THE MULTIVITAMIN QUESTION: WHAT THE COSMOS TRIALS MEASURED

A 21,442-person randomized trial tested two supplements at once in adults over 60 — cocoa flavanols and a daily multivitamin. The one everybody expected to win didn't. The one nobody was watching produced the more interesting result.

The trial

COSMOS — the COcoa Supplement and Multivitamin Outcomes Study — enrolled 21,442 US adults: 12,666 women aged 65 and over, 8,776 men aged 60 and over, all free of major cardiovascular disease and recent cancer. Randomized, double-blind, placebo-controlled, and built as a 2×2 factorial design, meaning every participant was independently assigned to cocoa extract or placebo and to a multivitamin or placebo. The cocoa arm delivered 500 mg flavanols per day, including 80 mg (−)-epicatechin. The multivitamin was Centrum Silver. The intervention ran from June 2015 through December 2020 (Sesso et al., American Journal of Clinical Nutrition 2022; PMID 35294962).

That scale matters because most supplement research at this level is observational — people who already take a supplement compared with people who don't, with all the lifestyle differences that come attached. COSMOS randomized. It also ran long enough to count real clinical events rather than blood markers.

What the cocoa arm found

Over a median 3.6 years of follow-up, 410 participants taking cocoa extract and 456 taking placebo had a confirmed total cardiovascular event. Hazard ratio 0.90, 95% CI 0.78 to 1.02, P = 0.11.

That is the primary endpoint, and it did not reach statistical significance. The authors' own conclusion says so plainly: cocoa extract supplementation did not significantly reduce total cardiovascular events among older adults.

The secondary endpoints are where the headlines came from. Cardiovascular death: HR 0.73, 95% CI 0.54 to 0.98 — a 27% reduction, and the confidence interval clears 1.0. Myocardial infarction: 0.87 (0.66–1.16). Stroke: 0.91 (0.70–1.17). Coronary revascularization: 0.95 (0.77–1.17). All-cause mortality: 0.89 (0.77–1.03). A per-protocol analysis that censored follow-up at non-adherence supported a lower risk of total cardiovascular events (HR 0.85, 0.72–0.99). There were no safety concerns.

So: one significant secondary endpoint, a supportive per-protocol analysis, and a primary endpoint that came up short. The authors wrote that additional research is warranted to clarify whether cocoa extract may reduce clinical cardiovascular events. That is a fair summary of where it sits.

Then the cognition studies

Three separate cognitive substudies ran inside COSMOS, and they are the reason this trial is worth a second look.

COSMOS-Mind (Baker et al., Alzheimer's & Dementia 2023; PMID 36102337) enrolled 2,262 participants, mean age 73, and assessed cognition by telephone annually for three years. The primary endpoint was change in a global cognition composite with cocoa extract. Result: no effect. Mean z-score 0.03, 95% CI −0.02 to 0.08, P = 0.28.

The pre-specified secondary endpoint was the multivitamin. That one moved: mean z 0.07, 95% CI 0.02 to 0.12, P = 0.007, with benefits also observed for memory and executive function. The effect was most pronounced in participants with a history of cardiovascular disease, though the authors flag that interaction as nominal.

COSMOS-Web (Yeung et al., AJCN 2023; PMID 37244291) tested 3,562 older adults with an internet-based battery over three years. Its pre-specified primary outcome was change in episodic memory at one year, measured as immediate recall on the ModRey test. The multivitamin group performed significantly better at one year (P = 0.025) and across three years on average (P = 0.011). Secondary outcomes — novel object recognition, executive function — showed no significant effect.

COSMOS-Clinic (Vyas et al., AJCN 2024; PMID 38244989) was the in-person version, with detailed neuropsychological testing in 573 participants. Global cognition over two years: mean difference 0.06 SD units, 95% CI −0.003 to 0.13 — a modest benefit that did not clear significance. Episodic memory did: 0.12 SD units, 95% CI 0.002 to 0.23. Executive function and attention did not.

That paper also ran a meta-analysis pooling the non-overlapping participants across all three substudies — 573 plus 2,158 plus 2,472. Pooled global cognition: 0.07 SD units, 95% CI 0.03 to 0.11, P = 0.0009. Pooled episodic memory: 0.06 SD units, 95% CI 0.03 to 0.10, P = 0.0007.

How big is that, honestly

Both research groups translated the effect into years, which is the most useful way to read a standard-deviation number. The COSMOS-Web authors estimated the multivitamin improved memory performance above placebo by the equivalent of 3.1 years of age-related memory change, based on a cross-sectional analysis of age and ModRey performance. The pooled meta-analysis put the global cognition effect at the equivalent of reducing cognitive aging by 2 years.

Those are estimates derived from the data, not measured outcomes, and the effect sizes are small in absolute terms — 0.06 to 0.07 standard deviation units. What makes them worth reporting is that they came from a randomized comparison in tens of thousands of people, and they replicated across three separate substudies using three different testing methods: telephone, internet, and in-person.

What the trial did not show

Cocoa extract did not benefit cognition in the study designed to test it, and it did not hit its primary cardiovascular endpoint.

None of the cognitive substudies measured dementia incidence. They measured test scores over two to three years. A change in a cognition composite is not the same outcome as whether someone develops Alzheimer's disease, and the papers do not claim it is.

The COSMOS-Mind authors also noted the cohort was 89% non-Hispanic White and wrote that additional work is needed to confirm the findings in a more diverse cohort and to identify the mechanisms behind the multivitamin effect. Nobody has established why a multivitamin would do this.

And the funding is disclosed in the papers themselves, which is the standard I'd want applied to any supplement research I read: "COSMOS is supported by an investigator-initiated grant from Mars Edge, a segment of Mars dedicated to nutrition research and products, which included infrastructure support and the donation of study pills and packaging. Pfizer Consumer Healthcare (now part of GSK Consumer Healthcare) provided support through the partial provision of study pills and packaging." NIH grants supported the trial as well. The investigators state that neither NIH nor Mars nor Pfizer contributed to the design, conduct, analysis, or decision to publish. Mars makes cocoa products; Pfizer made Centrum. Both supplements tested were donated by companies that sell them.

The Protocol

What the trials actually tested, stated as what was tested:

• A standard commercial daily multivitamin-mineral (Centrum Silver), taken daily for two to three years in adults aged 60 and over, produced a small but repeated benefit on global cognition and episodic memory across three substudies.
• Cocoa extract at 500 mg flavanols per day produced no cognitive benefit and missed its primary cardiovascular endpoint, with one significant secondary endpoint (cardiovascular death) and no safety concerns.
• Nothing in COSMOS tested a multivitamin against eating well. It tested a multivitamin against a placebo pill.

A multivitamin is roughly the cost of a coffee per month, the trial reported no safety concerns across 21,442 people over 3.6 years, and the cognitive signal replicated three times. That combination is unusual in supplement research. The effect size is small, the mechanism is unknown, and no trial has shown it changes anyone's odds of dementia. Both of those things are true at once, and I'd rather report them together than pick the half that makes a better headline.

Multivitamins can interact with prescription medication and with certain conditions — the vitamin K content matters if you take warfarin, and iron and vitamin A content matter in other situations. If you take prescription medication, that conversation belongs with your physician or pharmacist, who can see your actual list.