Multivitamins: The COSMOS Memory Trials and What Else Holds Up

Multivitamins are the most common dietary supplement in the United States; the US Preventive Services Task Force cites a survey in which 52 percent of adults had used at least one supplement in the previous month and 31 percent a multivitamin-mineral. For most of the time they have been trialled, the honest summary of the evidence was “little either way”. Between 2022 and 2024 that changed a little. Three sub-studies of one large trial, COSMOS, found that a daily multivitamin nudged memory scores upward in older adults. The effect is statistically real, practically small, and sits beside trial data on heart disease and lifespan that did not move, and cancer data that are split.

The two trials that matter

Almost everything solid comes from two randomized, placebo-controlled trials run by overlapping teams at Brigham and Women’s Hospital in Boston.

Physicians’ Health Study II (PHS II) randomized 14,641 male US doctors aged 50 or over to a daily multivitamin or placebo starting in 1997 and followed them for a median of 11.2 years. Its investigators describe it as the first long-term trial of a multivitamin, and its participants were all men and all physicians.

COSMOS (the COcoa Supplement and Multivitamin Outcomes Study) randomized 21,442 US adults, women 65 and over and men 60 and over, in a two-by-two design testing a cocoa-flavanol extract and a daily multivitamin from 2015 to 2020, median follow-up 3.6 years. It excluded anyone with a prior heart attack or stroke, or a cancer diagnosis in the previous two years. The multivitamin was Centrum Silver. Study pills and packaging were donated by Mars Edge and Pfizer Consumer Healthcare, Mars Edge also funded infrastructure, and the COSMOS-Mind paper states that neither the NIH nor either company had a role in design, conduct, analysis or publication.

Both trials used the same daily multivitamin-mineral, Centrum Silver, against placebo, which is the only “dose” the evidence speaks to.

What COSMOS found on memory

COSMOS ran three cognitive sub-studies, each with its own testing method and partly overlapping groups of volunteers.

• COSMOS-Mind (Baker 2023) tested 2,262 people, mean age 73, by telephone once a year for three years. The multivitamin group’s global cognition score finished 0.07 standard deviation units above placebo (95% CI 0.02 to 0.12, P = 0.007). The gap was widest in those with prior cardiovascular disease (0.14 versus 0.06 without). The cocoa extract had no significant effect (0.03, 95% CI −0.02 to 0.08).
• COSMOS-Web (Yeung 2023) tested 3,562 people online. The primary endpoint, immediate recall on the ModRey memory test after one year, favoured the multivitamin (P = 0.025), as did the three-year average (P = 0.011). The authors translate the gap as the equivalent of about 3.1 years of age-related memory change. The secondary measures, novel object recognition and executive function, did not differ significantly from placebo.
• COSMOS-Clinic (Vyas 2024) gave 573 people in-person neuropsychological testing over two years. Episodic memory finished 0.12 units ahead of placebo (95% CI 0.002 to 0.23), a result that only just clears significance; global cognition (0.06, CI −0.003 to 0.13) and executive function (0.04, CI −0.04 to 0.11) did not.

The same paper pooled the three studies’ non-overlapping participants, 5,203 in all. Across them the multivitamin improved global cognition by 0.07 units (95% CI 0.03 to 0.11, P = 0.0009) and episodic memory by 0.06 units (95% CI 0.03 to 0.10, P = 0.0007); the global-cognition effect is the one the authors equate to roughly two years of cognitive ageing. That is the whole memory case: three randomized comparisons pointing the same way and a convincing pooled P value, with memory moving in all three and executive function only in COSMOS-Mind.

How big 0.07 standard deviations is

Small. An effect of 0.07 means the treated group averaged seven hundredths of one standard deviation above placebo, a shift no individual could notice. The “years of ageing” conversions compare that gap with how scores fall with age in cross-section, and sound large only because the average change per year of age is itself tiny.

Three other caveats matter. COSMOS-Mind was 89 percent non-Hispanic white, and the parent trial excluded recent cancer and prior heart attack or stroke, so the result may not transfer to other groups. Memory was the primary endpoint in COSMOS-Web; in COSMOS-Mind the multivitamin was the pre-specified secondary question and cocoa extract the primary one. And all three measured test scores, not dementia diagnoses. The checklist in how to read a supplement study scores this evidence well on size and funding disclosure and poorly on effect size.

PHS II is the other reason for caution. Grodstein 2013 followed 5,947 of the physicians aged 65 and over through up to four telephone cognitive assessments over 12 years and found no difference at any of them: the global score differed by −0.01 units (95% CI −0.04 to 0.02) and verbal memory by −0.005. Why COSMOS found a small effect where PHS II found none is not settled. Grodstein’s paper suggests the doses may have been too low or the physicians too well nourished to benefit; the tests differed too, and the effect, if real, is small enough for a long trial to miss.

Cancer, heart disease and lifespan

Here the record is mostly flat.

Cancer. In PHS II, men on the multivitamin had an 8 percent lower rate of total cancer diagnoses over 11 years (17.0 versus 18.3 per 1,000 person-years; hazard ratio 0.92, 95% CI 0.86 to 0.998, P = 0.04), with no effect on prostate, colorectal or other individual cancers and no significant difference in cancer deaths. COSMOS could not repeat it: 518 invasive cancers on the multivitamin versus 535 on placebo (HR 0.97, CI 0.86 to 1.09, P = 0.57). COSMOS did report fewer lung cancers (HR 0.62, CI 0.42 to 0.92), one unreplicated secondary outcome among many.

Heart disease. PHS II found no effect on major cardiovascular events (11.0 versus 10.8 per 1,000 person-years, HR 1.01), heart attack, stroke or cardiovascular death. COSMOS agreed: 429 versus 437 cardiovascular events (HR 0.98, CI 0.86 to 1.12). A secondary analysis of 13,316 PHS II men with baseline diet data found no consistent sign that foods, individual nutrients, diet-quality scores or supplement use at entry changed the null heart result (Rautiainen 2017).

Mortality. Neither trial saw a significant difference in all-cause deaths: HR 0.94 (CI 0.88 to 1.02) in PHS II and 0.93 (CI 0.81 to 1.08) in COSMOS. Loftfield 2024 followed 390,124 US adults free of chronic disease at entry for up to 27 years, with 164,762 deaths, and found multivitamin use went with, if anything, a marginally higher death rate (HR 1.04, CI 1.02 to 1.07 in the first half of follow-up; 1.04, CI 0.99 to 1.08 in the second). A gap that small can reflect who chooses to take a multivitamin, but it leaves no room for a lifespan benefit.

The US Preventive Services Task Force, weighing this evidence in 2022, judged it insufficient to recommend for or against multivitamins for preventing cardiovascular disease or cancer, and recommended against beta-carotene and vitamin E supplements for that purpose.

What the pill does, and who it applies to

A multivitamin does change blood chemistry. In a COSMOS biomarker sub-study of 399 participants, two years on the multivitamin raised total serum carotenoids by 15.5 percent, lutein by 17 percent, beta-carotene by 47 percent and alpha-tocopherol (vitamin E) by 29.5 percent relative to placebo. Lycopene, zeaxanthin, alpha-carotene and beta-cryptoxanthin did not differ significantly from placebo. The study measured blood levels only, not any health outcome. Vitamin D at the doses deficiency trials used is a separate question, covered in vitamin D supplementation, as are the higher amounts of vitamin K2 and D3 studied together.

Every positive cognitive result comes from adults aged 60 and over, and the largest point estimate was a COSMOS-Mind subgroup with prior cardiovascular disease, whose own confidence interval (−0.02 to 0.31) crossed zero. None of these trials enrolled adults in their 30s or 40s, so the memory finding says nothing about them. The “marginal diet” story, that the pill helps people whose food is short of micronutrients, is the explanation Grodstein’s paper raises for its null result in well-nourished physicians; the one PHS II analysis that tested it for heart outcomes found no consistent pattern, and none of the three COSMOS cognitive abstracts reports a diet subgroup.

Doses are what the trials used: a daily Centrum Silver multivitamin-mineral, for two to three years in the COSMOS sub-studies and a median of 11.2 years in PHS II. COSMOS-Web’s authors describe the approach as “safe and accessible”; even so, anyone who is pregnant, takes prescription medication or has a diagnosed condition should ask a clinician first. The Supplement Stack Bible covers the single-nutrient trials in the same sceptical frame.

Bottom line

Three COSMOS sub-studies, pooled to 5,203 older adults, show that a daily multivitamin raised memory and global cognition scores by about 0.06 to 0.07 standard deviations over two to three years, which the investigators translate as roughly two to three years of age-related change. It is a genuine randomized finding and a small one, tested only in adults aged 60 and over, with its largest point estimate in a subgroup whose own confidence interval crossed zero. On cancer the two big trials disagree and the newer, larger one is null; on heart disease both are null; on mortality nothing suggests a benefit. A multivitamin is a low-dose supplement with a small measurable effect on memory tests in older adults and no consistent effect on cancer, heart disease or lifespan.