Caffeine: Part Three: The Long View

Abstract

Parts One and Two covered what a dose of caffeine does over hours and what daily intake does over days. This part covers the question people actually ask: does drinking coffee for forty years protect the brain, or damage it?

A systematic review in the Journal of Alzheimer's Disease assembled 61 studies published between 1990 and 2020 and sorted them by whether their outcome was positive, negative or null.

The answer it produces is genuinely mixed, and the shape of the mixture is more informative than any single number.

1.The Method, and Why It Is Unusual

Jessie Chen, Philip Scheltens, Colin Groot and Rik Ossenkoppele did not pool effect sizes. They assessed overall study outcomes — positive, negative or no effect — and then examined whether study characteristics explained the differences.

That choice reflects the material. These are epidemiological studies with different caffeine sources, different quantities, different follow-up periods and different confounders. Pooling them into a single number would produce a figure with no defensible interpretation.

The authors open by stating the position plainly: epidemiological studies have provided inconclusive evidence for a protective effect of caffeine consumption on the risk of dementia and cognitive decline.

2.The Numbers

Of 57 studies examining the association between caffeine and risk of dementia or cognitive decline, 16 — 28 per cent, covering 40,707 of 153,070 participants — reported positive outcomes outright.

A further 30 studies, 53 per cent, covering 71,219 participants, showed positive results that were dependent on study characteristics.

That second category is the largest and the most interesting. A majority of this literature found a benefit under some conditions and not others — which is either a real effect with real boundaries, or an artefact of testing many subgroups until one is significant. Both readings are available and the review does not force a choice.

Arithmetic on those figures leaves roughly one study in five reporting no effect or a negative one.

3.Where the Conditions Point

Three characteristics recurred among the positive findings, and they are consistent enough to be worth naming.

Quantity: effects were more often positive at moderate intake, defined here as 100 to 400 mg a day — roughly one to four cups of coffee.

Source: effects were more often positive when caffeine came from coffee or green tea rather than from other sources.

Sex: effects were more often positive in women.

The source finding deserves a caution the review's readers routinely skip. Coffee and green tea are not caffeine — they are complex mixtures containing polyphenols, chlorogenic acids and, in green tea, the compounds covered in our L-theanine series. If the benefit tracks the beverage rather than the molecule, the molecule may not be the active part.

A dose-dependent effect that appears only within a window, and only from particular sources, is not a straightforward story about caffeine.

4.The Already-Impaired

Four studies examined caffeine and cognitive function in people who already had mild cognitive impairment or dementia. Three of the four, covering 272 of 289 participants, reported positive outcomes.

Three studies out of four is not a basis for anything, and the review does not present it as one. We include it because the direction is worth recording, and because it is a small enough number that a reader can hold the whole evidence base in mind at once.

5.What Observational Data Cannot Do

Everything in this part is observational, and the standard limitation applies with unusual force to coffee.

People who drink moderate amounts of coffee over decades differ from those who do not, in ways that predict cognitive outcomes independently: they are more likely to be employed, socially active, and physically well enough to tolerate a stimulant. And reverse causation is specifically plausible here — early cognitive decline changes habits, and someone in the prodromal phase of dementia may reduce their coffee intake years before diagnosis, making low consumption look like a cause of an outcome it followed.

The three possibilities we set out in our gut–brain series apply here identically: causative, consequential, or bystander. This literature cannot distinguish them, and the review's refusal to pool the data is the appropriate response to that.

6.Where Three Parts Leave It

Acutely, in pooled randomised crossover trials around exercise, caffeine improves attention, accuracy and speed — and only those.

Taken daily at 450 mg by habitual consumers, it did not disturb sleep architecture or subjective sleep quality, while leaving a persistent reduction in spindle-band EEG power that survived a day of abstinence.

Across decades, the observational evidence is inconclusive, with a majority of studies reporting benefits that depend on how the study was conducted, concentrated at moderate doses and in particular beverages.

That is a substantial and genuinely useful body of knowledge about a compound most people take without thinking. It is also considerably less than the certainty with which caffeine is discussed in either direction.

Editorial Comment

Boost contains 100 mg of caffeine, from anhydrous caffeine and guarana — inside the 100 to 400 mg range in which the observational findings above cluster, though we would not present that as support for anything. A dose falling within a window where epidemiological studies more often report benefits is not evidence that our product produces those benefits, and treating it as such would be exactly the reasoning this series argues against.

As in Part One, we quote no authorised claim wording for caffeine because we have not re-verified the current register entry, and its conditions attach to dose and carrier in ways that require checking rather than recalling.

One practical note that costs nothing. If caffeine has stopped working for you, the most likely explanation in this literature is not tolerance to be overcome with a larger dose. It is that the comparison has changed: a habitual user's baseline is a caffeinated baseline, and the felt effect is measured from there.