A drugstore multivitamin, three trials, and one stubborn little number

COSMOS tested a daily multivitamin against placebo in more than 21,000 American seniors and found a small, repeatable slowing of cognitive aging. It is one of the only supplement results in dementia prevention that has held up, and it is smaller than the headlines suggest.

A daily multivitamin bought off a drugstore shelf slowed cognitive aging in a national placebo-controlled trial, and then did it again, and then again. The effect is small. In a field where almost every supplement has failed outright, small and repeatable is unusual enough to be news.

The trial is COSMOS, the Cocoa Supplement and Multivitamin Outcomes Study, run out of Brigham and Women's Hospital. It randomized 21,442 adults across the United States, with 35,669 people starting a placebo run-in first, a two-by-two design testing cocoa extract containing 500 mg a day of flavanols and a commercial multivitamin-mineral against matching placebos. The parent trial was built to ask about heart attacks and cancer. Cognition rode along as ancillary studies.

The first of those was COSMOS-Mind, led by Laura Baker and Mark Espeland at Wake Forest, published in Alzheimer's & Dementia on 14 September 2022. It followed 2,262 participants aged 65 and over who did not have dementia, testing them by telephone at baseline and once a year for three years. Cocoa extract did nothing for cognition. The multivitamin group ended three years ahead on the global cognition composite by 0.07 of a standard deviation.

Hold that number for a second, because it is the whole argument. Scores on these tests drift down with age at a rate the investigators measured in their own participants, whose average age was 73. Against that yardstick, the authors estimated the gap amounted to roughly 1.8 years of age-related decline forestalled.

That is a statistical statement about group averages, and a group average is not a morning at a kitchen table. Nobody in the trial reported finding a name faster.

What makes COSMOS interesting is what happened next. COSMOS-Web, led by Lok-Kin Yeung and Adam Brickman at Columbia, published in the American Journal of Clinical Nutrition on 24 May 2023, used web-based testing in more than 3,500 participants aged 60 and over. Its primary outcome, immediate recall on the ModRey word list, favored the multivitamin at one year and across three years. The secondary outcomes, object recognition and executive function, showed nothing. The authors put the memory difference at about 3.1 years of age-related change.

Then came the in-person check, which matters most, because telephone and web testing are blunt instruments. COSMOS-Clinic, reported by Chirag Vyas and colleagues in the same journal on 18 January 2024, brought 573 participants into a room with a neuropsychologist at baseline and at two years. Episodic memory favored the multivitamin by 0.12 standard deviation units. Global cognition moved by 0.06 units and its confidence interval crossed zero, so that one fell short of significance.

Pooling the three substudies, more than 5,000 people with no one counted twice, the meta-analysis found 0.07 standard deviation units on global cognition (95% CI 0.03 to 0.11, P = 0.0009) and 0.06 on episodic memory (95% CI 0.03 to 0.10, P = 0.0007). The authors describe the global cognition effect as equivalent to reducing cognitive aging by about two years.

Everyone improves a little the second time they take the same memory test, which is exactly why randomization and a placebo group exist. The cocoa arm is the quiet proof that this machinery was working: same participants, same tests, same mail-order bottles, and cocoa produced nothing at all.

So what should the 55-year-old daughter, the one who already keeps a list of her mother's medications in her phone, do with this?

First, note who was studied. These were community-dwelling volunteers aged 60 and over without dementia at enrollment. Nothing here speaks to someone already diagnosed. If your mother is three years into a diagnosis, COSMOS has no message for her.

Second, note what was measured. Test scores over two to three years. Not dementia diagnoses, not driving, not whether anyone stayed in their own house longer.

Third, note the money. The pills were a commercial multivitamin-mineral product, and the work was supported in part by investigator-initiated grants from Mars Edge, a segment of Mars Inc, which has a cocoa business. The cocoa arm failed anyway, which is the kind of detail that makes a funded trial easier to read.

Vyas, speaking to Medscape when the third paper landed, declined to turn the result into a recommendation and said only that the evidence supports the promise of multivitamins for preventing cognitive decline. That restraint is appropriate, and it belongs in this story rather than in a footnote.

For twenty years the honest summary of supplements and memory was a shrug. COSMOS makes the summary slightly longer: one cheap, widely available formulation produced a small benefit on cognitive testing, three times, in the same large randomized trial, and the chocolate in the next bottle did not. Whether that small benefit is worth anything to a particular family is a conversation for a particular clinician, with that family's medications and kidneys and diet on the table.