The distinction
Supplementing a nutrient does not answer one question — it answers two, and which one depends on where the person already sits. A nutrient’s effect on a patient-important outcome is status-dependent, so the same pill means different things to different people:
- Deficiency repletion — the person is below the requirement; correcting the shortfall can move an outcome. This is the lower arm of the curve.
- Enhancement — the person is already replete; adding more sits on the plateau (buys nothing) or past the upper bound (harms). This is the upper arm.
These are different questions on different parts of one dose-response curve, not two rival claims about whether the nutrient works. The central supplements error is to run a trial on one arm and read the result as a verdict on the other: a null (or harm) from supplementing the already-replete says nothing about repletion of the deficient, and vice versa.
It is one curve — the dose-response frame
The distinction is a specific reading of the nutrient dose-response curve, whose features the method layer already names (threshold · knee · plateau · upper bound):
- Threshold — the deficiency edge; below it, repletion is on the steep part and can benefit.
- Plateau — replete; more buys nothing (the enhancement null).
- Upper bound — excess harms (the antioxidant / beta-carotene arm below).
So does the supplement work? is ill-posed until the person’s position on the curve is fixed. The question is not whether but where.
The held evidence, by arm
The parameter that decides everything is the population’s baseline nutrient status, so read every supplement trial by which arm it actually sampled:
| Tested quantity | VITAL (Manson 2019) | USPSTF 2022 | Bjelakovic 2007 / CARET 1996 | Same arm? |
|---|---|---|---|---|
| Population nutrient status | replete: mean 25(OH)D 30.8 ng/mL, 12.7% below 20 | community-dwelling, non-deficient («does not apply to … known nutritional deficiency») | mostly replete / at-risk (CARET: smokers + asbestos-exposed) | VITAL and USPSTF both sample the non-deficient stratum; Bjelakovic/CARET push dose past need |
| Intervention | vitamin D 2000 IU + omega-3 1 g/day, RCT | multivitamins / individual vitamins, RCT-pooled | antioxidants (47 low-bias RCTs); CARET beta carotene 30 mg + retinol | supplement-in-the-non-deficient vs high-dose antioxidant |
| Finding on a hard outcome | null: cancer HR 0.96, major CV 0.97, fractures 0.98 | no net benefit (multivitamins I; beta carotene / vitamin E graded D) | harm: antioxidants all-cause RR 1.05; beta carotene 1.07; CARET lung cancer 1.28 | plateau (null) vs upper bound (harm) — same status-dependent shape, different nutrients (not one nutrient’s measured curve) |
Provenance for the cells: VITAL status + null (Manson et al., 2019), (Manson, Cook, Lee, Christen, Bassuk, Mora, Gibson, Albert, et al., 2019), fractures + the <12 ng/mL tail (LeBoff et al., 2022); USPSTF grades + carve-out (US Preventive Services Task Force, 2022); antioxidant harm (Bjelakovic et al., 2007); CARET (Omenn et al., 1996).
Independence caveat — this is NOT type-E convergence. VITAL and USPSTF do not independently confirm each other: USPSTF 2022 pools VITAL, and CARET is a primary trial inside both the USPSTF beta-carotene evidence and the Bjelakovic meta-analysis. The value here is not several sources agree — it is the structure the sources jointly imply, which none states alone.
Enhancement in the replete tests null-to-harm
Two of the three arms are well-populated, and both land against the supplement:
- VITAL dosed a replete population and found nothing move — «Supplementation with vitamin D was not associated with a lower risk of either of the primary end points» (Manson, Cook, Lee, Christen, Bassuk, Mora, Gibson, Gordon, et al., 2019); and within the range it could study, the effect did not vary by baseline 25(OH)D, so the null is not rescued by it only helps the low-D people inside the studied range -> Vitamin and Mineral Supplements for Disease Prevention.
- USPSTF graded supplementation for a non-deficient population and found no net benefit — and it draws the repletion line in its own scope statement: the recommendation «does not apply to … persons who are chronically ill, are hospitalized, or have a known nutritional deficiency. The USPSTF separately recommends that all persons who are planning or capable of pregnancy take a daily supplement containing 0.4 to 0.8 mg (400-800 μg) of folic acid» (US Preventive Services Task Force, 2022). The folic-acid carve-out is the enhancement-vs-repletion line made explicit by the guideline itself: a need (repletion) is a different recommendation from a prevention claim (enhancement).
- Bjelakovic + CARET populate the upper bound: in 47 low-bias trials the antioxidant supplements «significantly increased mortality (RR, 1.05; 95% CI, 1.02-1.08)» (Bjelakovic et al., 2007), and CARET’s beta-carotene arm raised lung cancer (RR 1.28) (Omenn et al., 1996). More was not merely useless; past need it was harmful.
The enhancement-null now has a SAME-OUTCOME second witness — vitamin D/calcium and FRACTURES (Kahwati 2018)
The fracture endpoint was previously held only through VITAL (LeBoff 2022) — a single trial’s replete-null. Kahwati 2018 (USPSTF gold SR, 11 RCTs, N=51,419) supplies an independent, same-outcome confirmation, and — unusually — the review names the deficiency-repletion split in its own voice. It sampled the enhancement arm by exclusion: «studies of populations that were institutionalized or had known vitamin D deficiency, osteoporosis, or prior fracture were excluded» (Kahwati et al., 2018), and found no fracture benefit — «Vitamin D supplementation alone or with calcium was not associated with reduced fracture incidence among community-dwelling adults without known vitamin D deficiency, osteoporosis, or prior fracture» (Kahwati et al., 2018) (vit-D+Ca total fracture ARD -0.35% [-1.02 to 0.31], hip -0.14%; SoE low).
- The review states the repletion pattern itself: «The fracture benefits overall appear to be largely attributable to benefits among the high-risk populations, with little to no benefit in lower-risk populations (1 fewer hip fracture per 1000 community-dwelling adults per year [95% CI, 0 to 2])» (Kahwati et al., 2018) — Kahwati reporting the 2014 Cochrane review, which included institutionalized/deficient/osteoporotic populations. So on fractures the two arms are now both named: null in the community-dwelling (direct), benefit in the high-risk/deficient (reported, second-hand via Cochrane).
- Independent same-outcome convergence on the fracture enhancement-null
[E-independent]. Kahwati’s 11 RCTs (search cutoff March 2017) do not include VITAL (published 2019/2022), share no trials, no research group, and no lineage with it — so VITAL (LeBoff) and Kahwati are two independent trial sets reaching the same result on the same outcome (no fracture benefit from D +/- Ca in the non-deficient). This is genuine type-E backing, raising confidence in the fracture enhancement-null specifically — strictly that claim, not any point estimate (inferred from Kahwati et al., 2018; LeBoff et al., 2022; Manson, Cook, Lee, Christen, Bassuk, Mora, Gibson, Gordon, et al., 2019). - Route-c harm on this arm: vitamin D + calcium raised kidney stones — pooled ARD 0.33% (0.06-0.60), RR 1.18 (1.04-1.35), SoE moderate (Kahwati’s only above-low grade); calcium alone did not (ARD 0.00%) and vitamin D alone was unstudied (Kahwati et al., 2018). A concrete contraindication-stratum harm from combining, not a pure upper-bound overdose effect.
The high-risk/institutionalized BENEFIT leg is now DIRECT, not second-hand (Yao 2019) [F, NOT E]
Kahwati named the deficient/high-risk benefit only by reporting Cochrane (the «1 fewer hip fracture per 1000» line above, second-hand). Yao 2019 (gold SR+MA) supplies the direct, within-review demonstration of that leg — so both arms of the fracture curve are now anchored in primary syntheses, not one direct + one reported:
- D + calcium reduces hip fracture, and the benefit concentrates in the institutionalized (the repletion arm). «found a 6% reduced risk of any fracture (RR, 0.94; 95% CI, 0.89-0.99) and a 16% reduced risk of hip fracture (RR, 0.84; 95% CI, 0.72-0.97)» (Yao et al., 2019), with the effect «more extreme … in the RCTs of older participants (ie, aged 80 years) living in an institution than those younger than 80 years living in the community (P for heterogeneity = .02)» (Yao et al., 2019). That is the deficient/high-risk arm shown directly, where Kahwati (community-dwelling only) sees the enhancement null.
- Vitamin D ALONE is null — calcium is the active co-ingredient. D alone «did not find a reduced risk of any fracture (RR, 1.06; 95% CI, 0.98-1.14) or hip fracture (RR, 1.14; 95% CI, 0.98-1.32), but these trials were constrained by infrequent intermittent dosing, low daily doses of vitamin D, or an inadequate number of participants» (Yao et al., 2019) — so the fracture lever, where it exists, is the combination, not vitamin D by itself (and the D-alone null is partly dose-constrained: median 25(OH)D difference 8.4 ng/mL, most trials sub-2000 IU/day).
- This is F/refinement, NOT independent type-E. Yao’s D+Ca pool shares its dominant trial with
Kahwati (WHI/Jackson 2006 dominates both pools; Porthouse 2005 also shared), so it is not an independent
second witness — it upgrades the second-hand high-risk-benefit leg to direct evidence and adds the
D-alone/D+Ca split, no
[E-independent]claimed (Kahwati et al., 2018; inferred from Yao et al., 2019). The net-effect of the D+Ca benefit against its CV and kidney-stone harms lives on Vitamin D and Calcium Supplementation for Fracture Prevention.
The fracture enhancement-null now generalizes across EXPOSURE FORM — dietary dairy fails too (Malmir 2019) [E-independent + F]
The supplement witnesses (VITAL, Kahwati) share a form: an isolated Ca/D pill in an RCT. That leaves an escape hatch — supplements are the wrong delivery; real dairy food, with its calcium-plus-protein matrix, would protect bone. Malmir 2019 (gold observational SR-MA, 34 studies, 616k people for fracture) removes its empirical support (in cohorts): dietary milk/dairy is null for hip fracture in prospective cohorts (total dairy RR 0.90 [0.73-1.11]; milk 0.93 [0.75-1.15], with milk trending to +9%/200 g harm), protective only in the reverse-causation-prone cross-sectional/case-control designs -> Dairy and Bone Health. Malmir also draws the parallel to the supplement evidence in his own voice — supplemental calcium and/or vitamin D «was not associated with a lower risk of fractures among community-dwelling older adults» (Malmir et al., 2019) (Zhao 2017, secondhand) — but that parallel-drawing is not part of the independence (Malmir citing the same supplement literature Kahwati summarizes adds no corroborative weight); the independence rests only on Malmir’s own dairy-cohort data.
- A THIRD independent route to the fracture enhancement-null
[E-independent]— the most methodologically distant of the three. Malmir (observational cohorts, food-source) shares no trials, no research group, no lineage with the supplement RCTs (VITAL, Kahwati), and it differs on both axes at once — a different exposure form (whole-food matrix vs isolate) AND a different design family (observational vs randomized). So it is not merely a third witness of the same kind: it establishes that the fracture-null does not depend on the supplement form. More calcium, however delivered, does not lower fracture in the enhancement stratum — Malmir’s general/unscreened (predominantly replete) adult and the RCTs’ verified-non-deficient are distinct populations, joined only in being enhancement-territory, not the deficient arm (Kahwati et al., 2018; LeBoff et al., 2022; inferred from Malmir et al., 2019). - Scope guards (strict). The convergence is on the null direction only, not on any point estimate — the magnitudes are non-commensurable (RR-per-200 g food vs RCT ARD; the parameter table on Dairy and Bone Health carries the NO cells). And Malmir does not screen calcium/vitamin-D status, so it samples more dairy across the general (predominantly replete Western) adult, which is enhancement-territory but not verified-replete; it therefore strengthens the enhancement leg, not the deficient-repletion arm, which stays a separate question (a genuinely calcium/D-deficient or osteoporotic person is not what Malmir sampled — Kahwati’s secondhand high-risk benefit still governs there).
The enhancement-null now extends to a NEW ENDPOINT — MORTALITY (Zhang 2019) [F, NOT E]
The enhancement arm was held on fractures/cancer/CVD (VITAL, Kahwati, Yao, Malmir). Zhang 2019 (gold SR-MA, 50 RCTs of vitamin D alone, 74,655) adds MORTALITY as a new enhancement-null endpoint, in a population that was largely replete: «more than half of participants (50 466/66 546) from trials reported a baseline mean 25 hydroxyvitamin D concentration of more than 50 nmol/L» (Zhang et al., 2019). Pooled all-cause mortality was null (RR 0.98, 0.95-1.02, GRADE high), and the authors read it as an enhancement problem in their own voice: «The baseline 25 hydroxyvitamin D concentrations of trial participants have not been low enough, which could partly contribute to the null finding» (Zhang et al., 2019). This is type-F (Zhang pools VITAL, so not an independent second witness), broadening the enhancement-null to mortality.
- A weak, NS hint of the lower arm on this same endpoint. Baseline-25(OH)D subgroup (all-cause): >=50 nmol/L RR 1.04 (0.97-1.12) vs <50 RR 0.95 (0.90-1.01), P-interaction 0.07 (Zhang et al., 2019) — the predicted direction (more benefit lower down), but non-significant and route-(b), so a hint, not the demonstrated lower arm Martineau supplies for ARI. The trial population barely sampled the deficient edge (<25 nmol/L in 4 trials / 886 people), so it cannot power the repletion arm on mortality.
- The observational gradient the RCTs do NOT reproduce — a reverse-causation candidate. Cohort data (Zhang citing Gaksch 2017 IPD MA) show a steep low-25(OH)D mortality signal: HR vs the 75-100 nmol/L referent was 1.67 (1.44-1.89) at <30 nmol/L, 1.33 at 30-40, 1.15 at 40-50 (Zhang et al., 2019). Supplementing largely-replete trial populations does not confirm this gradient as causal — exactly the pattern where reverse causation / confounding-by-frailty inflate an observational low-nutrient mortality arm -> The U-Shaped Association Artifact. (inferred from Zhang et al., 2019) the RCT-vs-cohort divergence is the artifact-check: the cohort HR is not evidence the deficient would benefit from repletion on mortality; that arm stays untested here.
The lower (repletion) arm — now demonstrated for one nutrient x outcome
The lower arm — does correcting an actual deficiency move an outcome? — was, until Martineau, asserted as a live question the held trials barely touched. Martineau 2017 (gold IPD meta-analysis, 25 RCTs, 10,933 participants, GRADE high) supplies the first held lower-arm demonstration — and it is exactly the effect-modification-by-baseline-status the concept predicts. Vitamin D reduced acute respiratory tract infection (ARI) overall («adjusted odds ratio 0.88, 95% confidence interval 0.81 to 0.96» (Martineau et al., 2017)), and the benefit scaled with baseline deficiency — the signature of a status-dependent curve:
| Baseline 25(OH)D | Adjusted OR (ARI) | NNT | Arm |
|---|---|---|---|
| <25 nmol/L, daily/weekly dosing | 0.30 (0.17-0.53) | 4 | deficient — steep lower arm |
| >=25 nmol/L, daily/weekly dosing | 0.75 (0.60-0.95) | 15 | replete-ish — shallower |
| <25 vs >=25, unstratified | 0.58 vs 0.89 (P-interaction 0.01) | 8 vs — | the interaction itself |
Rows 1-2 are the daily/weekly-stratified table 3 (P-interaction 0.006); row 3 is the unstratified subgroup (P-interaction 0.01) (Martineau et al., 2017).
- The source states the mechanism as the repletion 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)
- This is the LEGITIMATE effect-modification case (route (b)), not the false-positive kind. The modifier was «specified a priori and measured at baseline, P values for interaction remained significant after adjustment … and subgroup effects were consistent when analysed as proportions and event rates» (Martineau et al., 2017) — it fulfils the Sun/Guyatt subgroup-credibility criteria and carries a mechanism, so it is not a data-dredged interaction. It still rests on few trials per deficient cell (234 participants, 6 studies), and the authors themselves flag it: «We therefore suggest caution when interpreting the results in table 3.» (Martineau et al., 2017)
The cross-outcome caveat — this is NOT one measured dose-response curve. Martineau’s outcome is ARI; VITAL’s nulls were fractures / cancer / CVD. So the composed picture is vitamin D helps the deficient on ARI; it does not help the replete on fractures/cancer/CVD — a status-dependent effect-modification pattern read across different outcomes, not a single nutrient-and-outcome curve traced from its deficient edge to its plateau. The concept (benefit concentrates in the deficient) is instantiated; a clean same-outcome dose-response for any one endpoint is still not held (LeBoff et al., 2022; Manson, Cook, Lee, Christen, Bassuk, Mora, Gibson, Gordon, et al., 2019; inferred from Martineau et al., 2017).
Independence guard — this is F/A composition, NOT type-E. Martineau (ARI) and VITAL (fractures/cancer/ CVD) do not independently confirm the same claim — they answer different outcome questions, and jointly compose the arc (deficient benefits / replete does not). The value is the structure they imply together, not two witnesses of one effect (Manson, Cook, Lee, Christen, Bassuk, Mora, Gibson, Gordon, et al., 2019; inferred from Martineau et al., 2017).
- VITAL’s frankly-deficient tail (<12 ng/mL, only 2.4% of the trial) was underpowered, and the authors note a trial in people «well below the 20 ng per milliliter» might show stronger effects (LeBoff et al., 2022), (Manson, Cook, Lee, Christen, Bassuk, Mora, Gibson, Gordon, et al., 2019). So repletion on VITAL’s own endpoints (fractures/cancer/CVD) is left open — Martineau answers it only for ARI.
- Repletion maps onto route (a) of Baseline Risk and the Relative-Absolute Split: deficiency status is a baseline-risk / support-factor variable, so the absolute benefit of repletion scales with deficiency severity even at a constant relative effect (Martineau’s NNT falls from 20 overall to 4 in the deficient) — and the safe route needs no subgroup-interaction claim, though here a credible route-(b) interaction is also present (inferred from Martineau et al., 2017).
Second instance — iron -> cognition, anaemia-stratified (Falkingham 2010)
The lower arm now has a second, independent worked instance in a different nutrient on a different outcome: iron supplementation and cognition, subgrouped by baseline iron status (Falkingham 2010, gold meta-analysis, 14 RCTs, children aged 6+/adolescents/women; SMD, random effects). The signature is the same — benefit concentrates in the deficient:
| Cognitive outcome | Overall | Anaemic at baseline | Non-anaemic / replete | Arm |
|---|---|---|---|---|
| Intelligence (IQ) | null: SMD 0.10 (-0.14 to 0.33), I2 81%, n=2365 | SMD 0.54 (0.26 to 0.81), = +2.5 IQ points (1.24-3.76), I2 0%, n=209 | null, well-powered (>1500) | deficient benefits; replete does not |
| Attention/concentration | SMD 0.59 (0.29 to 0.90), I2 0%, n=179 | benefit | benefit (but n=33, mixed category) | reported status-independent — see refinement |
| Memory / psychomotor / scholastic | null | null | null | no arm moves |
Provenance: IQ overall + anaemic subgroup (Falkingham et al., 2010); the +2.5-point back-transform and the headline «improved intelligence quotient (IQ) by 2.5 points (95% CI 1.24 to 3.76), but had no effect on non-anaemic participants» (Falkingham et al., 2010); the replete null being well-powered («adequately powered to detect an effect, and … the lack of effect observed in these iron replete samples is likely to be reliable») (Falkingham et al., 2010). The authors name the repletion mechanism directly: «positive effects of iron supplementation were clearest in those with deficiencies which were corrected by the intervention» (Falkingham et al., 2010).
Absolute + relative form (the IQ arm). Relative/standardised: SMD 0.54 in the anaemic. Absolute: +2.5 IQ points vs placebo in the anaemic, versus zero in the well-powered replete subgroup — a decision-relevant magnitude only for the deficiency-corrected stratum.
Subgroup caveat (route (b), symmetric standards). Baseline-status modification is route (b) — the false-positive generator. It is credible here because it was a-priori (status was the planned subgrouping variable), carries the repletion mechanism, and the anaemic IQ subgroup is homogeneous (I2 0%) with the replete null well-powered. But it is weaker than Martineau on trial quality: the included RCTs are «generally small, short and methodologically weak», the funnel plot «suggested modest publication bias», and the anaemic IQ cell rests on n=209 (3 studies) (Falkingham et al., 2010). Hold the iron instance at lower certainty than the vitamin-D one, which was GRADE-high IPD.
The refinement — the modification is OUTCOME-specific, not nutrient-universal
Falkingham does not simply replicate the pattern; it bounds it. Within iron, the status-modification appears for IQ but not for attention/concentration, which improved «irrespective of baseline iron status» (SMD 0.59 in both subgroups) (Falkingham et al., 2010). Two readings, and the evidence does not fully separate them (inferred from Falkingham et al., 2010):
- Genuine outcome-specificity — some cognitive functions respond to iron only when a deficit is corrected (IQ), others may respond more broadly (attention). If so, “benefit concentrates in the deficient” is a per-outcome claim, not a blanket property of the nutrient.
- A power artifact on the attention side — the attention non-anaemic arm was only 33 people in the mixed «iron deficient and/or replete» category, far too thin to exclude a gradient, whereas the IQ replete arm (>1500) was powered to find a null. On this reading attention is simply measured too weakly on the non-deficient side to decide.
Either way the decision-relevant finding is unchanged: for IQ — where the replete arm is well-powered — correcting the deficiency helps and adding iron to the replete does not (scholastic, also well-powered, was a null in both arms, so it is a powered double-null, not a deficiency benefit). The refinement the fabric keeps is that the deficiency-repletion gradient is claimed per outcome, and demonstrated where the non-deficient stratum is powered enough to show a null — not asserted as a universal property of every nutrient x outcome pair.
The principle now has independent cross-nutrient backing [E-independent]
Three [updated 2026-08-04: strengthened from two to three by Bougma — see the iodine instance below] independent routes now reach the same general principle — supplementation benefit is
modified by baseline nutrient status; it concentrates in the deficient and is small-to-absent in the
well-nourished:
- Vitamin D -> ARI (Martineau IPD MA, 25 RCTs): deficient OR 0.30 vs replete-ish 0.75, P-interaction 0.006. Both legs within-study.
- Iron -> IQ (Falkingham MA, 14 RCTs): anaemic +2.5 IQ points vs a well-powered replete null. Both legs within-study.
- Iodine -> child IQ (Bougma MA, 24 studies): repletion of the deficient +7.4 IQ points (range 6.9-10.2). Deficient-benefit leg only directly; replete-null leg indirect (see the iodine instance below for the leg asymmetry).
These share no trials, no research group, no lineage, run on different nutrients, different
outcomes, different populations, and none cites the others — so their agreement on the pattern
is genuine independent backing, raising confidence in the principle in a way no single instance
does [E-independent]
(Bougma et al., 2013; Falkingham et al., 2010; inferred from Martineau et al., 2017).
The convergence is uneven across the principle’s two legs — do not read three as three-for-both.
The principle has a benefit-in-the-deficient leg and a small-to-absent-in-the-replete (contrast)
leg. The deficient-benefit leg is backed by all three independent nutrients. The contrast
leg — that adding more to the replete does not help — rests on only two clean within-study
demonstrations (Martineau, Falkingham); Bougma supports it only indirectly (the Murcia sufficient-area
null), because its design holds no supplement-the-replete arm. So the honest strengthening is: leg-1
now three-independent; leg-2 still two-independent-plus-one-indirect
[inferred from @martineau2017; @falkingham2010; @bougma2013].
Scope of the E claim — strict. The independence holds only for the general principle. The
specific effects (vit-D -> ARI, iron -> IQ, iodine -> child IQ) are different claims and do NOT
corroborate each other — Falkingham is not a second witness that vitamin D cuts ARI, nor Bougma that
iron raises IQ. What is independently corroborated is the status-dependence structure all three
instantiate.
Do not read the [E-independent] mark as evidence for either point estimate
(Falkingham et al., 2010; inferred from Martineau et al., 2017).
Third instance — iodine -> child mental development (Bougma 2013)
The lower arm now has a third, independent worked instance, and it is the largest-magnitude repletion effect the wiki holds: iodine and child mental development, in iodine-deficient populations (Bougma 2013, gold meta-analysis, 24 studies of children aged <=5 years; SMD d, random effects). Correcting maternal/child iodine deficiency raised child IQ substantially:
| Study design | Effect size (SMD d) | Absolute (IQ points, SD=15) |
|---|---|---|
| Intervention RCT (2 trials) | 0.68 | ~10.2 |
| Intervention non-RCT (8) | 0.46 | ~6.9 |
| Cohort, maternal iodine status (9) | 0.52 | 7.8 |
| Cohort, infant iodine status (4) | 0.54 | 8.1 |
| Best estimate (pooled intervention, 16 effects/10 studies) | 0.49 | 7.4 |
The headline: «the best estimate to date of the effect size of iodine supplementation on mental development in children 5 years old and under is 0.49, which translates into 7.4 IQ points lost due to iodine deficiency» (Bougma et al., 2013); the four-design range is «6.9 to 10.2 IQ points lower in iodine deficient children compared with iodine replete children» (Bougma et al., 2013).
Absolute + relative form. Relative/standardised: d = 0.49 (intervention). Absolute: +7.4 IQ points (range 6.9-10.2) for the deficiency-corrected stratum — nearly 3x iron’s +2.5-point anaemic-IQ effect. This tracks mechanism: iodine -> thyroid hormone (T4/T3) -> CNS cell migration/differentiation/myelination, and severe deficiency causes cretinism, so the deeper the deficiency the larger the repletion return (inferred from Bougma et al., 2013). This is the mechanistically cleanest lower arm the trio expected — severe deficiency -> severe, measurable impairment — though evidentially it is the trio’s weakest design (2 RCTs, heavy confounding; see Caveats). Big, clear signal on a shaky evidence base: the direction is secure, the exact size is not.
Synthesis check 1 — the principle now has THREE independent instances, with a stated asymmetry
Iodine shares no trials, no research group, no lineage with Martineau (vit-D) or Falkingham (iron),
and runs on a different nutrient, outcome (child neurodevelopment), and population — so it is a
genuine third independent route to the general principle repletion benefit concentrates in the
deficient. The [E-independent] mark below is strengthened from two to three nutrients.
But the structure of Bougma’s evidence differs, and the strict scope must be kept. Martineau and Falkingham each demonstrated both legs within-study (a deficient-vs-replete interaction: benefit in the deficient, null in the replete). Bougma directly demonstrates only the deficient-benefit leg — its intervention studies replete iodine-deficient populations; it holds no supplement-the-already-replete RCT arm. The replete-null leg for iodine is supported only indirectly: the largest iodine-sufficient-area cohort found «children of iodine-sufficient and deficient mothers in the first trimester of pregnancy, based on UIE, had similar mental development scores (Mean 100.38 vs. 99.10; … d = -0.09)» (Bougma et al., 2013) — a null on iodine-intake stratification in a replete population — but other sufficient-area cohorts stratifying on maternal thyroid function (a functional-deficiency contrast) did find effects. So iodine’s contribution is: strong, large, independent confirmation of the deficient-benefit leg; only weak, indirect support for the replete-null leg (inferred from Bougma et al., 2013).
Synthesis check 2 — iodine does NOT show Falkingham’s benefit irrespective of status; it adds a TIMING window
Falkingham’s iron->attention was the outcome-specific exception (benefit irrespective of baseline status). Does iodine->development replicate that exception, or fit the clean deficient-only gradient? Neither, cleanly — because Bougma cannot test adding iodine to the replete (no replete arm). What it does add is a different conditionality axis: developmental timing. The repletion benefit is gated by the fetal-brain window — «comparisons between early pregnancy and control groups produced a large average effect size of 0.51 while comparisons between late pregnancy and control groups produced a small average effect size of 0.17» (Bougma et al., 2013), with a direct early-vs-late comparison of d = 0.82 favouring early. So the benefit is not irrespective — it is conditional on the intervention reaching the deficient physiology during its window of susceptibility (inferred from Bougma et al., 2013).
What this does to the fabric’s existing refinement — keep TWO distinct axes apart. The three instances do not all decorate one support factor — they populate two different axes, and collapsing them would be a category error (Bougma et al., 2013; inferred from Falkingham et al., 2010; Martineau et al., 2017):
- Axis 1 — WHERE the gradient appears (outcome-specificity). Falkingham: the deficiency-modification shows on IQ but the attention benefit looked status-independent. This is about which outcome carries the gradient, not a factor that must co-occur — the fabric’s existing “gradient is claimed per outcome” refinement.
- Axis 2 — WHETHER a given benefit fires (gating support factor). Martineau: the ARI benefit needs daily/weekly dosing (bolus was null even in the deficient); Bougma: the child-IQ benefit needs the early-pregnancy developmental window (late supplementation was null). Same-outcome, same-deficient benefit, gated by a co-present exposure condition.
This is not a bold new emergent claim — it is the held transportability principle («a mechanism
fires only when its whole set of conditions is present») instantiated for repletion, a type-F
refinement of it, not a standalone type-A. The decision-relevant upshot: works only in the deficient
is necessary, not sufficient — repleting the right stratum still fails if the outcome does not carry
the gradient (axis 1) or the exposure misses its gating condition (axis 2), and which condition binds
is instance-specific, not transferable across nutrients
[inferred from @bougma2013; @falkingham2010; @martineau2017].
Caveats (symmetric standards) — hold iodine below the vitamin-D certainty
Bougma’s design bar is the weakest of the trio: only 2 RCTs (one, Pretell, with unclear randomization and post-hoc child selection), median sample 50/group (underpowered), and the source’s own list of «weak study designs, the omission of important confounders, small sample sizes, the lack of cluster analyses». Confounding is uncontrolled in the non-RCT/observational studies — «Failure to include confounds in the statistical analysis may serve to inflate effect sizes» (Bougma et al., 2013) — and a maternal-behaviour channel may partly mediate/confound the child-IQ effect: «we cannot determine how much of the difference is explained by the child’s iodine sufficiency and how much is explained by the mother’s provision of positive stimulation rather than the child’s iodine sufficiency» (Bougma et al., 2013). The magnitude is large and design-robust (consistent across four designs); the precise size and the pure-iodine attribution carry real uncertainty. The direction and the deficient-benefit leg are secure; hold the point estimate loosely.
A DISTINCT flavor of repletion — periconceptional folic acid and NTDs (window-gated, stratum-DEFINED, not baseline-status-modification)
The trio above (vit-D/iron/iodine) are one structural kind: route-(b) baseline-STATUS modification — the supplement’s benefit concentrates in the measured-deficient and fades in the measured-replete, within one population. Periconceptional folic acid for neural-tube-defect (NTD) prevention is a different kind of repletion, and forcing it into the trio’s pattern would be a category error (inferred from De-Regil et al., 2015). It is the clearest supplement-benefit the wiki holds on a HARD patient-important outcome, and it earns a clean statement:
- Magnitude (De-Regil 2015 SR/MA, GRADE high). Periconceptional folate vs no folate cut NTDs to RR 0.31 (95% CI 0.17 to 0.58); 5 RCTs; 6708 births; recurrence RR 0.34 (0.18-0.64); absolute 35 -> 11 per 1000 (6 to 20) (De-Regil et al., 2015). Authors’ conclusion: «Folic acid, alone or in combination with vitamins and minerals, prevents NTDs, but does not have a clear effect on other birth defects» [EXTRACTED chunk 01] — the benefit is specific to NTDs. No harm signal (miscarriage RR 1.10, 0.94-1.28) [EXTRACTED chunk 01].
Why it is a distinct flavor, in three appraisal moves:
- The stratum is defined by LIFE-STAGE, not by a lab value. The “who” is women capable of / planning pregnancy, periconceptionally — a reproductive-window stratum, not a biochemically-deficient subgroup. The whole stratum benefits, and the effect is not modified by dose above ~0.36 mg or by folate-alone vs -combined (a plateau, no route-(a) scaling by measured shortfall within the stratum) (De-Regil et al., 2015). Contrast the trio, where absolute benefit scales with measured deficiency severity.
- The protective target sits ~9x ABOVE the deficiency threshold. Deficiency is RBC folate <100 nmol/L, but «the optimal folate level associated with lowest risk of NTDs is 906 nmol/L or above» (De-Regil et al., 2015). So this is not correcting a frank deficiency — it is reaching a supra-adequate, outcome-specific target. A woman with “adequate” folate by the anaemia-prevention standard can still be NTD-insufficient. This is why the effect is population-wide for the stratum rather than concentrated in the lab-deficient tail.
- It is window-gated — Axis 2, the same gating structure as iodine. The neural tube closes within ~28 days of conception (De-Regil et al., 2015), so the exposure must be present periconceptionally (before most women know they are pregnant); trials that started supplementation after the first trimester were out of scope. This is the fabric’s Axis 2 (WHETHER a benefit fires — gating support factor) instantiated again, and the closest structural parallel to Bougma’s fetal-brain window (early-pregnancy iodine d 0.51 vs late 0.17) — repletion after the window largely fails (Bougma et al., 2013; inferred from De-Regil et al., 2015).
Higher-risk sub-stratum (route (a), baseline risk). For women with a prior NTD pregnancy, diabetes, or on anticonvulsants, the recommended dose is 5 mg/day (vs 0.4 mg general) (De-Regil et al., 2015) — a higher-baseline-risk stratum getting a larger dose, route (a) on Baseline Risk and the Relative-Absolute Split.
A related repletion-TERRITORY case whose outcome-transmission is NOT yet held — B12 in vegetarians. Vitamin B12 in a vegetarian/vegan is a genuine lower-arm case by provenance of the deficit: B12 has essentially no plant source, so the diet creates a real shortfall (11-90% biochemically depleted/deficient by accurate markers, cutoff-driven) -> Vitamin B12 Status in Vegetarian and Vegan Diets. But it is structurally weaker than the trio: the held source (Pawlak 2013) is a descriptive prevalence review with no patient-important-outcome effect and no omnivore comparator, and in the one study that looked, two-thirds were biochemically deficient with zero clinical symptoms — so the repletion benefit on a hard endpoint is a gap, not a demonstration, and the biochemical rate is a surrogate -> Surrogate Outcomes. It is repletion-territory (correcting a real dietary deficit is cheap insurance), not a quantified lower-arm effect like vit-D/iron/iodine (inferred from Pawlak et al., 2013).
Kept OUT of the [E-independent] count, deliberately. Folate/NTD is not a fourth
baseline-status-modification instance — it does not demonstrate a measured-deficient-vs-measured-replete
interaction on one outcome. It is a different structural flavor (window-gated, stratum-defined, supra-adequate
target), so the three-nutrient E-claim on the baseline-status-modification principle is unchanged; folate
adds breadth to what repletion can look like, not a fourth witness of the trio’s specific pattern
(Bougma et al., 2013; inferred from De-Regil et al., 2015; Falkingham et al., 2010; Martineau et al., 2017).
Decision-change. For a woman planning or capable of pregnancy, 400 ug/day periconceptional folic acid (5 mg if higher-risk) is a real lever on a hard outcome — it flips supplementation for this stratum from bottom-of-hierarchy (where the general-population disease-prevention null puts it, -> Vitamin and Mineral Supplements for Disease Prevention) to a genuine big rock. This is the standing counter-example to supplements are a distraction: for this stratum, on this outcome, a supplement clearly works. The population-scale analogue is folic-acid fortification of staple foods (delivery vehicle, not a different exposure) (De-Regil et al., 2015).
Guard — this is the REAL dose-response U, not the artifact
Do not conflate this status-dependent curve with The U-Shaped Association Artifact. There, a U in observational data is often manufactured by reverse causation, confounding by frailty, or unequal reporting precision — an artifact to be checked and usually discounted. Here the U is causal and mechanistic, and its upper arm is demonstrated in RCTs (Bjelakovic low-bias pool; CARET, stopped early for harm) — the harm of too much is real and randomized, not a reverse-causation artifact. Both arms are now RCT-demonstrated, though on different outcomes: the lower arm — benefit of correcting too little — is demonstrated for vitamin D on ARI (Martineau IPD MA, deficient OR 0.30), and the upper arm — harm of too much — for antioxidants/beta-carotene on mortality/cancer. So the mechanism’s full U is instantiated across the evidence, not proved as a single traced curve on one endpoint. Same shape, opposite epistemic status from the artifact — a curve to dose along, not a signal to explain away (Bjelakovic et al., 2007; inferred from Martineau et al., 2017; Omenn et al., 1996).
Decision consequence
- Generally-nourished person. Supplementation is enhancement — the plateau-to-harm arm — so it ranks at or near the bottom of the intervention hierarchy, and megadose pushes toward the harm arm, not a bigger benefit -> Layer 1 - Ranking Interventions for a Stratum.
- Deficient stratum. Repletion is a different question with a demonstrated benefit on patient-important outcomes across three nutrients now: in the vitamin-D deficient (<25 nmol/L), daily/weekly D cut ARI risk with NNT=4 (vs NNT=33 overall, NNT=20 daily/weekly) (Martineau et al., 2017); in the anaemic, iron raised IQ by +2.5 points (null in the replete) (Falkingham et al., 2010); and repleting iodine-deficient mothers/children raised child IQ by +7.4 points (range 6.9-10.2) (Bougma et al., 2013) — each a decision-change that flips supplementation for that deficient stratum from bottom-of-hierarchy to a real lever. Each also carries a support-factor condition beyond deficiency itself: the exposure must be daily/weekly, not bolus (vitamin D: bolus null even in the deficient, OR 0.82, 0.51-1.33) (Martineau et al., 2017), it is outcome-specific (iron: IQ yes, some functions irrespective), and it is window-gated (iodine: early pregnancy d 0.51 vs late 0.17 — repletion after the fetal-brain window largely fails) (Bougma et al., 2013). It still requires first establishing the deficiency (a prescriber act needing this person’s labs, out of the wiki’s scope). The right move stays test, then replete the confirmed shortfall with the right schedule, in the right window, not supplement broadly to prevent disease.
- The demarcation, stated once. An enhancement-null does not read as the nutrient does not matter — it reads as more, in the replete, does not help (and past need can harm) The nutrient can be essential and the supplement still worthless for the already-replete; the two claims live on different arms and must not be collapsed.
Gaps
No held deficiency-repletion effect size on a patient-important outcomeCASHED 2026-08-04 by Martineau (vitamin D x ARI, deficient OR 0.30, NNT=4). The lower arm is now quantified for one nutrient-and-outcome pair, GRADE high. What remains open: a same-outcome dose-response for any one endpoint (Martineau’s ARI benefit and VITAL’s fracture/cancer/CVD nulls are different outcomes), and repletion effect sizes for other nutrients. Iron -> cognition CASHED 2026-08-04 by Falkingham (anaemic +2.5 IQ points, null in the well-powered replete) — a second nutrient x outcome, and the independent backing that lifts the principle to[E-independent](above). Iodine -> child mental development CASHED 2026-08-04 by Bougma (repletion of the deficient +7.4 IQ points, range 6.9-10.2) — the trio’s third independent nutrient, lifting the[E-independent]principle to three nutrients (see the iodine instance above). Still open: a same-outcome dose-response traced from deficient edge to plateau for any one endpoint (all three instances read the deficient-benefit leg on outcomes their replete-arm evidence does not directly test), and repletion effect sizes for further nutrients.Gfurther cashed. Fractures now have BOTH arms named in one source (Kahwati 2018): a direct community-dwelling replete-null and a reported high-risk/deficient benefit (via Cochrane) — the closest the fabric holds to a same-outcome two-arm read, though the deficient arm is second-hand (Kahwati’s citation of Avenell 2014), not a within-review deficient-vs-replete interaction.- Threshold locations are per-nutrient and per-outcome and mostly unheld — where the deficiency edge and the toxicity edge sit differs by nutrient (25(OH)D in ng/mL is not beta-carotene in mg), so the curve is a shape the fabric holds, not a set of numbers. Martineau does locate one edge: the ARI benefit concentrates below 25 nmol/L 25(OH)D, with no significant interaction across the 25/50/75 nmol/L cuts above it (Martineau et al., 2017).
(inferred from Martineau et al., 2017)