What’s new?
Genome-wide association studies (GWAS) and hidden statistical flaws invalidate most popular health articles about the coffee-cognition link, regardless of whether coffee gets good or bad press.
Why It Matters
For far too long we have been exposed to conflicting messages about the health effects of coffee. The research into the world’s most consumed beverage (by raw-product weight) needs a critical second look. Its remarkable result: Coffee is no medicine, unless your lifestyle makes it so.
What’s Next
Apply the findings of this post to the next article you read about the alleged health effects of coffee (or any other food item).
Your Takeaways
- A clear (and lay-friendly) understanding of the statistical flaws of coffee-health research.
- A downloadable library of cognitive tests as applied in research.
- Links to relevant and reliable internet resources.
Sturgeon’s Law
Says 90% of everything is crap. That’s Sturgeon’s law named after the science fiction writer who proposed it in the 1950s. Today, it’s also the principle the media try very hard to live up to. No less so when it comes to health news.
Now here’s their latest edition, the coffee-cognition conundrum, the one where the headlines oscillate between a ‘too-much-coffee-turns-you-into-a-blithering-idiot’ meme and its exact opposite. If at least one of those was true, Sturgeon’s 90% should be revised to 50%. However, I will show you how the 90% survive when you do what the journalists should have done: Critically check the science before committing pen to paper.
We Love Coffee. Really.
Mankind devours 10.7 million tons of coffee per year (dry-weight roasted coffee beans). If you were to line up the traders’ standard 60-kg bags side by side they’d fit around the globe — two and a half times. Expressed as average per-capita consumption, each human on Earth, from newborn to supercentenarian, consumes 1.3 kg of the roasted beans every year. Given this degree of popularity, it is no surprise that scientists seek to uncover the java’s effects on all aspects of health. And one thing has become abundantly clear: Equating coffee with caffeine is far too simplistic. Coffee is one of the most concentrated sources of polyphenols and other micronutrients. Multiply that with the humongous amount we consume and you realize that:
“At the level of populations in the developed world, coffee provides more dietary phenolic phytochemicals than vegetables and fruits.”
So, let’s have a quick look at those nutrients in the next section before we dissect the two health messages of the above image.
Coffee’s Cognition Components
Only three of the 250 different coffee species end up in your brew: Predominantly Coffee arabica (Arabica coffee) and Coffea canephora (Robusta coffee), and to a much lesser extent Coffee liberica (Excelsa coffee). What they all have in common is a high concentration of health-promoting micronutrients.
Caffeine
This dominant physiologically active compound in coffee is a so-called alkaloid. I’m not throwing chemistry terms at you which you could do very well without. In this case, though, you might want to remember alkaloids as the type of substances that plants learned to produce during evolution as a protective because unpalatable (bitter-tasting), defense against their herbivorous animal enemies.
Then our species entered the scene. In Mother Nature’s arms race, we not only became tolerant to those alkaloids, we get our thrills from them. Think nicotine, morphine, and cocaine. But also piperine, which gives black pepper its pungency.
Humans absorb caffeine quickly, with peak concentration in the blood occurring within 30–120 minutes. Its half-life for excretion is between 2.5 and 5 hours. Neither age nor concomitant drug or stimulant use seem to have any appreciable effect on absorption and excretion.
Caffeine also diffuses to all body fluids and organs. It easily crosses the blood-brain barrier, which is why we love coffee in the first place. It also gets through the placenta, which you may want to keep in mind during pregnancy. In that situation, limiting daily caffeine intake to less than 300 mg is the recommendation of a well-designed and comprehensive Canadian review. But others advise more caution.
Studies trying to isolate the effects of caffeine on cognitive performance suggest that (a) chronic consumption has some protective effects against the onset and progression of Alzheimer’s disease, and (b) caffeine is a cognitive normalizer rather than a cognitive enhancer. In other words, no caffeine-loaded drink will turn you into a cognitive superhero. It just gets you to your standard normal self.
Caffeic Acid (CA)
Caffeic acid (CA) is a polyphenol with a strong antioxidant capacity. It’s also the main representative of the group of hydroxycinnamic acids. CA not only occurs in coffee (>80 mg/100 ml) but even more so in black chokeberries, some herbs (e.g. spearmint, thyme, sage), and spices (cinnamon, star anis). These data are taken from the phenol explorer database.
Aside from its role as an antioxidant, in vitro studies have shown caffeic acid to have anti-bacterial, anti-viral, and anti-proliferative (in cancer) qualities. In vivo experiments demonstrated improved memory and cognitive function in rodent models of Alzheimer’s disease (AD).
CA-treated AD-mice performed significantly better in memory tasks than their untreated AD-peers. Expressed as mg CA/kg body weight (assuming a CA content of 80mg/100ml), a 75-kg human would need to drink roughly half a liter of coffee to get the equivalent amount as the CA-treated mice. That’s within every java junkie’s reach.
By the way, translating dosages from animal models to humans is not as simple as scaling by body weight. Fortunately, you can also get high doses of caffeic acid (for which there exists no upper intake limit recommendation) from other sources such as fruits (cranberries, huckleberries, olives) and vegetables (Swiss chard).
Chlorogenic Acid (CGA)
Contrary to what you might think, chlorogenic acid does not contain chlorine. The “chloro” part of the name has the same Greek root as the “chloro” in chlorophyll (the green pigments of plants). It simply means ‘green’. CGA consists of two acid molecules welded together, one of those two is CA. Ironically, the decaffeinated version often contains more (70–350 mg per cup of coffee) than ‘uncastrated’ java.
Due to its high concentration in coffee, CGA is one of the most abundant polyphenols in regular coffee consumers. But you can get it also from plums, blueberries, and sunflower seeds.
Like CA and caffeine, CGA readily crosses the blood-brain barrier. Once there, its antioxidant capacity comes in handy because the brain is particularly vulnerable to oxidative damage. While not many clinical studies have been conducted with humans, in rodents CGA prevents the formation of senile plaques that are typical for AD. Consequently CGA treatment improved memory, and learning, and reduced anxiety in AD-mice.
Takeaway
Here are the links to the nutrient sites that tell you which foods contain the components that I listed above.
Trigonelline (TRG)
Trigonelline is the second most abundant alkaloid in coffee after caffeine. It is also probably the most tricky one. While it seems to do all the things that I mentioned for the previous coffee components, namely improving memory and reducing or inhibiting the progression of Amyloid-beta (Aβ) in AD-mice, it is also known to inhibit an intracellular pathway that protects cells from oxidative damage.
Inside each cell resides a protein, Nrf2, which is tied to a sensor for oxidative stress. Once this sensor (Keap1) sounds the alarm, Nrf2 moves into the cell’s nucleus to do a job on the cell’s DNA.
Long story short, Nrf2 triggers several genes to express antioxidant and anti-inflammatory enzymes. That’s typically a good thing, and CA and CGA fire up that pathway, which seems to be the acids’ dominant anti-oxidant effect. Trigonelline, on the other hand, does exactly the opposite.
That doesn’t necessarily make TRG a bad actor in the health theater. Chronic Nrf2 activation can become problematic in certain disease scenarios, like cancer and diabetes. Hence, TRG putting the brakes on such runaway activation can be a good thing.
So, the jury is still out on whether TRG’s net effect on cognition is beneficial or neutral. Add to that the fact that different types of roasting affect CGA’s and CA’s potential to activate Nrf2 differently (the darker the roast the better the antioxidant effect). I’m not mentioning all this just for the fun of it. You’ll see in the next section how it relates to the two studies that we are going to take apart.
The Natural Plant Oils of Coffee: Kahweol and Cafestol
That the names of these coffee components rhyme with cholesterol is a coincidence. But one that helps you remember what they do: they slightly raise cholesterol and triglyceride levels in the blood. But that’s not all there is to them. Thanks to their anti-inflammatory, liver detox, and anti-carcinogenic actions these so-called diterpenes have a net positive effect on health.
Whether you find them in your coffee depends entirely on you, or rather on your way of brewing it. Filtering your coffee will catch those two compounds before they get into your cup. Boiled coffees and espressos, on the other hand, are a rich source of kahweol and cafestol.
Takeaway
There are three takeaways from our discussion. All three emerge from the simple fact that coffee beans are a plant food. Hence:
- Coffee engages similar physiological pathways as vegetables and fruits to promote health.
- Coffee is not the only source of its constituent nutrients.
- Coffee’s health effects cannot be reduced to any one of its components.
If you have a ‘Duh’ moment right now because these insights sound trivial, most of them were lost on the two health news which we will dissect now.
Coffee is Bad for Cognition
In August 2024, Medscape entertained its readers with the following headline:
“Too Much Coffee Linked to Accelerated Cognitive Decline.”
We have to take this seriously because Medscape’s motto is,
“…to improve patient care with comprehensive clinical information and resources essential to physicians and healthcare professionals.”
That sounds very nice, and since all of this is free of charge for anyone we wonder where the money comes from to operate this >200-employees organization. A look at the owners of Medscape puts our skepticism to rest: Medscape is owned by WebMD, which is owned by Internet Brands, which is owned by Kohlberg Kravis Roberts (KKR) (good thing the third partner’s name didn’t start with the letter K).
WebMD tells us that they,
“… generate revenue primarily through the sale of various types of advertising and sponsorship products to our clients, which include: pharmaceutical, biotechnology and medical device companies; hospitals, clinics and other healthcare services companies; health insurance providers; consumer products companies whose products or services relate to health, wellness, diet, fitness, lifestyle, safety and illness prevention; and various other businesses, organizations and governmental entities. Advertisers and sponsors use our services to reach, educate and inform target audiences of consumers, physicians and other healthcare professionals.”
How reassuring! Medscape, the self-proclaimed,
“leading online global destination for physicians and healthcare professionals worldwide, offering the latest medical news and expert perspectives; essential point-of-care drug and disease information; and relevant professional education and CME” is run by the stalwarts of scientific objectivity: The companies that wanna sell us their products and services.
Why am I digressing into the nitty-gritty of Medsacpe’s business model? Because to “follow the money” is a brilliantly simple way of arming your BS alarm.
Since it obviously wasn’t the coffee industry that paid for Medscape’s article, we have to assume that it presents the “comprehensive clinical information” that Medscape prides itself on. Let’s see whether that is the case.
The “Study”
The study behind coffee’s alleged dementia-promoting qualities doesn’t have a published paper to it. The investigators presented their findings at the Alzheimer’s Association International Conference in July 2024. Medscape didn’t even pretend to do a proper referencing to the source. It took some digging to find the “study’s” one-page summary that made it into the National Library of Medicine.
Unsurprisingly, it is a rehash of publicly available UK Biobank (UKB) data. No actual clinical work required. Simply select your cohort, compare the participants’ self-reported coffee intake at baseline with cognitive outcomes a few years (8.8 years) later, and voila, there is your coffee correlation: compared to zero cups per day, 1–3 cups reduce dementia risk, 4 and more cups increase it.
It would have been helpful to know what “cup” actually means in terms of ml or ounces. But that lack of accuracy is only a minor issue in this “study”. Far more relevant is the lack of controlling for potential confounding.
The last sentence of the Medscape article informs us that, upon being asked, the presenter stated that the study didn’t control for smoking, and (here comes the most hilarious part) that “it would be interesting to explore its impact on cognition.”
The correct follow-up question should have been: If it’s interesting, and you obviously have the data (because smoking habits are part of the UKB questionnaires), why didn’t you control for it? Maybe because your correlations weren’t significant anymore?
The smell of statistical calisthenics to support some preconceived hypothesis is too damned strong to be ignored. After all, we know from epidemiological studies in the US and Europe that regular coffee intake scales with smoking (and also with regular alcohol intake). And neither of those two is exactly a cognition enhancer.
Had the Medscape author read the “study’s” one-page summary, she would have noticed that its study authors controlled for far too few known confounders, as we will see in the next section.
Maybe that’s the reason why we still can’t find a full paper on this ‘study’ in any journal. If peer reviewers are doing their job (the one that Medscape should have done), they won’t give their nod for publication.
Coffee Boosts Cognition
On 1st January this year, The Sun put our coffee-craving minds at ease:
“Heavy coffee drinking in later life could slash dementia risk.”
Let’s do the bias check first. Being a newspaper, the business model of The Sun isn’t camouflaged. Neither are its intentions:
“The Sun is more than a newspaper. It is an instigator, an entertainer, a cultural reference point.”
Not the words of a more left-leaning critic, it’s what The Sun prides itself on.
So, let’s see whether there is some base to these “instigating” and “entertaining” coffee news.
The article reports on a study done in Switzerland. In this case, we have a full paper to analyze. The study investigated the correlation between habitual coffee intake and cognitive performance in participants of the Swiss Atrial Fibrillation (Swiss-AF) study. AF patients have a high risk of stroke and hence future cognitive impairments.
The primary outcome measures were several types of cognitive performance tests that are used to detect mild cognitive impairments (I have compiled those tests for you to download and apply to yourself or your loved ones. The link is below).
A Brief on The Tests
The Montreal Cognitive Assessment (MoCA)
MoCA is a brief cognitive screening tool with high sensitivity and specificity for detecting mild cognitive impairement (MCI) in otherwise normally perfoming persons.
The Trail Making Test (TMT)
The TMT measures how quickly participants connect numbered circles (Part A) or numbers and letters (Part B) in order. The results are indicative of participants’ processing speed, flexibility, and attention.
The Semantic Fluency Test (SFT)
The SFT is an assessment semantic of access to word memory. It consists of verbally naming as many words from a single category (e.g animals) as possible in sixty or ninety seconds.
The Digit Symbol Substitution Test (DSST)
The DSST measures how quickly participants can copy simple graphic symbols to the digits 1–9 based on a key within two minutes. The total score reflects cognitive skills like attention, speed, and memory.
The Cognitive Construct Score (CoCo)
Is a score that condenses the most predictive factors of the above-mentioned scores into one “derived factor” score specifically for AF patients.
I have prepared the MoCa tests, TMT, and DSST (except the CoCo score) for you to download.
As you can see from the image above there was a dose-response relationship between coffee consumption (expressed as cups per day) and scoring on all tests. The difference between the java junkies (>5 cups/day) and the coffee abstainers (<1 cup/day) was significant only on the DSST score, not on the other three scores. That’s probably because there were only 40 java junkies in the overall group of 2413 participants.
As is the case for the UK Biobank study, we don’t know what “cup” means in terms of volume.
Contrary to the UK Biobank study, the Swiss researchers controlled for smoking status. They also controlled for all the variables in the UK study plus physical activity, the presence of depression, hypertension, medication, and diabetes. Is that why the results of this study contradict those of the UKB study?
There are two reasons why neither study should impress you or make you believe that there is a causal relationship between coffee and cognition. The first reason has to do with plain statistics. The second reason is my personal, and admittedly contrarian interpretation of coffee studies, no matter whether the examined correlation is with cognitive performance, heart health, or any other health metric.
First, the statistics.
The Flawed Statistics
Confounding in General
I mentioned the problem of confounding.
To be a confounder of a relationship between a causal variable (e.g. coffee intake) and an outcome (e.g. cognitive impairment) the confounding variable,
- Must be causally associated with the outcome (in this case, cognitive performance).
- Must be associated (causally or not) with the exposure (in this case, coffee intake).
- Must not be on the causal pathway between the exposure and the outcome.
Smoking and alcohol intake are obvious candidates for confounding as suggested by physiology and by epidemiological studies. Here is why controlling for these confounders is so important.
Let’s say, high coffee intake has no impact on dementia but correlates positively with smoking. If smoking promotes dementia then lack of controlling for smoking will incorrectly suggest that coffee causes cognitive impairment.
The same false conclusion could even be made if coffee had a protective effect, but the deleterious effects of smoking outweighed coffee’s benefits. In that case, we will still see a net negative effect.
In more practical terms. If you, as a non-smoker, get scared away from drinking coffee, simply because your favourite health evangelists didn’t untangle the confounding effect of smoking from the benefits of coffee, you’d be screwed (well, maybe not screwed, but at a disadvantage).
Takeaway
If you read a post or article, that lists an impressive amount of variables that had been controlled for (often referred to as ‘covariates’), don’t be too impressed. In the grand scheme of nutrition science, the most contentious of the health sciences, it is a known but rarely acknowledged fact that the researchers’ choice of controlled-for variables can swing the outcomes one direction or another. It is called vibration of effects (VoE).
So, how do we know which confounders should be controlled for? Simple, we don’t.
But we can get a fairly good idea.
The Confounders of Coffee Studies
In the coffee-cognition scenario, we could look at the results of genome wide association studies (GWAS). GWAS search for genes (or more precisely for gene variants, so-called single nucleotide polymorphisms — SNPs) that correlate with coffee drinking. Those variants that are also known to be causally related to other health behaviors or conditions (traits) of their carriers are automatically good candidates for confounders.
Thanks to large genomic databases, such as the UK Biobank and the ’23andme’ database in the US, we have such data available.
A very recent study methodically searched these two for coffee-drinking correlates. First here is the most surprising finding. In ’23andme’ 75 traits correlated significantly with coffee consumption. In the UKB cohort, it was 74. But these 70+ traits were not all the same. Only 30% of traits that were significant in one cohort were significant in the other cohort, too.
And 40% of those that were significant in one cohort had the opposite direction of correlation in the other cohort. That’s surprising as both cohorts are largely made up of people of European descent.
In any case, genetic correlations between coffee and alcohol consumption and cigarette smoking were unidirectional and strongly significant in both cohorts.
To cut to the chase. Since June last year, researchers can refer to these GWAS results to identify potential confounders for the coffee-whatever associations they want to examine. Neither the UKB nor the Swiss study did that.
To be fair, not every potential confounder in a GWAS will necessarily have a significant effect. But one confounder should definitely be considered. It has nothing to do with genes but everything with Starbucks & Co.
If your idea of coffee is a Mocha Cookie Crumble Frappuccino with a calorie content equal to a Big Mac (560 kcal), you should expect some different health outcomes than I do from my unadulterated espressos (no milk, no sugar).
Did any of the studies that we discussed account for the type of coffee consumed? No. At best, they look at caffeinated vs. decaf versions, but not at the whacky world of coffees out there.
Takeaway
Fastfoodnutrition.org — your link to the nutrient contents of your favorite fast-food chain.
So, unaccounted confounders make study results unreliable. But there is one more reason why they specifically may not apply to you. Remember what I said about the different confounders in the UKB vs. the ’23andme’ populations? If you are Australian, Indian, or Chinese, all the confounder assumptions go up in smoke. You don’t match those populations.
What if you are Swiss? The results of the Swiss study only apply to you if you are living there AND have AF. Because that’s the population that was investigated. You can’t simply extrapolate from a study population if you don’t meet the study’s inclusion and exclusion criteria.
Takeaway
Always pay attention to the studied population and their inclusion and exclusion criteria.
Now to my second point why those studies that made it into Medscape and The Sun shouldn’t impress you.
A Radically Different Point of View
Coffee is no medicine, unless our lifestyle makes it so.
Think about it. One point of my admittedly lengthy introduction into the cognition-relevant components of coffee was to show that coffee is a regular plant food whose nutrients are found in other plant foods, too. Due to their exceptionally high concentration in coffee, the latter rescues us to some extent from the nutrient starvation of our more or less shitty diets.
“Coffee employs similar pathways of beneficial physiological effects as were recently identified for vegetables or fruits. At the level of populations in the developed world, coffee provides more dietary phenolic phytochemicals than vegetables and fruits.”
That’s why it’s not that coffee reduces your dementia risk, it’s your less-than-sterling lifestyle habits that increase it in the first place. Coffee simply attenuates their negative effects.
You can use the three tests that I have prepared for your download to check your cognitive ability scores. Or the scores of your loved ones.
What I Do
I love unadulterated espresso, no milk no sugar. Each of the latter two spoils the unique, pleasantly bitter taste of coffee. That’s a totally subjective view of course.
I have my 4–6 cups daily until no later than 1 pm. I wouldn’t mind drinking espresso in the afternoon too, but this has simply not become a routine. Which, by the way, brings me to the recent news about timing your coffee intake. You’ll read that limiting intake to the morning provides health benefits, which disappear if you extend your intake to afternoons. When you read those news, just keep in mind what I have laid out here.
Over to You
You read articles like the ones in Medscape and The Sun in the hope that their messages and conclusions apply to you. As I have tried to argue, they mostly won’t. That’s why I led into this post with the reference to Sturgeon’s Law. Though it was meant to be more tongue-in-cheek than serious, I have come to recognize it as being quite spot-on in many areas, particularly in the media and internet universe.
But you may see it all differently. What I would like to hear from you is why or why not you disagree with me.
