Vitamin D and All-Cause Mortality: The U-Shaped Curve the Megadose Crowd Ignores.

The shape of the curve — what 247,574 Danes showed

The single most important paper to read on vitamin D and mortality is the Copenhagen General Practitioners (CopD) study, published by Durup and colleagues in The Journal of Clinical Endocrinology & Metabolism in 2012, and extended in a 2015 follow-up looking specifically at cardiovascular endpoints [1][2]. The first paper analysed serum 25-hydroxyvitamin D — abbreviated 25(OH)D, the standard marker of vitamin D status — in 247,574 individuals drawn from the general Danish population, linked to the national death registry.

The finding was not "more is better." All-cause mortality was lowest at a serum 25(OH)D concentration of approximately 70 nmol/L (28 ng/mL). Below that level, mortality rose progressively as 25(OH)D fell — consistent with everything the deficiency literature had been claiming for a decade. But above approximately 100–125 nmol/L (40–50 ng/mL), mortality rose again. The hazard ratio at very low 25(OH)D (around 10 nmol/L) was approximately 2.13, and the hazard ratio at very high 25(OH)D (around 140 nmol/L) was approximately 1.42, both relative to the 50–60 nmol/L reference. The shape was a clear, statistically robust U.

The 2015 follow-up, using the same cohort, showed the same U-curve for major cardiovascular events and acute myocardial infarction specifically. The pattern was not driven by a single endpoint. It showed up in cardiovascular mortality, in non-cardiovascular mortality, and in the composite of both.

This is the dataset most "vitamin D = lower mortality" infographics quietly omit, because it makes the linear narrative untenable. The Danish data do not say vitamin D doesn't matter — they say the relationship is non-monotonic, and that the upper limb of the curve is real.

The Schöttker BMJ meta-analysis: 26 cohorts, same shape

If a single Danish cohort were the only evidence, you could argue ancestry, latitude, or measurement quirk. The reason the U-curve has held up is that it has been reproduced — most rigorously by Schöttker and colleagues in a 2014 BMJ meta-analysis of 26 prospective cohort studies, pooling 26,018 men and women aged 50–79 across Europe and the United States [3].

The Schöttker analysis pre-specified quintiles of 25(OH)D and tracked all-cause and cause-specific mortality. Compared with the highest quintile, participants in the lowest quintile had a 57% higher risk of all-cause mortality (hazard ratio 1.57, 95% CI 1.36–1.81). But within the upper half of the distribution, the protective signal flattened and, in cardiovascular-disease subgroup analyses, modestly reversed. The authors concluded that "25-hydroxyvitamin D deficiency is associated with increased all-cause and cause-specific mortality" — but the per-quintile data, examined carefully, show the same non-linear shape as the Danish cohort.

Sempos and colleagues, working at the U.S. National Institutes of Health, performed a parallel analysis on the NHANES III cohort and reported essentially the same finding: a J-shaped curve, with the nadir of mortality risk between approximately 75 and 100 nmol/L and a modest rise above 125 nmol/L [4]. Three independent cohorts on two continents, all producing the same curve, is not a statistical fluke.

Vitamin D is one of the few nutrients where the population evidence is strong, the trial evidence is null, and the dose–response refuses to be a straight line.

VITAL: when a properly powered RCT lands null

The VITamin D and OmegA-3 TriaL (VITAL), led by Manson and colleagues at Harvard, is the largest randomized trial of vitamin D supplementation ever conducted. It enrolled 25,871 U.S. adults — men aged 50 or older and women aged 55 or older — and randomized them in a 2×2 factorial design to 2,000 IU/day of vitamin D3, 1 g/day of marine omega-3, both, or matching placebos. Median follow-up was 5.3 years. The primary results published in The New England Journal of Medicine in 2019 [5].

The result that landed with a thud through the supplement narrative: vitamin D supplementation did not reduce the primary endpoint of invasive cancer of any type (hazard ratio 0.96, 95% CI 0.88–1.06), and did not reduce the co-primary endpoint of major cardiovascular events (hazard ratio 0.97, 95% CI 0.85–1.12). The pre-specified secondary endpoint of all-cause mortality was also null.

Subsequent ancillary VITAL analyses have looked for benefit in fracture, autoimmune disease, depression, cognitive function, and diabetes incidence. The autoimmune-disease arm was a notable positive signal, with vitamin D supplementation associated with a 22% reduction in new autoimmune disease diagnoses over 5 years [6]. The diabetes arm in adults with prediabetes (D2d trial, related literature) was null on the primary endpoint but suggestive in subgroup analyses.

The single most important nuance about VITAL is baseline status. The mean serum 25(OH)D at randomization was approximately 77 nmol/L — i.e., already in the protective range identified by the cohort curves. VITAL did not test "correct deficiency"; it tested "add vitamin D to a largely replete population." That is a different question. The null result tells us that once you are in the protective range, adding more does not extend life. It does not tell us that correcting frank deficiency is useless.

DO-HEALTH and Bischoff-Ferrari's older adults

DO-HEALTH was a multicenter European RCT led by Bischoff-Ferrari and colleagues, published in JAMA in 2020. It enrolled 2,157 community-dwelling adults aged 70 or older, across five European countries, in a 2×2×2 factorial design testing 2,000 IU/day of vitamin D3, 1 g/day of marine omega-3, and a home exercise program [7].

After 3 years, none of the three interventions — individually or in combination — improved the primary endpoints: change in systolic and diastolic blood pressure, Short Physical Performance Battery score, Montreal Cognitive Assessment score, incidence of non-vertebral fractures, or incidence of any infection. The vitamin D arm specifically did not reduce fractures, did not improve physical function, and did not slow cognitive decline.

DO-HEALTH is significant because it tested the population most likely to benefit — older European adults, with higher baseline rates of suboptimal vitamin D status, in countries with limited winter sun. It still landed null on every primary endpoint. Combined with VITAL, we now have two large, properly powered RCTs in the populations most commonly told to supplement, both showing that 2,000 IU/day of vitamin D3 does not meaningfully extend healthy lifespan in non-deficient adults.

Mendelian randomization and the causality problem

Mendelian randomization (MR) is the cleanest tool epidemiology has for testing causality without a randomized trial. The logic: certain common genetic variants are reliably associated with higher or lower lifetime 25(OH)D. Those variants are inherited at random at conception, before any of the lifestyle confounders that plague observational studies. If lower 25(OH)D causally raises mortality, people with the genetic variants for lower 25(OH)D should have higher mortality. If the association is purely a marker of underlying poor health, they should not.

Across the published MR literature, the results have been mixed but instructive. A 2021 MR analysis using UK Biobank data, with over 300,000 participants, found a non-linear causal relationship: a substantial increase in all-cause mortality risk for people in the bottom 25(OH)D quartile compared with those in the highest, but essentially no further benefit above approximately 50 nmol/L [8]. A separate large MR analysis on cardiovascular outcomes was largely null for the normal range, with a signal restricted to severe deficiency.

The Autier and colleagues 2014 umbrella review in The Lancet Diabetes & Endocrinology reached the same overall conclusion from a different angle: across hundreds of trials and cohorts, the discrepancy between robust associational data and weak interventional data is most parsimoniously explained by reverse causation, not by undetected causal benefit at higher doses [9].

The current consensus from the best MR work: severe deficiency (<25–30 nmol/L) probably causally raises mortality risk. Insufficiency (30–50 nmol/L) probably does too, with weaker evidence. Above 50 nmol/L, the case for causal benefit of higher levels weakens substantially. Above 125 nmol/L, the observational upper limb of the U-curve has no causal genetic signal supporting benefit and some signal suggesting harm.

Reverse causation — low D as marker, not driver

Here is the uncomfortable possibility the MR work keeps gesturing at: the strong observational link between low 25(OH)D and bad outcomes may be largely driven by sick people having low 25(OH)D, rather than low 25(OH)D making people sick.

Mechanisms that lower 25(OH)D as a consequence of underlying illness:

• Systemic inflammation reduces circulating 25(OH)D directly. Active inflammatory states (chronic infection, autoimmune flares, advanced cancer) all lower 25(OH)D acutely, independent of any change in vitamin D status itself.
• Reduced sunlight exposure in people who are ill, frail, hospitalized, or chronically fatigued. The sickest people in any cohort spend the least time outdoors.
• Obesity and metabolic dysfunction sequester vitamin D in adipose tissue, lowering measured circulating 25(OH)D without necessarily reflecting true cellular insufficiency. Higher BMI and metabolic disease are independently linked to mortality.
• Lower physical activity tracks with both lower 25(OH)D (less outdoor time) and higher all-cause mortality through a hundred separate mechanisms.

The careful version of this argument was articulated in the Autier umbrella review: vitamin D is a near-perfect biomarker for being healthy in general. A high circulating 25(OH)D is consistent with being outdoors, mobile, lean, and not chronically inflamed. The downstream causality from the vitamin itself, beyond a deficiency threshold, is much smaller than the observational data suggests.

This is not a reason to ignore vitamin D status. It is a reason to stop treating it as a longevity dial you turn higher for more protection.

What "optimal" looks like in nmol/L and ng/mL

Pulling the cohort data, the MR data, and the trial data together, the optimal range for all-cause mortality risk converges on:

• 75–100 nmol/L (30–40 ng/mL) — the nadir of the U-curve across CopD, NHANES III, and the Schöttker meta-analysis. This is the range most major guidelines target for "sufficient."
• 50–75 nmol/L (20–30 ng/mL) — adequate for most outcomes, with mortality risk only modestly elevated. The Endocrine Society's clinical practice guideline (Holick et al.) defines >75 nmol/L as sufficient and 50–75 nmol/L as insufficient, but the line is not sharp [10].
• <50 nmol/L (<20 ng/mL) — frank deficiency. The Institute of Medicine treats this as the cut-off for inadequacy. This is the range where correcting status produces the clearest signal in MR and the cleanest gradient in the cohort data.
• >125 nmol/L (>50 ng/mL) — the upper limb of the U. Not toxic, but not protective either, and associated with modestly higher all-cause mortality in multiple large cohorts. This is the territory the megadose narrative pushes people into.
• >250 nmol/L (>100 ng/mL) — territory associated with frank hypercalcemia in case reports. Avoidable.

Unit conversion, because every supplement label and lab makes this unnecessarily annoying: 1 ng/mL = 2.5 nmol/L. A 25(OH)D of 30 ng/mL is 75 nmol/L. A level of 50 ng/mL is 125 nmol/L. North American labs report in ng/mL; most international literature uses nmol/L.

If you are low: how to actually supplement

Test first. There is no defensible reason to supplement vitamin D without a baseline serum 25(OH)D, because the entire shape of the evidence is non-linear. If you do not know where you are on the curve, you do not know which direction to move.

For most adults with a 25(OH)D below 50 nmol/L (20 ng/mL): 1,000–2,000 IU/day of cholecalciferol (vitamin D3) is sufficient to bring most people into the 75–100 nmol/L range over 8–12 weeks. D3 is preferable to D2 (ergocalciferol) on bioavailability grounds. Take with a fat-containing meal; vitamin D is fat-soluble and absorption with fat-containing meals is meaningfully better than on an empty stomach.

Recheck at 8–12 weeks. Adjust based on the result. People with higher body mass typically need higher doses to achieve the same serum level — somewhere between 1.5× and 2× the standard dose is a reasonable starting estimate, though individual variation is wide.

Megadosing — 10,000 IU/day or large infrequent bolus doses without medical supervision — is the part the data does not support. The clearest harm signal is the Sanders trial (JAMA, 2010): 2,256 community-dwelling women aged 70+ were randomized to a single annual oral dose of 500,000 IU of cholecalciferol or placebo, and the high-dose arm had a 15% higher rate of falls and a 26% higher rate of fractures, with the excess concentrated in the first three months after dosing [11]. Whatever the mechanism, "more is better" is not how this nutrient behaves.

Cofactors — magnesium and K2, briefly

Magnesium is a required cofactor for nearly every enzyme in the vitamin D activation pathway. Without adequate magnesium, the liver and kidney enzymes that convert vitamin D3 to 25(OH)D and then to the active 1,25(OH)2D struggle. Population magnesium intake is below recommended levels in most adults, which makes magnesium repletion the highest-leverage cofactor move. We've written about the magnesium–vitamin D switch in detail elsewhere, and about magnesium's well-established role in sleep regulation in the RCT literature.

Vitamin K2 (menaquinone, MK-7 in supplements) is proposed to direct calcium toward bone and away from soft tissue when vitamin D raises intestinal calcium absorption. The mechanistic case is reasonable. The clinical evidence for K2 co-supplementation extending lifespan or reducing vascular calcification in the general population is weak and based on small trials. We are not arguing against K2; we are arguing that it is a secondary, not foundational, lever.

For longevity-oriented readers, two other nutrients have cleaner per-paper signal than escalating vitamin D once you are replete: taurine as a population-level longevity associate, and urolithin A's effect on immune aging. Neither is a replacement for vitamin D status, but both are better-spent attention than another 5,000 IU.

A practical framework

We do not write protocols. We write frameworks to take to a clinician. With that established:

Frequently asked questions

What is the optimal serum vitamin D level for lowest mortality risk?
Across the largest cohort studies, the lowest all-cause mortality risk clusters around 75–100 nmol/L (30–40 ng/mL). Risk rises below 50 nmol/L and rises again above roughly 125 nmol/L. The curve is U-shaped, not linear.

Does taking more vitamin D give more benefit?
No. The dose–response is non-linear. Once you correct deficiency, more supplementation does not produce more mortality reduction in randomized trials. VITAL and DO-HEALTH both showed null primary endpoints in already-replete adults.

Is the link between low vitamin D and death actually causal?
Probably yes for severe deficiency, increasingly debated for the normal range, and likely confounded by reverse causation across most of the population distribution. Mendelian randomization studies support a causal effect at the bottom of the curve, not across the whole range.

How much vitamin D should I take if my level is low?
Get tested first. For 25(OH)D below 50 nmol/L, 1,000–2,000 IU/day of D3 is typically sufficient over 8–12 weeks. Recheck before adjusting. Not medical advice.

Do magnesium and K2 matter?
Magnesium is a required enzymatic cofactor and most adults are mildly insufficient — addressing it is worthwhile. K2's clinical evidence is weaker; the mechanistic case is reasonable but it is a secondary rather than foundational lever.