your risk of developing cancer if you also smoke.
This is contrary to the finding of decreases in mortality from the other causes that are noted in smoking.
Figure 3 illustrates these points very well in graphic form:
Before we get too excited with these results we should note that some of the previous caveats in terms of sampling apply here.
The data considered for smoking and non smoking is considerably smaller in terms of number of studies. I have summarised this in the table below:
Studies Used for Comparing Smoking and Non Smoking
| Mortality | Smoking (n=studies) | Non Smoking (n=studies) |
|---|---|---|
| All cause | 5 | 5 |
| CVD | 4 | 5 |
| CHD | 2 | 3 |
| Stroke | 2 | 3 |
| Cancer | 4 | 5 |
The smaller number of studies highlights an important problem with meta-analyses, that of failing to control for confounding factors.
The vast majority of the studies used did not differentiate between smokers and non smokers.
Even those that did may not have done so adequately and that may be responsible for the discrepancy between the number of studies examined for smokers vs non-smokers.
This inevitably reduces the statistical strength of the findings in this specific instance.
Meta-analyses are a fantastic tool but they are not perfect. A well designed study of equivalent size could in theory give you more statistically valid results.
Unfortunately money, time and other resource limitations mean that in most cases that is an impossible dream to achieve.
Why Would Coffee Increase Cancer in Smokers But Reduce it in Non-Smokers?
As to why smoking would seem to not only negate the benefits of coffee in relation to cancer but actually cause a reversal it is hard to say without more information.
One hypothesis I can think of is that drinking hot drinks may irritate the oesophagus (gullet).
In normal individuals this may not have as significant effect but in smokers they may already have an existing irritation of the oesophagus as a result of smoking.
This may result in a greater sensitivity to developing oesophageal cancers.
Further there may be some as yet unknown interaction between smoking and coffee that has a carcinogenic effect in combination.
It would be interesting to see if the higher risk in smokers translates to cancers in general or a specfic subset such as oesophageal cancers.
Conclusion
This is a fascinating study which supports the growing body of evidence that coffee is good for us.
As I have discussed there are some problems with the data and these are common issues with meta analyses.
I don't believe that these should detract from the main finding that coffee drinking reduces mortality from all causes.
When it comes to smoking and coffee drinking (in relation to cancer mortality) I think we need a lot more research before we can say what is going on with any degree of certainty.
If I was a smoker I wouldn't give up coffee drinking but would use it as an excuse to try to stop smoking.
Although the cancer risk seems to increase according to these results there are other benefits such as to CVD health that are maintained in smokers.
I suppose it depends on what method of death you fear more.
Anyway I don't need an excuse to continue enjoying my coffee drinking!
References
- Grosso, Giuseppe, Agnieszka Micek, Justyna Godos, Salvatore Sciacca, Andrzej Pajak, Miguel A. Martínez-González, Edward L. Giovannucci, and Fabio Galvano. 2016. “Coffee Consumption and Risk of All-Cause, Cardiovascular, and Cancer Mortality in Smokers and Non-Smokers: A Dose-Response Meta-Analysis.” European Journal of Epidemiology 31 (12): 1191–1205.
Thank you for reading
Before you go have you filled in the Coinbase form to list STEEM? It only takes a few seconds. THIS POST shows you how.
If you like my work please follow me and check out my blog - 
Tables were constructed by msyself using the data in the paper. Figure 3 was taken from the paper an all uncredited images are taken from my personal Thinkstock Photography account. More information can be provided on request.
Are you new to Steemit and Looking for Answers? - Try:
