For an adult who already eats reasonably well, the supplements that fill most shopping baskets — a daily multivitamin, vitamin D, fish oil, a cocoa-flavanol capsule — do essentially nothing for the diseases they are sold against. In large randomized trials they sit at the bottom of the levers worth pulling, and no amount of dosing moves the ranking.
A short list of supplements are genuine levers — but each earns its place only in a specific circumstance: correcting a documented deficiency, the weeks around conception, or a high-risk patient taking a drug-strength dose under a clinician. The benefit belongs to the circumstance, not to the pill.
One class of supplement actively harms. High-dose antioxidants — beta-carotene above all — have raised cancer and death in randomized trials, not lowered them. More of a good nutrient is not a safe default.
The rule that sorts all of it: a supplement that corrects a shortfall in someone who is short answers a completely different question from the same pill added to a person who already has enough. The marketing rarely marks which question it is answering; the sections below do.
Ask whether the person is short before asking whether the supplement works
Every supplement decision splits on one prior fact: is this person deficient (below the requirement, where correcting the shortfall can move an outcome) or already replete (at the plateau, where more buys nothing, or past the upper bound, where more harms)? These are two arms of one dose-response curve, not two rival opinions about whether a nutrient matters (Manson et al., 2019; inferred from Martineau et al., 2017) -> Deficiency Repletion vs Enhancement.
The field makes one central error: it runs a trial on one arm and reads the result as a verdict on the other. A null from supplementing the already-replete says nothing about repletion of the deficient, and a benefit in the deficient does not license the replete to supplement. Almost every dispute about whether supplements work dissolves once you fix which arm was actually sampled.
For the well-nourished, the marketed stack is a null lever
VITAL is the cleanest test. It randomized 25,871 US adults selected by age alone (no cardiovascular disease or cancer at entry) to vitamin D3 2000 IU/day and/or marine n-3 1 g/day in a 2x2 design, median 5.3 years. Both agents missed both of their co-primary endpoints (Manson et al., 2019):
| Agent | Endpoint | HR (95% CI) | State |
|---|---|---|---|
| Vitamin D3 | Invasive cancer | 0.96 (0.88-1.06) | no meaningful effect |
| Vitamin D3 | Major CV events | 0.97 (0.85-1.12) | no meaningful effect |
| Marine n-3 | Major CV events | 0.92 (0.80-1.06) | no meaningful effect |
| Marine n-3 | Invasive cancer | 1.03 (0.93-1.13) | no meaningful effect |
«Supplementation with vitamin D was not associated with a lower risk of either of the primary end points.» (Manson et al., 2019) Fractures were null too — vitamin D3 did not reduce total (HR 0.98, 0.89-1.08) or hip (1.01, 0.70-1.47) fractures (LeBoff et al., 2022).
VITAL’s cohort was not deficient — the load-bearing fact here: mean baseline 25(OH)D was 30.8 ng/mL, only 12.7% below 20 ng/mL (Manson et al., 2019). So every null reads as adding a nutrient to someone who already has enough buys nothing — enhancement, not deficiency correction (LeBoff et al., 2022; inferred from Manson et al., 2019).
VITAL is one trial; a pooled anchor settles the mortality endpoint a single trial could not. Zhang’s meta-analysis of 50 RCTs of vitamin D alone (calcium-combination trials excluded), 74,655 participants, found supplementation did not lower all-cause mortality — RR 0.98 (0.95-1.02, I2=0%) — and this is a robust null, not an empty one: trial-sequential analysis reached the information size to exclude a 7.5-10% relative-risk reduction (Zhang et al., 2019) -> Vitamin and Mineral Supplements for Disease Prevention. One narrower endpoint did move: cancer mortality fell 15% (RR 0.85, 0.74-0.97), a signal confined to vitamin D3 trials (Zhang et al., 2019). Do not read that as an all-cause benefit — it is cancer death, not cancer incidence (VITAL’s own incidence stayed null at 0.96), it does not lift overall survival, and the authors hold the D3 subgroup as needing further evidence (inferred from Zhang et al., 2019).
One stratum breaks the vitamin-D null — and it breaks on baseline disease risk, not on repletion. VITAL recruited by age alone, so its cohort ran at average diabetes risk and was largely vitamin-D-replete, and it found nothing. Pittas’ 2023 individual-participant meta-analysis instead pooled three low-risk-of-bias RCTs built specifically for diabetes prevention (D2d, Tromso, DPVD; 4,190 adults with prediabetes, mean BMI 30, mean baseline 25(OH)D 63 nmol/L ~25 ng/mL), and there vitamin D slowed progression to type 2 diabetes: «Vitamin D reduced risk for diabetes by 15% (hazard ratio, 0.85 [95% CI, 0.75 to 0.96]) in adjusted analyses, with a 3-year absolute risk reduction of 3.3% (CI, 0.6% to 6.0%). The effect of vitamin D did not differ in prespecified subgroups.» (Pittas et al., 2023) (unadjusted ITT HR 0.88, 0.77-0.99; NNT 30).
This is route-(a) baseline-risk, not a reversal of VITAL — a different outcome in a different stratum. VITAL tested cancer, cardiovascular events and fractures in a replete, average-risk cohort; Pittas tested diabetes incidence in a prediabetic, higher-risk one, where a small relative effect on a high baseline risk turns into a non-trivial absolute one. It is not deficiency repletion: the cohort was not deficient (63 nmol/L) and the effect did not vary by baseline 25(OH)D — one route-(b) modifier aside, cholecalciferol worked in the leaner through a named activation-pathway mechanism -> Vitamin and Mineral Supplements for Disease Prevention.
The benefit tracks disease risk, not vitamin-D status, and Pittas draws the line the marketing blurs: «the results do not apply to the general healthy population and should not be extrapolated to those who are at average risk» (Pittas et al., 2023) (Manson et al., 2019; inferred from Pittas et al., 2023) -> Baseline Risk and the Relative-Absolute Split, Deficiency Repletion vs Enhancement.
Even in its own stratum the lever is small. The 15% relative reduction sits against 58% for intensive lifestyle change and 31% for metformin in the Diabetes Prevention Program — «a number of persons with prediabetes needed to treat of 30 (compared with 7 with intensive lifestyle modification and 14 with metformin in the Diabetes Prevention Program study)» (Pittas et al., 2023). For a prediabetic the prevention rock is lifestyle first, then metformin; vitamin D is a cheap, low-harm add-on, not a substitute -> Lifestyle vs Metformin for Diabetes Prevention, Layer 1 - Ranking Interventions for a Stratum.
One scope bound on the omega-3 null: the tested dose is the 1 g/day low-dose consumer form in the replete. It does not speak to high-dose prescription marine omega-3 (~4 g/day ethyl ester) in the hypertriglyceridaemic, which is a different exposure on a different stratum; the one high-dose benefit signal carries a contested comparator, and a clean-comparator outcome trial is not held here — a named gap, not a settled null .
But omega-3 supplementation is not merely inert as the dose rises — it carries a demonstrated harm. Gencer’s meta-analysis of 7 marine-omega-3 CV-outcome RCTs (81,210 participants) found the supplements raised incident atrial fibrillation, dose-dependently: overall HR 1.25 (1.07-1.46); low-dose <=1 g/day HR 1.12 (1.03-1.22); high-dose >1 g/day HR 1.49 (1.04-2.15); and per 1 g/day increment HR 1.11 (1.06-1.15) (Gencer et al., 2021). The signal is robust — it survives dropping REDUCE-IT (HR 1.23) (Gencer et al., 2021). AF here is an incident, patient-important RCT outcome, not a surrogate, so this is a genuine dose-dependent harm arm on the supplement (isolate) form — running opposite the dose-dependent CV benefit — and it does not touch any fish-as-food claim -> Omega-3 Supplementation and Atrial Fibrillation.
An independent trial set reaches the same fracture-null. Kahwati’s USPSTF review pooled 11 RCTs
(N=51,419) of community-dwelling adults — using no VITAL trial, its search closing before VITAL
published — and found vitamin D +/- calcium did not reduce fractures (D+Ca total-fracture ARD -0.35%,
-1.02 to 0.31) (Kahwati et al., 2018). Because
the two share no trials, no group, no lineage, this is genuine independent agreement on the
enhancement-arm fracture-null, scoped to that claim [E-independent]
(inferred from Kahwati et al., 2018; LeBoff et al., 2022).
The guideline map agrees and keeps the evidence states distinct (US Preventive Services Task Force, 2022):
| Supplement | USPSTF grade (general adults) | State |
|---|---|---|
| Beta carotene | D — recommend against | harm outweighs benefit |
| Vitamin E | D — recommend against | no meaningful effect on hard outcomes |
| Multivitamins | I — insufficient | evidence lacking, balance undeterminable |
| Other single/paired nutrients | I — insufficient | evidence lacking |
Keep those two nulls apart: vitamin E is a graded null; multivitamins are insufficient evidence, a different state (a modest cancer-incidence signal, discordant with mortality, on short follow-up) — not demonstrated benefit, and not the same as a demonstrated null (inferred from US Preventive Services Task Force, 2022) -> Rating Certainty of Evidence.
COSMOS ran the same test on a newer marketed supplement — a cocoa-flavanol extract, sold for heart health. It randomized 21,442 older US adults free of major cardiovascular disease to a cocoa-extract capsule (500 mg/day cocoa flavanols) and/or a multivitamin, median 3.6 years, and its primary endpoint was null: total cardiovascular events HR 0.90 (0.78-1.02, P=0.11), an annualized 1.08% vs 1.20% (Sesso et al., 2022). Do not let its one favourable secondary stand in for that null: CVD death fell to 0.73 (0.54-0.98) and a per-protocol total-CVD figure to 0.85 (0.72-0.99), but the authors flag the per-protocol result «should be interpreted with caution» — no multiplicity control, sitting under a null composite (Sesso et al., 2022). That is the same discipline VITAL’s total-MI secondary required.
The null is not an adherence artifact. Treatment produced «a more than 3-fold increase compared with placebo in gVLM concentrations, with an overall ratio of geometric means (95% CI) of 3.23 (2.84, 3.67; P < 0.001)» (gVLM is a validated flavanol-intake biomarker), so the flavanol was absorbed and the composite still did not move (Sesso et al., 2022).
COSMOS sits in this section’s frame twice over: the 500 mg/day dose «substantially exceeds the mean intake reported in Europe of 105 mg/d», so it is enhancement in the already-replete; and even this RCT does not isolate the flavanol — the capsule contained «all naturally occurring bioactive components of the cocoa bean … thus, we cannot disentangle the effects of its individual components» (Sesso et al., 2022). A cocoa-extract supplement is a different exposure from cocoa the food -> Is the Food Category Doing Any Work.
Some antioxidant supplements shorten lives rather than lengthen them
The harm arm is the sharpest decision-change here, because it runs opposite to the nutrients are good, so more must be better intuition. Beta carotene is the paradigm case. CARET randomized 18,314 smokers and asbestos-exposed workers to beta carotene 30 mg/day plus retinol and was stopped 21 months early: «The active-treatment group had a relative risk of lung cancer of 1.28 (95 percent confidence interval, 1.04 to 1.57 …» (Omenn et al., 1996), with all-cause death RR 1.17 and lung-cancer death RR 1.46 in this stratum (Omenn et al., 1996).
The harm is broader than one nutrient in one stratum. A meta-analysis of antioxidant RCTs found that, restricted to the methodologically sound trials, supplemental antioxidants raised all-cause mortality: «In 47 low-bias trials with 180 938 participants, the antioxidant supplements significantly increased mortality (RR, 1.05; 95% CI, 1.02-1.08) … beta carotene (RR, 1.07 …), vitamin A (RR, 1.16 …), and vitamin E (RR, 1.04 …), singly or combined, significantly increased mortality.» (Bjelakovic et al., 2007) The method lesson is load-bearing: the all-trials pool was null (RR 1.02, 0.98-1.06) and the harm surfaced only after excluding high-bias trials, in which the supplements looked spuriously protective (Bjelakovic et al., 2007) — risk-of-bias appraisal was decision-determining, and the finding is about synthetic isolates, which the authors say «should not be translated to potential effects of fruits and vegetables» (Bjelakovic et al., 2007) -> Is the Food Category Doing Any Work.
A cancer authority converts this into an instruction. WCRF’s Third Expert Report recommends «Do not use supplements for cancer prevention» (World Cancer Research Fund & American Institute for Cancer Research, 2018), resting on the same beta-carotene trials — a same-evidence echo, not independent backing. It adds one wrinkle worth the net-effect discipline: «Although taking calcium supplements helps protect against colorectal cancer, some trials for other cancer sites have shown potential for unexpected adverse effects» (World Cancer Research Fund & American Institute for Cancer Research, 2018) — a site-specific benefit does not license the supplement, because the whole-body ledger is what a recommendation weighs.
Correcting a documented deficiency is a different, real lever
Where a person is genuinely short, repletion moves patient-important outcomes — and the benefit scales
with how deficient they were. Repletion is the other arm of the same curve, and three independent nutrients
now demonstrate it, sharing no trials, no group, and no lineage [E-independent] on the principle
(Bougma et al., 2013; Falkingham et al., 2010; inferred from Martineau et al., 2017):
| Nutrient -> outcome | Deficient stratum | Replete stratum | Certainty |
|---|---|---|---|
| Vitamin D -> acute respiratory infection | OR 0.30 (0.17-0.53), NNT 4 (<25 nmol/L, daily/weekly) | OR 0.75, NNT 15 | GRADE high (25 RCTs) |
| Iron -> IQ (children) | +2.5 IQ points (1.24-3.76) if anaemic | null, well-powered | moderate (small RCTs) |
| Iodine -> child IQ | +7.4 IQ points (supplementation trials) | no supplement-the-replete arm | low (2 RCTs) |
Provenance: vitamin D ARI overall OR 0.88 (0.81-0.96), NNT 33, deficient/replete split (P for interaction 0.006) and the null of bolus dosing (OR 0.82, 0.51-1.33) (Martineau et al., 2017); iron IQ (Falkingham et al., 2010). The iodine 7.4 is the supplementation-trial effect; separately, deficient children score 6.9-10.2 IQ points lower than sufficient children in observational data — a different quantity, not an interval on 7.4 (Bougma et al., 2013). Martineau states the mechanism as the principle itself: «Increased efficacy of vitamin D supplementation in those with lower baseline vitamin D status is more readily explicable, based on the principle that people who are the most deficient in a micronutrient will be the most likely to respond to its replacement.» (Martineau et al., 2017)
A fourth repletion stratum is defined by the diet, not a lab value: vegetarians and vegans have no appreciable plant source of vitamin B12, so biochemical depletion is common — though the reported prevalence is wide and cutoff-driven, «from about 11 to 90%» across 18 studies using accurate biomarkers, with serum B12 itself unreliable (MMA and holo-TCII are the accurate markers) (Pawlak et al., 2013).
But this case stops where the three above do not: repletion corrects a marker, not a demonstrated clinical harm. In the one study that looked, «none of the vegetarians included in their study had clinical symptoms despite the fact that about two-thirds of the sample had B12 depletion or deficiency, as indicated by both low holo-TCII and elevated MMA» (Pawlak et al., 2013). So marker-to-outcome transmission is insufficient evidence, held apart from the well-supported prevalence and measurement facts -> Vitamin B12 Status in Vegetarian and Vegan Diets.
Two disciplines keep this honest. First, repletion is necessary, not sufficient: the benefit is often gated by a further condition — vitamin D worked only on daily/weekly (not bolus) dosing, and iodine only inside the fetal-brain window (early pregnancy d 0.51 vs late 0.17) (Bougma et al., 2013). Second, establishing the deficiency is a prescriber act needing this person’s labs — out of scope here. The move is test, then replete the confirmed shortfall, not supplement broadly and hope.
Folic acid before pregnancy is the clearest supplement win the evidence holds
Periconceptional folic acid is the standing counter-example to supplements are a distraction — the cleanest benefit on a hard outcome (a birth defect, not a surrogate) the fabric holds. Folate versus none cut neural-tube defects to RR 0.31 (95% CI 0.17-0.58) — 5 RCTs, 6708 births, GRADE high; absolute risk 35 -> 11 per 1000; recurrence RR 0.34; no harm signal (miscarriage RR 1.10) (De-Regil et al., 2015).
It is a distinct flavour of repletion, and forcing it into the deficiency-status pattern above would misread it (inferred from De-Regil et al., 2015). Life-stage defines the stratum (a woman planning or capable of pregnancy), not a lab value; the protective target sits ~9x above the deficiency threshold (optimal RBC folate ~906 nmol/L), so it is not correcting a frank shortfall (De-Regil et al., 2015); and it is window-gated — the neural tube closes within ~28 days of conception, before most women know they are pregnant, so supplementation must be in place beforehand (De-Regil et al., 2015). Higher-risk women (prior NTD pregnancy, diabetes, anticonvulsants) take 5 mg/day rather than the general 0.4 mg (De-Regil et al., 2015). For this stratum a supplement flips from bottom-of-hierarchy to a genuine big rock.
Calcium plus vitamin D helps only the frail elderly, and carries its own harms
Bone brings the repletion/enhancement split, the stratum question, and the net-effect ledger together — and the answer is not one number. A gold meta-analysis (Yao) runs the arms in parallel and they disagree by design (Yao et al., 2019):
| Arm | Any fracture | Hip fracture | Read |
|---|---|---|---|
| Vitamin D alone (11 RCTs) | RR 1.06 (0.98-1.14) | RR 1.14 (0.98-1.32) | null (dose-constrained) |
| Vitamin D + calcium (6 RCTs) | RR 0.94 (0.89-0.99) | RR 0.84 (0.72-0.97) | 6% / 16% reduction |
Vitamin D alone does not prevent fracture at the doses trialled; “dose-constrained” is literal — the pooled D-alone RCTs never delivered a large 25(OH)D increment, so this null bounds the studied dose range, not vitamin D as such, and the higher-dose D-alone question stays open (inferred from Yao et al., 2019). The fracture benefit is the combination, so calcium is the active co-ingredient — and it concentrates in the older institutionalized: «more extreme changes in risk of any fracture in the RCTs of older participants (ie, aged 80 years) living in an institution» (Yao et al., 2019). That is the repletion arm (older, often vitamin-D-deficient, low dietary calcium); the community-dwelling replete get little to nothing — which is exactly why Kahwati, sampling only community-dwellers, finds the null. The two are a distinction by population, not a contradiction: their hip-fracture point estimates barely differ (0.84 vs 0.88) and their intervals overlap (Kahwati et al., 2018; inferred from Yao et al., 2019).
Then net that benefit against the calcium supplement’s own harms, and the ledger flips by stratum:
- Kidney stones (D+Ca), moderate certainty. Vitamin D combined with calcium raised incident stones — pooled ARD 0.33% (0.06-0.60), RR 1.18 (1.04-1.35); calcium alone did not (Kahwati et al., 2018).
- Cardiovascular (contested). Bolland found calcium +/- D raised myocardial infarction (RR 1.24, 1.07-1.45) and MI-or-stroke (RR 1.15, 1.03-1.27) (Bolland et al., 2011). Hold it as a counterweight rather than a settled harm — subgroup-derived and disputed — but the direction has firmed, not faded: a later pooling of 13 double-blind RCTs (28,935, though CV was mostly a secondary endpoint) reproduced it (CVD RR 1.15 [1.06-1.25], MI RR 1.25 [1.07-1.45], I2=0%), concentrated in the low-risk-of-bias trials (Myung et al., 2021). Myung firms without settling it, re-pooling Bolland’s own trials and unpublished data (type-F, not an independent witness) (Bolland et al., 2011; inferred from Myung et al., 2021). Bolland’s own ledger — six extra CV events per 1000 over five years (NNH 178) versus three fractures prevented (NNT 302) (Bolland et al., 2011) — was computed in community-dwelling healthy women, the enhancement arm, where the fracture benefit is weakest.
The exposure is the supplement bolus, not food calcium: Bolland’s effect is «independent of dietary calcium intake», its proposed mechanism the abrupt serum-calcium spike after a pill (Bolland et al., 2011), so it does not transport to dairy -> Is the Food Category Doing Any Work. Decision by stratum: for the community-dwelling replete, the whole-strategy ledger runs against the supplement (no benefit + stone harm + a contested CV signal), and the fracture lever is exercise, not the pill -> Exercise for Preventing Falls in Older Adults; for the institutionalized / deficient / low-calcium stratum, daily D+Ca is a small, defensible hip-fracture lever (Bolland et al., 2011; Kahwati et al., 2018; inferred from Yao et al., 2019).
Protein and creatine work on the gym floor, not in the medicine cabinet
Two supplements have solid randomized evidence — but on surrogates (muscle mass, strength, performance), not on disease or death, and both only as adjuncts to training that is doing the real work. They belong to a different decision than the disease-prevention stack above, and pooling them with it would be a category error (inferred from Kreider et al., 2017; Morton et al., 2017) -> Surrogate Outcomes.
- Protein added to resistance training raised 1RM strength by 2.49 kg (0.64-4.33) and lean mass by 0.30 kg (0.09-0.52) — small increments on top of what the training delivers, with a diminishing-returns region near ~1.6 g/kg/day (a weak, non-significant knee, p=0.079, CI 1.03-2.20 — hold it loosely) (Morton et al., 2017), (Morton et al., 2017). Protein quality qualifies the gram target: low-DIAAS plant sources deliver fewer indispensable amino acids per gram than milk (peas 64, wheat 40 vs milk 122) (Food and Agriculture Organization of the United Nations, 2013). Detail: Protein and Resistance Training for Muscle and Strength.
- Creatine monohydrate raises high-intensity, repeated-effort performance — «generally increased by
10-20% depending on the magnitude of increase in muscle PCr» — is safe at 3-30 g/day («no
compelling evidence that creatine supplementation negatively affects renal function»), and the cheapest
monohydrate form is the best-evidenced one
(Kreider et al., 2017),
(Kreider et al., 2017). Its long list of
clinical claims (neuroprotection, neurodegenerative disease) is insufficient / animal / null — the
stand’s own large trials found no benefit in Parkinson’s or ALS — and the source is an industry-funded
position stand, so the page is held at
confidence: low(Kreider et al., 2017). One useful decision-change: creatine raises serum creatinine without harming the kidney, so disclose it before a creatinine test (inferred from Kreider et al., 2017). Detail: Creatine Supplementation.
For the elderly, where muscle and strength are themselves patient-important (sarcopenia), both rise in rank — creatine plus resistance training beat training alone on muscle, strength and function (Kreider et al., 2017). Still an adjunct; the training is the driver.
Fibre and probiotic supplements are small or strain-specific levers, not tonics
- Fibre supplements move a surrogate modestly. As a dosable isolate, viscous/soluble fibre lowers LDL by about -0.057 mmol/L per gram in the practical 2-10 g/day range, which the author frames plainly: 3 g soluble fibre «can decrease total and LDL cholesterol by <0.13 mmol/L … can make only a small contribution to dietary therapy to lower cholesterol» (Brown et al., 1999). The larger benefits of fibre are for food patterns, appraised at Dietary Fibre and Health, not a fibre pill.
- Probiotics are not a general product — «the effects of probiotics are strain-specific and dose-specific» (World Gastroenterology Organisation et al., 2023), so take a probiotic for gut health is not a claim the evidence supports. Where specific strains have evidence, they earn it by indication: the standout patient-important outcome is preventing necrotizing enterocolitis in preterm neonates (NNT 20 to prevent one death), with strain-specific benefit for antibiotic-associated and C. difficile diarrhoea and some IBS symptoms — but «not all probiotic preparations tested are effective» (World Gastroenterology Organisation et al., 2023). The strain, not the word, is the exposure -> Gut Microbiome and Health.
Attention runs opposite to effect, and the aisle carries its own risk
The supplements with the most marketing and shelf space are, with few exceptions, the ones with the smallest or null effects — attention is an anti-signal here, and ranking by expected effect x certainty puts the whole disease-prevention stack near the bottom of what a well-nourished person should do -> Layer 1 - Ranking Interventions for a Stratum. The contrast with the big rocks is stark: quitting smoking, for one, buys back most of a lost decade — «Cessation before the age of 40 years reduces the risk of death associated with continued smoking by about 90%.» (Jha et al., 2013) No attainable precision about a multivitamin changes that ordering.
Two further costs are real and specific to supplements. The money and adherence spent on a null stack are a layer-3 opportunity cost the trial evidence leaves to the person (US Preventive Services Task Force, 2022). And the industry is loosely regulated: contamination, adulteration and mislabelled dose are a harm vector independent of any compound’s own evidence.
What the evidence here cannot yet say
These are named gaps — absence of held evidence, not evidence of absence :
- The mechanistically-plausible, heavily-marketed stack items — magnesium, CoQ10, curcumin, berberine, ashwagandha, NAC, glycine, inositol — have no held human-outcome meta-analysis. Each sits at insufficient evidence, neither asserted nor dismissed, until a source lands.
- A dedicated deficiency-prevalence / repletion reference for the Tier-2/3 minerals (magnesium, zinc, iron, iodine, calcium) is not held beyond the vitamin-D family, so the repletion sub-question for those is anchored only indirectly.
- A same-outcome dose-response traced from a nutrient’s deficient edge to its plateau is not held for any single endpoint — the repletion instances above each read the deficient-benefit arm on outcomes their replete-arm evidence does not directly test.
The bottom line
- If you are a generally-nourished adult: skip the multivitamin, vitamin D, fish oil, cocoa-flavanol and antioxidant stack for disease prevention — the best trials show no benefit, and spend the effort on the big rocks (not smoking, activity, weight, sleep, a decent dietary pattern) instead.
- Do not take high-dose antioxidant supplements, beta-carotene especially — in randomized trials they raised cancer and death, not lowered them.
- If you are planning or capable of pregnancy: take 400 ug/day folic acid periconceptionally (5 mg if higher-risk) — a real lever on a hard outcome, and one of the few supplements that clearly works.
- Correct a documented deficiency, don’t supplement blind. Repletion (vitamin D, iron, iodine) helps the deficient on real outcomes and scales with severity; confirm the shortfall with a clinician first, and use the schedule the evidence used (e.g. daily/weekly vitamin D, not a bolus).
- If you have prediabetes: vitamin D is a small, cheap add-on, not the lever. In trials built for the question it slowed progression to type 2 diabetes by about 15% (NNT 30) — but intensive lifestyle change (NNT 7) and metformin (NNT 14) are far larger; use those first. This is a higher-risk stratum, not the general population, and it does not revive the multivitamin/vitamin-D case for the well-nourished.
- Reserve calcium + vitamin D for the frail/institutionalized/deficient elderly, where a small hip-fracture benefit exists; for a healthy community-dweller it does not, and it carries kidney-stone and contested cardiovascular harms — exercise is the fracture lever.
- Treat protein and creatine as training adjuncts on surrogates, not disease-prevention supplements — useful if you resistance-train (and if muscle preservation is the goal in older age), inert otherwise.
Caveats
- The loop is open. Nothing here grades a supplement against a realized outcome in a real person; the wiki checks internal coherence and source-fidelity, never validity. A clean audit is not a validated recommendation.
- This appraises evidence; it does not prescribe. Selecting, dosing, screening for the deficiency that would justify repletion, and managing interactions (vitamin K vs anticoagulants, iron loading, fat-soluble toxicity) are prescriber acts requiring this person’s labs, medications and history.
- Findings are population-level by default. The estimates above describe strata (deficient vs replete, community-dwelling vs institutionalized, training vs not), and the right supplement for a given person depends on which stratum they are in — established per case, not read off the average.
- One health axis only. This cut weighs supplements on longevity, healthspan, function and absence of deficiency. It prices no cost, sustainability or convenience axis; those weightings are the person’s.
Evidence box
Question What does the evidence show about dietary supplements — where, for whom, and how large is any effect on patient-important outcomes, and how do they rank against the big rocks? Evidence included 26 sources — 11 gold, 14 high, 1 moderate Overall certainty Medium (see Rating Certainty of Evidence) Source-selection note 1 source(s) below the gold evidence bar feed this page: Kreider (consensus, moderate). Each labelled by tier; none load-bearing for the core claims. Last updated 2026-09-05 · Independently reviewed: No · Full edit history