Nucleus of the cardiometabolic-exposures cluster. WHO 2012, all recommendations strong.
The word conditional does not appear in the document.
The recommendations
“WHO recommends a reduction in sodium intake to reduce blood pressure and risk of cardiovascular disease, stroke and coronary heart disease in adults (strong recommendation). WHO recommends a reduction to <2 g/day sodium (5 g/day salt) in adults (strong recommendation).” (World Health Organization, 2012)
For children: reduce to control blood pressure (strong), with the adult 2 g/day figure “adjusted downward based on the energy requirements of children relative to those of adults” — a scaling rule, with no outcome evidence and no certainty rating attached to it.
No certainty rating is attached to any recommendation line — certainty lives only in the per-outcome tables and an Annex 8 bullet list.
The narrow, checked contrast with a later WHO guideline. WHO’s 2023 SFA/TFA guideline uses the
same bare “(strong recommendation)” form on its recommendation lines — there is no
(strong recommendation, moderate-certainty evidence) pairing in it, zero occurrences. What SFA 2023
does add is an overall certainty stated per recommendation in its Rationale (“The overall
certainty in the evidence for recommendation 1 was moderate, and for recommendation 2 was very low”);
this 2012 guideline issues no overall certainty anywhere.
(World Health Organization, 2023)
The evidence split — this is the whole story
| Outcome | Effect (95% CI) | Design | Certainty |
|---|---|---|---|
| Resting SBP | 3.39 mmHg lower (2.46 to 4.31) | 36 RCTs, 49 comparisons, n=6736 | HIGH |
| Resting DBP | 1.54 mmHg lower (0.98 to 2.11) | 36 RCTs | HIGH |
| Ambulatory SBP | 5.51 mmHg lower (3.16 to 7.87) | 6 studies | HIGH |
| All-cause mortality | RR 1.06 (0.94 to 1.20) — inconclusive | 7 cohorts, n=21 515 | VERY LOW |
| Stroke | RR 1.24 (1.08 to 1.43) | 10 cohorts, n=72 878 | VERY LOW |
| Cardiovascular disease | RR 1.12 (0.93 to 1.34) — inconclusive | 9 cohorts, n=46 483 | VERY LOW |
| Coronary heart disease | RR 1.04 (0.86 to 1.24) — inconclusive | 6 cohorts, n=37 343 | VERY LOW |
| Cardiovascular disease | RR 0.84 (0.57 to 1.23) — inconclusive | 2 RCTs, n=720 | MODERATE |
(World Health Organization, 2012)
The surrogate is rated HIGH; most — not all — hard cohort outcomes are rated VERY LOW. The exception matters and the table above omitted it: fatal stroke, RR 1.63 (1.27-2.10), is rated LOW, and fatal coronary heart disease, RR 1.32 (1.13-1.53), is the other statistically significant cohort result. Both were left out while the four null rows were listed — an omission running in the direction of the surrogate high, hard outcomes very low reading, which is exactly the direction such omissions tend to run. WHO’s own summary states the split: “Higher sodium intake was associated with higher risk of incident stroke, fatal stroke and fatal coronary heart disease. There was no association between sodium intake and all-cause mortality, incident cardiovascular disease and non-fatal coronary heart disease.” (World Health Organization, 2012)
But note the last row: a 2-RCT meta-analysis of cardiovascular disease is rated moderate — direct hard-outcome RCT evidence, above the cohort floor, though its interval is wide and null. WHO summarizes the split itself:
“The evidence regarding the relationship between sodium intake and blood pressure was of high quality, whereas the evidence regarding sodium intake and all-cause mortality, cardiovascular disease, stroke and coronary heart disease was of lower quality. Therefore, these recommendations should be reviewed when more evidence… becomes available.” (World Health Organization, 2012)
The published RCT meta-analysis behind the effect [2026-07-31, He 2013]
WHO 2012 is a recommendation summary whose four underpinning systematic reviews are unpublished in-house documents (this page’s Limits). He, Li & MacGregor 2013 (BMJ; a Cochrane SR + meta-analysis) is a published, verifiable, gold-tier RCT meta-analysis of the same sodium->BP effect, so the effect leg no longer rests on a recommendation summary alone. 34 RCTs, 3230 participants, modest reduction (>=4 weeks), pooled salt reduction achieved -75 mmol/24h (~4.4 g/day salt, close to the public-health target).
| Stratum | SBP (95% CI), I2 | DBP (95% CI), I2 | trials / n |
|---|---|---|---|
| All | -4.18 mmHg (-5.18 to -3.18), I2=75% | -2.06 (-2.67 to -1.45), I2=68% | 34 / 3230 |
| Hypertensive | -5.39 mmHg (-6.62 to -4.15), I2=61% | -2.82 (-3.54 to -2.11), I2=52% | 22 / 990 |
| Normotensive | -2.42 mmHg (-3.56 to -1.29), I2=66% | -1.00 (-1.85 to -0.15), I2=66% | 12 / 2240 |
«A modest reduction in salt intake for four or more weeks causes significant and, from a population viewpoint, important falls in blood pressure in both hypertensive and normotensive individuals, irrespective of sex and ethnic group. Salt reduction is associated with a small physiological increase in plasma renin activity, aldosterone, and noradrenaline and no significant change in lipid concentrations.» (He et al., 2013)
This is NOT independent backing of WHO — the check is the reference trail, again. He/MacGregor
studies sit inside WHO’s cited evidence base (established in the ESC section below), and He 2013’s own
included trials (MacGregor 1982/1989, Cappuccio 1997, He 2009, Swift 2005 …) are the same primary work.
So this is the primary systematic evidence in the same lineage, not a second witness — type F /
shared-evidence, no [E-independent]. The magnitudes even differ from WHO’s pooled figures (He’s SBP
-4.18 vs WHO’s 36-RCT/49-comparison -3.39), which confirms they are different pooled sets, not the
identical analysis: He 2013 is a verifiable SR the page can cite for the effect, not the literal
document WHO distilled.
(inferred from He et al., 2013; World Health Organization, 2012)
What it does and does not move. It raises the warrant on the BP-effect leg (a published gold SR
now stands behind the surrogate). It does not touch this page’s hard-outcome certainty split —
He’s hard-outcome claims (a re-pooled CV-events -20%, the Strazzullo cohort meta) are advocacy
re-analyses (competing interest below), not graded outcome evidence — so confidence: stays medium
and the surrogate high, hard outcomes very low reading is unchanged.
Competing interest, disclosed and material to the interpretive claims. FH is a member and GM the chairman of CASH / WASH (salt-reduction advocacy charities). The RCT-pooled BP effect is hard data and taken at face value; the interpretive extensions — the «3 g/day … long term target», the J-curve dismissal, the CV re-analyses — are advocacy-shaped and are caveated at point of use (symmetric standards: a favourable-direction source gets the same scrutiny as an unfavourable one).
The dose-response is significant, monotone, and estimated — where WHO’s was declared absent
WHO read its by-intake-band gradient as null (P=0.17) and concluded the effect is «independent of baseline sodium intake». He asks a different question — the magnitude of reduction achieved vs the BP fall — and finds a significant gradient: a 100 mmol/24h reduction (6 g/day salt) -> SBP -5.8 mmHg (2.5 to 9.2, P=0.001) adjusted for age, ethnic group and BP status (68% of between-study variance); steeper in hypertensives (-10.8 mmHg per 100 mmol) than normotensives (-4.3). These are not contradictory — WHO’s is baseline-intake-as-modifier, He’s is dose-of-reduction-vs-response — and both point the same decision way: no plateau in the studied range, larger reduction buys more.
«The observed significant association between the reduction in 24 hour urinary sodium and the fall in systolic blood pressure, indicates that larger reductions in salt intake will lead to larger falls in systolic blood pressure. The current recommendations to reduce salt intake from 9-12 to 5-6 g/day will have a major effect on blood pressure, but a further reduction to 3 g/day will have a greater effect and should become the long term target for population salt intake.» (He et al., 2013)
Width caveat (author’s own): the meta-regression dose-response «should be viewed as exploratory and could be prone to confounding» — a between-study (ecological) estimate; He flags that individual- participant-data meta-analysis would be needed to pin it. So larger reduction -> larger fall is supported as direction/shape, not as a precise per-gram coefficient. This is the domain’s every-reduction-pays default (telos dose-response section) instantiated on a recovery-biomarker exposure. The 3 g/day target is a policy extrapolation beyond the trial range and carries the advocacy caveat above. (He et al., 2013) Challenged since 2026-10-22 (Graudal 2020): trial range means the long-term range; Graudal’s trials (3-1100 days) reach a mean low arm of 65 mmol/day (SD 40). At that deeper contrast white normotensives fall ~0.8 mmHg per 100 mmol against He’s ~3.2 (pooled 2.42 mmHg / 75 mmol, a ratio; He’s own normotensive meta-regression slope is 4.3, 95% CI 0.1-8.5, adjusted for age and ethnicity), and Graudal, citing Brunner 1972, describes renin and aldosterone rising exponentially below ~100 mmol/day (no CI or fitted breakpoint). That is a further hint of concavity at the low end (weakly echoed by WHO’s non-significant by-baseline-intake gradient, absolute mmHg not per mmol; against it, Huang finds baseline intake not a modifier, P trend 0.20), confounded by ethnicity, baseline BP and duration (section below). No plateau shown still holds; proportionally more below ~100 mmol/day is not established. (inferred from Graudal et al., 2020)
Huang 2020 refines the dose-response with a duration axis — and it is the SAME lineage, not a witness [2026-07-31, Huang 2020]
Huang 2020 (BMJ; TRUE Consortium) is the He/MacGregor lineage’s larger successor — 133 RCTs,
12 197 participants on 24h-urinary-sodium, achieving -130 mmol -> SBP -4.26 mmHg (-4.89 to -3.62),
DBP -2.07. It is NOT independent backing: FJH + GAM co-author both papers, Huang’s ref 21 is He
2013, and Huang’s search strategy is developed from He’s. So this is type F / shared-evidence — a
refinement in the same lineage, no [E-independent] (same CASH/WASH advocacy interest; competing
interest below). It refines He on two axes He could not resolve.
1. The naive dose-response is null; it is real only after adjusting for duration. He’s steeper between-study slope was the estimate he himself flagged «exploratory… prone to confounding» — Huang names the confounder. The raw quintile trend and the univariable meta-regression both found NO association of SBP fall with sodium-reduction magnitude (P trend=0.09; univariable coeff 0.007, 95% CI -0.002 to 0.016, P=0.106). The relation appears only after multivariable adjustment:
«Likewise, the univariable meta-regression analyses including all studies (table 1) showed no association of blood pressure effect with magnitude of sodium reduction. However, the magnitude of the change in 24 hour urinary sodium excretion was positively associated with the change in blood pressure after adjusting for intervention duration, mean age, percentage of female sex, percentage of white ethnicity, and baseline blood» (Huang et al., 2020)
adjusted slope: each 50 mmol -> SBP -1.10 mmHg (0.66-1.54), DBP -0.33 (0.04-0.63) — i.e. ~2.2/100 mmol. This is the gate-6 caution made concrete on this exposure: the monotone dose-response here is recovered by the model specification, not visible in the raw contrast — but the direction survives (adjustment reveals a positive slope, and effects persist down to baseline SBP <120, so every-reduction-pays holds and no plateau is shown).
Parameter table (op-weave 2a) — He-vs-Huang dose-response slope, matched:
| Parameter | He 2013 | Huang 2020 | Same quantity? |
|---|---|---|---|
| SBP fall per 100 mmol, modest-reduction / >=2-wk trials | -5.8 mmHg (2.5-9.2), between-study meta-reg | ~-5.8, >14-day + <=100 mmol subgroup | YES — Huang states its subgroup «similar to» He’s, ref 21 |
| SBP fall per 100 mmol, ALL trials pooled | not computed (He restricted to >=4wk/modest) | -2.2 mmHg (multivariable, 133 trials) | NO — different trial mix (all durations vs >=4wk only) |
| Overall pooled SBP effect | -4.18 at -75 mmol achieved (34 trials) | -4.26 at -130 mmol achieved (133 trials) | NO — similar mmHg at nearly double the dose = the short-trial dilution below |
So the two are not in conflict: He estimated the long-trial / modest-reduction slope, Huang the across-durations slope, and Huang reproduces He’s 5.8 exactly when it restricts to He’s inclusion window. The apparent discrepancy (He 5.8 vs Huang 2.2 per 100 mmol) is the type-of-studies-included sensitivity, which is the next point.
2. The duration effect — short trials underestimate (a bias-direction finding). The overall duration relation is NULL (P trend=0.87, meta-regression NS) — it is not a longer-is-more gradient. What is real is a very-short-trial underestimate: each 50 mmol reduction bought -1.05 mmHg (0.40-1.70) SBP in <=14-day trials vs -2.13 mmHg (0.85-3.40) in longer — an ~doubling. Because 57% (77/136) of data points were <=14 days, the pooled estimate is biased downward.
«Our review identifies an approximate doubling of the effect of sodium reduction on blood pressure in studies of longer than two weeks’ duration versus shorter studies, indicating that the full effects of dietary sodium reduction require several weeks to become apparent. Very short term studies of sodium reduction are not a sound basis for drawing conclusions about the effects of sodium reduction on blood pressure and are not helpful in formulating policy recommendations for public health.» (Huang et al., 2020)
Huang reads this as a real biological lag, not a measurement artifact: short trials use very restricted diets -> sudden large reductions -> transient renin-angiotensin / sympathetic activation and adverse metabolic effects that do not persist chronically; the full effect needs several weeks (DASH-Sodium showed a larger fall at week 4 than earlier). (The activation-to-shortfall link is this page’s reading; Huang states the two side by side — see the opposing-Cochrane section.) Decision consequence for this page: the pooled surrogate effect it holds is a lower bound on the sustained effect — the chronic BP benefit of a maintained reduction is larger than a trial-duration-mixed meta-analysis shows. (inferred from Huang et al., 2020) Qualified since 2026-10-22: Huang’s own six multi-timepoint trials and its multivariable meta-regression are null on duration; the doubling is a subsidiary split with overlapping intervals. Lower bound — and the bias-direction finding / biased downward above — is therefore a possibility, not a finding -> the opposing-Cochrane section below. (inferred from Huang et al., 2020)
3. The modifier is continuous baseline SBP, not the hypertensive/normotensive dichotomy. Huang finds the BP-status dichotomy non-significant (normo/hyper P trend=0.08) while the continuous baseline-SBP gradient is significant (P trend=0.01), with effect present even at SBP <120 — and argues the dichotomy is «weak because the definition of hypertension is arbitrary». This refines the Response-heterogeneity section above (route-b effect modification): the real modifier this page should carry is a graded baseline-SBP interaction, of which the ~3:1 hypertensive:normotensive ratio is a coarsened summary. Age (older) and non-white ethnicity are independent modifiers; baseline sodium intake is NOT (P trend=0.20) — corroborating WHO’s «independent of baseline sodium intake». (Huang et al., 2020)
Competing interest (same as He’s lineage). FJH member / GAM chairman of CASH + WASH; NRCC + NC WASH; BN received salt substitute from manufacturers for other trials. The RCT-pooled effects are hard data; the pro-reduction interpretive framing (short-trial dismissal, the direct dismissal of PURE, the low-intake-artefact framing) is advocacy-shaped and read at face value only for the data, caveated for the interpretation — symmetric standards.
Lipids null — corroborates WHO’s high/no-effect rating (same lineage)
He finds no significant change in total cholesterol (+0.05, P=0.18), LDL (+0.05, P=0.11), HDL (-0.02,
P=0.19) or triglycerides (+0.04, P=0.22) with modest reduction — the small physiological renin/
aldosterone/noradrenaline rises being the compensations WHO also observed. This corroborates WHO’s
high-certainty, no-harmful-effect-on-lipids rating (this page’s Limits), but through the same
He/MacGregor lineage — a body line, not a sources:-padding independent backing.
(He et al., 2013)
Contested since 2026-10-22: Graudal 2020 (next section) finds total cholesterol and triglyceride
rising at a larger contrast, including in its >=7-day subgroup — and the two lipid estimates turn out to
be numerically compatible.
(inferred from Graudal et al., 2020; He et al., 2013)
The opposing Cochrane review — same hypertensive BP effect, a compensation arm the other lineage reads as transient [2026-10-22, Graudal 2020]
Graudal, Hubeck-Graudal & Jürgens 2020 (Cochrane CD004022, 4th update) is the review whose 2011/2017 editions the He/Huang lineage rebuts. It pools 195 RCTs, 12,296 people, durations 3-1100 days, at a larger and sharper contrast than He’s: 203 -> 65 mmol/day (salt 11.5 -> 3.8 g/day), with a methods-defined (no protocol) >=7-day, high arm <=250 mmol/day subgroup. Exposure is 24-hour urine or a sample of >=8 hours (looser than Huang’s 24-hour-only rule). In the authors’ words, «there is no protocol». (Graudal et al., 2020)
Not independent of He/Huang on the BP leg. The author lists are disjoint, but the trial pools
overlap heavily: MacGregor 1989, Cappuccio 1997, He 2009, Swift 2005, TOHP II and DASH-Sodium all sit in
Graudal’s lists and in He’s. Agreement on the BP effect is shared-trial agreement, no [E-independent].
The authors are also a position-holding camp (their own U-curve cohort meta-analysis, a guideline-process
critique, a 2018 Circulation letter co-written with Mente (Graudal et al., 2020)) — the mirror of the CASH/WASH interest above, read the same
way: pooled numbers at face value, interpretation caveated.
(inferred from Graudal et al., 2020; He et al., 2013)
Blood pressure: hypertensives agree across camps; white normotensives are small in every review
| Population (Graudal) | SBP MD (95% CI) | DBP MD (95% CI) | trials | GRADE |
|---|---|---|---|---|
| White normotensive | -1.14 (-1.65 to -0.63) | 0.01 (-0.37 to 0.39) | 95 (SBP) / 96 (DBP) | high |
| — >=7 d subgroup | -1.38 (-1.87 to -0.89) | -0.37 (-0.78 to 0.04) | 63 | high |
| White hypertensive | -5.71 (-6.67 to -4.74) | -2.87 (-3.41 to -2.32) | 88 | high |
| Black normo / hyper | -4.02 (-7.37 to -0.68) / -6.64 (-9.00 to -4.27) | -2.01 / -2.91 | 7 / 8 | low |
| Asian normo / hyper | -1.50 (-3.09 to 0.10) / -7.75 (-11.44 to -4.07) | -1.06 / -2.68 | 5 / 8 | moderate / low |
Hypertensive: convention held. Graudal’s -5.71 sits beside He’s -5.39 and WHO’s -4.06, and the review’s own conclusion is that in white people with elevated BP sodium reduction «may be used as a supplementary treatment for hypertension». For a hypertensive the direction of the recommendation is the same from both camps. (Graudal et al., 2020)
Normotensive — parameter table (op-weave 2a) before any comparison:
| Parameter | WHO 2012 | He 2013 | Graudal 2020 | Same quantity? |
|---|---|---|---|---|
| SBP fall, normotensive | 1.38 (0.02-2.74), 6 studies | 2.42 (1.29-3.56), 12 trials | 1.14 (0.63-1.65) white; 1.38 >=7 d | PARTLY — same outcome, different trial sets |
| Achieved contrast | ~-76 mmol overall (1.74 g/day; not reported for normotensives) | -75 mmol | ~-138 mmol (~-110 to -127 in the subgroup) | NO (WHO ~ He; Graudal ~1.8x) |
| Duration filter | >=4 weeks | >=4 weeks | any (>=7 d in subgroup) | NO |
| Ethnicity | all | all | white only (black/Asian separate) | NO |
| Baseline BP of normotensive | upper half of the distribution (Graudal’s reading) | 127/77 median (He chunk 01) | mean ~119/71 | NO |
(World Health Organization, 2012) (He et al., 2013) (Graudal et al., 2020)
Per 100 mmol of reduction that is roughly 3.2 mmHg in He’s normotensives (pooled 2.42 mmHg / 75 mmol, a ratio; He’s own normotensive meta-regression slope is 4.3, 95% CI 0.1-8.5) and 0.8 in Graudal’s . Contrast, duration, ethnicity and baseline BP all differ at once, and the held sources cannot say which carries the gap. The camps do agree on one candidate modifier: Graudal explains its gap with WHO by baseline BP, rejecting duration — «as duration has no impact on the BP effect (Table 2), a more reliable explanation for the difference between the WHO review and our review is that the study populations with normal BP in the WHO review generally have a high baseline BP in the upper 50% percentile of the population» (Graudal et al., 2020) — and Huang, from the other lineage, finds continuous baseline SBP the significant modifier (section above). Shared trial pool, so not an independent E; but it narrows the normotensive dispute to the SBP ~120 stratum, where both see a small non-zero fall and disagree on whether to call it worth having. (inferred from Graudal et al., 2020; Huang et al., 2020)
On race, the camps split by design: Huang’s between-trial meta-regression keeps non-white ethnicity as an independent modifier, while Graudal reads the group differences as mostly confounding by «baseline BP, age, and amount of sodium reduction» and cites a within-study comparison (Graudal 2015b, not held) with no racial difference. Graudal’s own black estimates and its Asian hypertensive estimate are low-certainty and split into large- and small-effect sub-clusters; the Asian normotensive estimate is moderate-certainty and homogeneous. Route (a) stays the default; ethnicity is not used here to personalize. (Graudal et al., 2020)
The compensation arm — the hormones rise in both lineages, TC and TG rise in Graudal, HDL and LDL are n.s. in both; the dispute is size and persistence
| Marker | Graudal 2020 (cross-over trials, all durations) | >=7 d subgroup | He 2013 (>=4 wk) | Same quantity? |
|---|---|---|---|---|
| Renin (ng/mL/h) | +1.56 (1.39-1.73), 82 trials, high | +1.29 (1.07-1.51) | +0.26 (0.17-0.36) | YES on marker/unit; NO on contrast + duration |
| Aldosterone | +104 pg/mL (CI printed «18.4 to 119.69», evidently 88.4; SoF 4: 102.4, 86.9-117.8) ~ +288 pmol/L, high | +96.6 pg/mL | +73.2 pmol/L (44.9-101.5) | same, after unit conversion; NO on contrast + duration |
| Noradrenaline | +62.3 pg/mL (41.9-82.8) ~ +370 pmol/L, high | +48.7 pg/mL | +187 pmol/L (39-336) | same, after unit conversion; NO on contrast + duration |
| Total cholesterol | +5.19 mg/dL (2.07-8.32) ~ +0.13 mmol/L, high | +4.31 (0.71-7.91) ~ +0.11 (0.02-0.20) mmol/L | +0.05 mmol/L (-0.02 to 0.11) | same, after unit conversion; NO on contrast + duration |
| Triglyceride | +7.10 mg/dL (3.14-11.07) ~ +0.08 mmol/L, high | +7.03 (2.01-12.04) | +0.04 mmol/L (-0.02 to 0.09) | same, after unit conversion; NO on contrast + duration |
| LDL / HDL | +2.46 (-0.97 to 5.9) ns, moderate / -0.30 ns | +2.74 ns / -0.68 ns | +0.05 ns / -0.02 ns | same, after unit conversion |
Graudal columns: (Graudal et al., 2020). He column: (He et al., 2013). Unit conversions (aldosterone x2.77, noradrenaline x5.91, cholesterol /38.7, triglyceride /88.6) are this page’s arithmetic. Neither review reports the sodium contrast of its hormone/lipid subset; the BP-analysis contrasts (~-138 mmol in Graudal, -75 mmol in He) differ, so NO on contrast is this page’s inference from those.
Not a contradiction on the numbers. Every marker’s point estimate moves the same way in both reviews. On cholesterol, He’s «no significant change» has an upper limit of ~0.11 mmol/L, and Graudal’s >=7-day estimate is ~0.11 — compatible estimates carrying opposite labels. He’s renin, aldosterone and noradrenaline rises at >=4 weeks are smaller than Graudal’s but statistically present, which He calls a «small physiological increase». (He et al., 2013)
The joined part is persistence, and both sides argue it from the same confound.
- He/Huang: He attributes Graudal’s lipid change to short trials — «in the meta-analysis by Graudal and colleagues the changes in lipids occurred only with short term trials, and a subgroup analysis including trials with a duration of four or more weeks showed no significant change in lipid concentrations» (that was the 2011 edition) (He et al., 2013). Huang says «The responses of the renin- angiotensin system and sympathetic nervous system as well as adverse metabolic effects associated with acute large falls in dietary sodium do not, however, appear to be present in longer term interventions», citing He 2013 and Rhee 2016, a renin study (population analysis plus meta-analysis of RCTs) the wiki does not hold (refs 21 and 165) (Huang et al., 2020). — the renin-specific warrant behind Huang’s not present. He’s own >=4-week data show those responses smaller, not absent.
- Graudal: «while short duration has been suggested to underestimate the BP effect, it has concomitantly been suggested to overestimate possible adverse effects on hormones and lipids. This idea that the duration of the intervention tends to underestimate some physiological outcomes and overestimate others has not been documented, but still has been used to disregard side effects shown in studies lasting less than four weeks» — and «Very few studies lasting more than four weeks have investigated side effects, and further more these studies do not reduce sodium to the recommended level, but to levels above 87 mmol/day» (Graudal et al., 2020). The 2020 edition reports no >=4-week lipid or hormone estimate, only >=7 days.
The hidden term: across trials, duration and contrast size travel together, and the sources say so themselves. He: the compensatory responses are «bigger with sudden and large decreases in salt intake and much smaller or minimal with a longer term modest salt reduction» (He et al., 2013). Huang: «large, short term reductions in sodium could elicit a different type of blood pressure response» (Huang et al., 2020). Graudal: the >4-week trials stop «above 87 mmol/day» (quoted above), and the renin/aldosterone funnel asymmetry arises because «large sodium reductions to low levels of sodium intake mainly was performed in small studies» (Graudal et al., 2020). Long trials’ low arms sit at ~80-90 mmol/day (He’s ~78-87, from medians 153-162 minus 75) .
Graudal’s account of renin and aldosterone, citing Brunner 1972, is an exponential dose curve («rise slowly as long as the intake is above 100 mmol/day, but exponentially» below it) (Graudal et al., 2020). Both low arms sit below ~100 mmol/day, where that curve places the steep rise; no CI or fitted breakpoint is given. On that curve He’s smaller rises are consistent with its smaller contrast and less-deep low arm, with no fading over time. Graudal’s quantification here is coarse: a reduction from high to moderate intake «induces a factor 2 increase in hormones whereas a reduction of sodium intake from a moderate level to a low level induces a factor 5 increase» (Graudal 1998, cited) (Graudal et al., 2020); nobody has fitted the curve to the two lineages’ contrasts, so this is unquantified. On a fading reading they are consistent with duration alone.
What this page adds is narrower than the bundle itself: for the markers, the bundle makes time-decay and dose observationally equivalent. No held source scales the hormone or lipid rises per mmol or models contrast and duration jointly, so between-trial pooling cannot separate them. Repeated within-trial measurement at a fixed contrast over months would, and Graudal states such trials are scarce. In the He lineage the primary-pool hormone data run four to six weeks; Graudal’s primary-pool durations are not extracted here (its all-duration pools include trials under 7 days; its subgroup is >=7 days, with no >=4-week estimate). He writes «It is likely that such effects might attenuate over time.» (He et al., 2013), backed by one multi-component trial; Graudal writes «The assumption that at least some of these effects may be persistent and not just temporary has been indicated in observational studies.» (Graudal et al., 2020). WHO pooled no renin or aldosterone. Its >=4-week total cholesterol (1-2 months) and noradrenaline (1-2.5 months) pools are n.s. and between-trial; their per-marker contrasts are not stated, but WHO’s overall trial contrast (~76 mmol, 1.74 g/day) is milder than Graudal’s, so they carry the same dose confound (World Health Organization, 2012). In WHO’s narrative, the lipids and catecholamines «transient» sentence cites Graudal’s 2011 edition (WHO ref 33); the renin and aldosterone «persisted» sentence carries only refs 40-41 (renin’s prognostic importance); by context it presumably rests on Graudal 2011’s >=4-week subgroup (ref 33, not held), so the persistence claim cannot be checked here (World Health Organization, 2012). No held source pools, or measures repeatedly within-trial, persistence at a fixed, sodium-only contrast over months. The one held months-scale record is not separately pooled: Jula 1994, a single open parallel trial (76 hypertensives, 365 days per Graudal’s study table), which Graudal excludes from its primary (crossover) marker pools, adds back only in all-duration sensitivity analyses («did not change the outcome»), and records as renin «increase from 0.7 to 1.1 ng/ml/h», aldosterone «increase from 297 to 363 pg/ml» and noradrenaline «unchanged from 610 to 520 pg/ml» during sodium reduction (Graudal et al., 2020). That fits renin and aldosterone still raised at 12 months and noradrenaline not raised then, from one open trial whose sodium-only status is not checked here and no earlier hormone timepoint is reported, so fading is only consistent with it, not observed; Graudal pools these data only in sensitivity analyses and does not argue persistence from them.
Graudal offers the activation as «a possible explanation» for the small BP fall («compensatory activation
of the renin-aldosterone system is proportional to the degree of sodium reduction») (Graudal et al., 2020). He
calls the responses «compensatory» to the volume fall. Huang places the acute activation and the
short-trial BP shortfall side by side but does not link them causally. That the lineages share one
damping mechanism and differ only on its time course is this page’s reading.
(inferred from Graudal et al., 2020; He et al., 2013; Huang et al., 2020) type-F type-G
The BP duration claim — the data agree; the dispute is interpretation. Graudal’s maximal efficacy at one week rests on within-trial repeated measures (its Table 2, data from Graudal 2015): SBP differences between weeks 1, 2, 4 and 6 of -0.50 to 0.39 mmHg, each interval about 2-3 mmHg (one ~1.8) either side of zero, 4-12 trials per comparison (the 12-trial week 2 vs 4 row wraps across a page break in extraction) (Graudal et al., 2020). Huang’s own within-trial and model-based evidence points the same way: «The six studies that recorded multiple measurements at different time points showed no apparent difference in the pattern of blood pressure lowering over time (supplementary fig 2), which corresponded to the meta-regression analyses.» and «Likewise, the univariable meta-regression analyses of all trials showed no association of blood pressure effect with duration of sodium reduction and neither did the multivariable meta-regressions.» (Huang et al., 2020)
Huang’s doubling comes from a subsidiary split: 1.05 (0.40 to 1.70) vs 2.13 (0.85 to 3.40) mmHg per 50
mmol in <=14-day vs longer trials. Huang affirms it: «In trials of more than two weeks’ duration, the
dose-response relation between sodium reduction and blood pressure fall was greater than that in trials
of shorter duration, but there was» [page break] «limited evidence that interventions of longer duration
further increased the effects of sodium reduction on blood pressure», and «Short
term studies underestimate the effect of sodium reduction on blood pressure.» Huang’s hedge covers only
further gains beyond two weeks (Huang et al., 2020).
That the intervals overlap, and that the main duration analyses and the six multi-timepoint trials are
null, is this page’s reading against that affirmation. Because Huang scales per
50 mmol (and its joint model, though not the stratified split, includes both terms), the duration-contrast
bundle confounds the split mainly if the BP dose-response is concave at deep cuts; other stratum
differences (baseline SBP, ethnicity) remain open. The exponential renin/aldosterone rise Graudal cites
(Brunner 1972) would supply such concavity, and the 3.2-vs-0.8 mmHg per 100 mmol gap above hints at it (confounded by
ethnicity, baseline BP and duration). So on BP both sides hold a null, imprecise within-trial result. The
joined part is interpretation: Graudal reads it as «duration has no impact»; Huang reads the earlier
analysis as «unable to determine» and leans on the subsidiary split.
(inferred from Graudal et al., 2020; Huang et al., 2020) type-D (on interpretation)
What it changes for a decision.
- High-intake hypertensive: reduce. Every camp agrees on direction, and the BP effect is ~4-6 mmHg SBP across all three reviews. Agreement stops at direction. Graudal hedges («People with hypertension may benefit due to the effect on BP, but may also suffer from harms due to the side effects.») and finds that «the effect on hormones and lipids were similar in people with normotension and hypertension.» (Graudal et al., 2020). The target level is disputed (WHO’s <2 g/day vs the PURE J below) -> Should Sodium Reduction Be Population-Wide or Targeted. In a hypertensive the marker cost is similar or smaller (aldosterone rises less in Graudal’s hypertensives; their total cholesterol rise is n.s.) and the BP benefit is ~2-5x larger (He ~2x, Graudal ~5x).
- Normotensive: the markers rise (high certainty in Graudal; present but smaller in He at 4-6 weeks), and the BP fall is ~1-2.4 mmHg SBP. Suppose BP lowering protects the same way whatever lowers it, which Graudal contests (beta-blockers lower BP without lowering mortality). Then BPLTTC’s HR 0.91 per 5 mmHg in people without CVD turns 1.1-2.4 mmHg into ~2-4.5% fewer major CV events, relative -> Blood Pressure Lowering and Cardiovascular Events. On a low baseline risk that is a small absolute gain. The marker rises have no held transfer to an outcome (LDL n.s.). Graudal’s no-net-benefit conclusion, «Due to the relatively small effects and due to the antagonistic nature of the effects (decrease in BP, increase in hormones and lipids), these results do not support that sodium reduction may have net beneficial effects in a population of white people with normal BP» (Graudal et al., 2020), puts a percent change in cholesterol beside a percent change in BP with no common outcome scale. For a low-risk normotensive the lever is small even if the compensation fades completely. That is a ceiling-is-a-finding result, and it holds whichever way the persistence dispute resolves. Net effect stays insufficient evidence -> Net Effect vs Intended Effect.
- For scale: the BP fall was measured at Graudal’s overall contrast (to ~65 mmol/day); the contrast behind the lipid rise (~0.1 mmol/L total cholesterol, LDL n.s.) is not reported. Graudal says a significant reduction is hard to hold: «it has been difficult to maintain a significant sodium reduction in longer-term studies» (Graudal et al., 2020). Other things equal, both effects should shrink at a contrast people can sustain.
(Blood Pressure Lowering Treatment Trialists Collaboration, 2021; inferred from Graudal et al., 2020; He et al., 2013; World Health Organization, 2012)
WHO’s no harmful effect verdict is a relevance call, not a no-change finding. WHO rates total cholesterol and noradrenaline high-certainty and «did not cross threshold of relevance» (Limits, below) against a threshold it never quantifies. Graudal’s significant rises and WHO’s verdict can therefore both hold. The disagreement sits partly in the unstated threshold and partly in the data: WHO’s own >=4-week estimates are much smaller than Graudal’s (total cholesterol +0.02 mmol/L, -0.03 to 0.07; noradrenaline +8.23 pg/mL, -27.84 to 44.29), and WHO reads the lipid and catecholamine changes as «transient and no longer present after 4 weeks of reduced sodium intake», while renin and aldosterone «persisted with longer term reduced sodium intake» — the first citing Graudal’s 2011 edition (WHO ref 33), the second carrying only refs 40-41 (renin’s prognostic importance); by context it presumably rests on Graudal 2011’s >=4-week subgroup (ref 33, not held), so the persistence claim cannot be checked here (World Health Organization, 2012). (Graudal et al., 2020; inferred from World Health Organization, 2012)
How the gap is bridged — a disclosed surrogate transfer
The bridge is explicit, and it is the guideline’s most consequential methodological act:
“Because of the well-established relationship between blood pressure and cardiovascular disease outcomes, the evidence of an effect of sodium on blood pressure was also considered moderate-quality evidence that reduced sodium is beneficial for reducing risk of cardiovascular disease, stroke and coronary heart disease.” (World Health Organization, 2012)
High-certainty surrogate evidence is transferred to hard outcomes one level down, and the direct very-low-certainty cohort evidence on those same outcomes is superseded rather than combined. There is no stated roll-up rule anywhere in the document — the string “overall certainty” does not occur, and no overall certainty is ever issued. The transfer rule in Annex 8 is the operative substitute.
Where this bites hardest — the <2 g/day threshold. The evidence table for that specific question carries four rows reading “There were no studies with [cardiovascular disease / stroke / coronary heart disease / all cause mortality] as an outcome which addressed this question”, each labelled (directly assessed). Annex 8 nonetheless states:
“High and moderate-quality evidence that consuming <2 g sodium/day compared with consuming >=2 g sodium/day is beneficial for reducing blood pressure and risk of cardiovascular disease, stroke and coronary heart disease in adults.” (World Health Organization, 2012)
This is not a contradiction, and “no studies” does not mean “no evidence.” Two qualifiers do the work, and both are printed on the cell itself:
- the rows say (directly assessed), and
- footnote 2 hangs off those very cells: “Though the effect of a decrease in sodium intake was not tested, there were 5 cohort studies that compared lower sodium intake (<2g/day) to higher sodium intake (>= 2 g/day). There was a significant increased risk of stroke in the group that consumed >=2 g/day relative to the group that consumed < 2 g/day (RR=1.30 (1.03 to 1.64)). Effects on the risk of cardiovascular disease, coronary heart disease and all cause mortality were not statistically significant.”
(World Health Organization, 2012)
So the threshold claim rests on high-certainty BP evidence at that threshold, transferred at moderate, plus 5 indirect cohort comparisons showing a stroke effect. What is absent is a directly assessed trial of the threshold against a hard outcome — which is a narrower and more accurate statement than “zero evidence.” (inferred from World Health Organization, 2012)
Response heterogeneity — measured, then deliberately not acted on
| Baseline status | SBP reduction (95% CI) | Certainty |
|---|---|---|
| Hypertensive | 4.06 mmHg (2.96 to 5.15), 24 studies | high |
| Mixed | 3.41 mmHg (1.69 to 5.13), 8 studies | high |
| Normotensive | 1.38 mmHg (0.02 to 2.74), 6 studies | moderate |
A roughly three-fold difference, which WHO records as “statistically significantly less” in normotensives — and then declines to stratify the recommendation, on a stated population-level ground: the high global prevalence of hypertension and the breadth of benefit. The reason given is a public-health-standpoint reason, not an evidentiary one (telos divergence class 1). (World Health Organization, 2012)
By baseline intake, point estimates rise monotonically with intake (1.79 -> 2.97 -> 3.07 -> 3.91 -> 5.74 mmHg across intake bands) and WHO reads the subgroup test as null (P=0.17), concluding the effect is “independent of baseline sodium intake.” A monotone gradient declared absent on a non-significant interaction test is a claim about power as much as about biology. (inferred from World Health Organization, 2012)
The J-curve — named as the reason for the review, then never adjudicated
WHO names the J-shaped hypothesis twice in framing sections, citing two 2011 cohort studies proposing that reducing sodium below 2 g/day “may be associated with increased risk of cardiovascular disease and stroke” as part of the justification for commissioning the review. Neither study is returned to anywhere in the evidence summary, the final considerations, or the annexes, and whether they were among the 15 included cohorts is not stated. The terms reverse causation, sick-quitter and U-shaped do not appear in the document. (World Health Organization, 2012)
Compounding this: the strata where a lower-arm harm is most plausible were excluded a priori — “individuals with illnesses or taking drug therapy that may lead to hyponatraemia… (e.g. patients with heart failure and those with type I diabetes)… were not considered in the review of the evidence.” (World Health Organization, 2012)
So the guideline cannot answer the J-curve question, and does not claim to. Per the expectancy test this is unprobed, not disproved — and it is the wiki’s largest open question on sodium.
He 2013 rebuts the J-curve — but from the advocacy side and without the strong check. He dismisses the two 2011 JAMA papers that WHO cited (Stolarz-Skrzypek: lower salt -> higher CV mortality; O’Donnell: a J-shaped association) as having «many methodological flaws, such as measurement error in assessing daily salt intake, confounding factors not controlled for, and reverse causality». The named mechanisms are exactly The U-Shaped Association Artifact’s recipe — so the hypothesis that the low-intake arm is artifact is now on the record with mechanism. But this is a partisan critique (He/MacGregor are CASH/WASH advocates) and, decisively, it is only a critique — no referent-corrected or Mendelian-randomization analysis of the low-intake arm, which is the strong check that concept requires. So He argues the arm away; he does not adjudicate it. (He et al., 2013)
The AWAITS narrows but stays open. PURE (Mente 2016) has now landed — see the direct-hard-outcome
section below — but it is a pole in the dispute, not the confounder-immune adjudicator this hold
seeks: it reports the J directly on hard outcomes yet ran only the weak checks (exclude early
events, exclude prevalent CVD/HTN/diabetes), never a referent-correction or a genetic/MR instrument on
the low arm. So the hold still AWAITS a source that adjudicates the low arm with a confounder-immune
method (a referent-correction, a Mendelian-randomization instrument on the low-intake arm, or a large
long-term RCT of sodium level rather than reduction) — the post-2011 literature (PURE observationally;
the He/MacGregor and Graudal RCT meta-analyses) supplies more poles, not the adjudication. The joined-issue write-up is
Should Sodium Reduction Be Population-Wide or Targeted.
Graudal 2020 (held since 2026-10-22) argues the low arm is real — also without the strong check. The opposing Cochrane review pools no RCT hard outcomes; its low-arm case is argument plus an observational pooled table (Discussion Table 4: all-cause mortality, adjusted cohorts, healthiest subgroups, pooled RR/OR 0.84 (0.76-0.93) or 0.87 (0.76-0.98) for usual vs low intake; cut-points <114 mmol/day (Graudal 2014) or <130 mmol/day (O’Donnell 2014, PURE). The table includes PURE itself (RR 0.62), so on the low arm this is not a pole independent of Mente/PURE.) (Graudal et al., 2020). Against reverse causation it cites adjusted cohorts and NHANES data showing sick and healthy people eat the same sodium. Against measurement error it argues the misclassification is «not systematic, but random… which contributes to an underestimation of the outcome, but do not cause a change of the direction of the outcome» (Graudal et al., 2020). That defence covers only classical random error. The mechanism this page flags for PURE — error that is unequal across the range, largest at the extremes — can bend a flat curve into a U, and a random-error argument does not reach it -> The U-Shaped Association Artifact. So one camp names the artifact recipe without testing it and the other runs weak checks (adjusted cohorts, healthiest subgroups, NHANES); neither runs a referent-correction or MR. (inferred from Graudal et al., 2020)
A third major body encodes the split as uncertainty rather than resolving it [2026-08-04, Afshin GBD 2017].
The GBD 2017 dietary-risks analysis needed an optimal sodium level to compute attributable burden,
and rather than pick one it published the RCT-vs-observational disagreement as a deliberately wide band:
«To reflect the uncertainty in existing evidence on optimal level of intake for sodium, 1—5 g per day was considered as the uncertainty range for the optimal level of sodium where less than 2·3 g per day is the intake level of sodium associated with the lowest level of blood pressure in randomised controlled trials and 4—5 g per day is the level of sodium intake associated with the lowest risk of cardiovascular disease in observational studies.» (Afshin et al., 2019)
This is the exact split this section holds, seen from the modelling side. The BP-minimising intake (RCTs, <2.3 g) and the CVD-minimising intake (observational, 4-5 g) point to different optima, which is what a J-curve on the hard outcome would produce — and GBD, an independent consortium with no salt-advocacy stake, does not adjudicate it either: it widens the uncertainty to 1-5 g and carries both. So the corpus now holds three postures on the low-intake arm — WHO (names it, cannot answer), He/MacGregor (rebut it from advocacy, no strong check), GBD (encode it as uncertainty) — plus the PURE and Graudal poles (sections above), and none runs the confounder-immune adjudication. GBD corroborates unprobed, not disproved; it does not close the gap -> The U-Shaped Association Artifact. (inferred from Afshin et al., 2019)
The hard-outcome RCT arrives — but on a confounded (Na-down + K-up) exposure [2026-08-04, Neal SSaSS]
This page’s central gap was that hard outcomes sit at very low certainty (all cohort evidence, the surrogate-transfer doing the work). SSaSS (Neal 2021, NEJM) is the large RCT with patient-important outcomes — but it tests a potassium-enriched salt substitute (75% NaCl / 25% KCl), so it moves two exposures at once and its benefit cannot be attributed to sodium reduction alone.
What it found (open-label cluster-RCT, 600 rural-China villages, 20,995 people with prior stroke or age >=60 + high BP, 4.74 y follow-up; events adjudicated blind to assignment):
| Outcome | Salt sub vs regular (per 1000 py) | Rate ratio (95% CI) | Absolute |
|---|---|---|---|
| Stroke (primary) | 29.14 vs 33.65 | 0.86 (0.77-0.96), P=0.006 | −4.5 / 1000 py |
| Major adverse CV events | 49.09 vs 56.29 | 0.87 (0.80-0.94), P<0.001 | −7.2 / 1000 py |
| All-cause death | 39.28 vs 44.61 | 0.88 (0.82-0.95), P<0.001 | −5.3 / 1000 py |
The achieved contrast: 24h urinary sodium −15.2 mmol (−350 mg), potassium +20.6 mmol (+803 mg), SBP −3.34 mmHg (−4.51 to −2.18). (Neal et al., 2021)
Why this is the decision-relevant caveat, not a footnote. The intervention raised potassium more than it cut sodium (+20.6 vs −15.2 mmol), and the authors state both exposures «independently lower blood pressure and… have synergistic effects». So SSaSS is strong evidence that a salt substitute reduces stroke, CV events and death in a high-risk stratum — it is not clean evidence that sodium reduction per se delivers that benefit. The corpus’s cleanest hard-outcome sodium RCT is therefore a joint Na:K lever, which is exactly the joint target both this page and Potassium Intake and Blood Pressure flag as un-adjudicated — SSaSS adjudicates the lever, not either electrolyte separately. (inferred from Neal et al., 2021)
The absolute benefit rides on high baseline risk (route-a). These are large absolute effects (−4.5 to −7.2 events per 1000 person-years) because the population is extreme: 72.6% prior stroke, mean BP 154/89, mean age 65. The ~12-14% relative reductions may transport to a lower-risk stratum, but the absolute benefit shrinks with baseline risk — a lean, normotensive, low-stroke-risk person gets a far smaller absolute effect from the same switch. This is Baseline Risk and the Relative-Absolute Split applied: SSaSS earns a strong recommendation for its own stratum, not a universal one. (inferred from Neal et al., 2021)
The lower-arm / contraindication stratum (route-c) — reassuring but scoped. The safety worry for a K-raising intervention is hyperkalaemia. SSaSS found no excess (3.35 vs 3.30 per 1000 py; RR 1.04, 0.80-1.37; P=0.76) and no excess sudden death — but it excluded known serious kidney disease, potassium-sparing diuretics and K-supplements by history, did no biochemical renal prescreening, and did not measure serum potassium serially. So the safety reassurance holds for the screened population; CKD / impaired potassium excretion remains a real harm-firing stratum where a salt substitute is contraindicated (concordant with the route-c boundary on Potassium Intake and Blood Pressure). (Neal et al., 2021)
The events slightly exceed what the measured BP fall predicts — a recurring pattern. Via Blood Pressure Lowering and Cardiovascular Events (BPLTTC: ~13% stroke reduction per 5 mmHg SBP), a −3.34 mmHg fall predicts a stroke RR near 0.91; SSaSS observed 0.86. The observed benefit outruns the BP-channel prediction — the same excess Potassium Intake and Blood Pressure found for potassium’s cohort stroke effect. Three non-exclusive reasons: (i) a BP-independent electrolyte effect (potassium); (ii) the measured −3.34 mmHg understates the sustained effect (adherence dilution + a single-day recovery-biomarker + Huang’s short-trial underestimate — the page’s lower-bound reading); (iii) noise in a single subgroup measurement. Direction is corroborated; the exact split between sodium, potassium and dilution is not identified by this trial. (Blood Pressure Lowering Treatment Trialists Collaboration, 2021; inferred from Neal et al., 2021)
PURE lands the J directly on hard outcomes — a pole, not the adjudicator [2026-08-27, Mente 2016]
The gap this page kept flagging — nobody measures the low arm directly on hard outcomes — is now directly measured, but by an observational study that is itself a pole in the dispute rather than the confounder-immune adjudicator. Mente 2016 pooled 133,118 people (63,559 hypertensive / 69,559 non-hypertensive) from 49 countries across three cohorts (PURE + ONTARGET/TRANSCEND + EPIDREAM), >10,000 events, median 4.2 y follow-up; composite outcome = all-cause death + major CVD event, with 4-5 g/day sodium as the lowest-risk reference. (Mente et al., 2016)
The headline is effect-modification BY hypertension (route-b), on a significant interaction test (P for heterogeneity = 0.0342 for the composite; 0.0135 for death; 0.0432 for major CVD). The stratified HRs vs the 4-5 g/day referent:
| Stratum | <3 g/day (low arm) | >7 g/day (high arm) |
|---|---|---|
| Hypertensive | HR 1.34 (1.23-1.47), p<0.0001 | HR 1.23 (1.11-1.37), p<0.0001 |
| Non-hypertensive | HR 1.26 (1.10-1.45), p=0.0009 | HR 0.90 (0.76-1.08), p=0.25 (NS) |
Two distinct shapes: the low-arm harm is present in BOTH strata (and survives BP adjustment — low-arm HR 1.35 [1.23-1.49] in hypertensives, still p=0.0011 in non-hypertensives — so Mente reads it as not BP-mediated), while the high-arm harm is confined to hypertensives (flat/null in normotensives). Only ~10% of the pooled population had both hypertension and >6 g/day, which is the entire basis for Mente’s targeting conclusion — reduce sodium only in high-intake hypertensives, not population-wide. (Mente et al., 2016)
The BP slope is itself hypertension-modified (systolic mmHg per g sodium: hypertensives 2.08 [1.96-2.21] vs non-hypertensives 1.22 [1.13-1.30]; P<0.0001 for interaction) — concordant with the salt-sensitivity gradient this page holds, and it makes the high-arm stratification mechanistically coherent. But it does not explain the low-arm harm, which is where the artifact question lives. (Mente et al., 2016)
Why PURE is a pole and not the adjudication this page’s AWAITS seeks. The exposure is a single fasting morning midstream urine run through the Kawasaki formula to estimate 24h sodium (validated ICC ~0.70 vs a 24h collection; regression-dilution corrected from 30-90-day repeats in 448 people). That is a one-day recovery biomarker, and its error is largest at the extremes — exactly where the J-arms sit — so a flat true relationship can be bent into a U by unequal measurement error alone, with no confounder and no reverse causation, and it would pass the checks Mente ran -> Measurement Error in Dietary Assessment, The U-Shaped Association Artifact. Mente ran only the weak artifact checks (exclude events in the first 2 years; exclude in turn those with known CVD/hypertension/diabetes; exclude anti-hypertensive users) — the pattern held — but ran no referent-correction and no genetic/Mendelian-randomization instrument on the low arm, which is the strong check The U-Shaped Association Artifact requires. Mente himself concedes observational analysis cannot prove causality and calls for large long-term RCTs of sodium level. So the low arm is directly observed but still unadjudicated — the [PRIOR] on the U-shaped artifact does not close here. (inferred from Mente et al., 2016)
The observed-vs-modelled discordance is the study’s sharpest single result. Mente built a simulation projecting CVD risk from sodium assuming the effect runs only through systolic BP, then compared it to the directly-observed HRs. The two diverge most at <3 g/day: the BP-only model predicts lower risk as sodium falls, while the observed data show rising risk — the exact discordance a hidden low-arm artifact would also produce, so it is evidence that BP-mediation is incomplete OR that the low arm is contaminated, and the study cannot separate those. This is why the WHO/He surrogate-transfer chain (sodium -> BP -> CVD, assumed monotone) and PURE’s direct observation give opposite signs below 3 g/day. The full joined-issue analysis is Should Sodium Reduction Be Population-Wide or Targeted. (inferred from Mente et al., 2016)
What is absent
- Absolute effects entirely. No per-1000, no risk differences, no baseline risk, no NNT. The mmHg figures are absolute in units but never translated to outcome risk. So this recommendation cannot be ranked against any other exposure in the fabric (telos layer 1).
- A numeric “threshold of relevance” — the criterion driving every imprecision downgrade is defined only qualitatively and never operationalized for any outcome.
- Salt sensitivity / hyperresponders — the terms do not occur. The hypertensive/normotensive split above is the closest the document comes.
- Sodium:potassium ratio — explicitly out of scope, while asserting that following this guideline plus the potassium guideline yields “approximately one to one, which is considered beneficial for health”, sourced to a 2003 expert consultation rather than to evidence graded here. The potassium half is now held — Potassium Intake and Blood Pressure (WHO’s companion Aburto 2013 SR+MA) — and it is the same WHO/CASH lineage, so the two are two arms of one programme, not independent witnesses; the Na:K ratio itself stays un-adjudicated on both pages (asserted 1:1-beneficial, never graded against a hard outcome). Note the asymmetry: sodium reduction lowers BP in normotensives too, potassium raising does not (its normotensive effect is a flat zero).
- Any low- or middle-income-country evidence. All RCTs were run in Australia, Europe, North America and New Zealand; cohorts in Europe, Japan and the USA. The PICO setting is “All countries” and transportability is never discussed — while the guideline’s stated motivation is LMIC-framed.
- Any salt-provenance distinction. The guideline treats “salt” as sodium chloride and never addresses whether the type of salt matters — see the standing note below, which the belief that sea/Himalayan salt is healthier makes worth stating.
(inferred from World Health Organization, 2012)
Salt type does not change the sodium — sea and Himalayan salt carry the same BP burden [2026-09-25, belief-harvest run 17 B2]
The BP lever is the sodium ion, and all edible salt is sodium chloride whatever its provenance — table, sea, and Himalayan salt deliver essentially the same sodium per gram, so specialty salts carry the same BP burden gram-for-gram. Their trace-mineral content differs, but the amounts are nutritionally negligible at culinary doses: to get a meaningful mineral dose from the trace fraction you would first have to eat a BP-harmful quantity of salt. The one real difference runs backwards for the “natural salt is healthier” belief — table salt is commonly iodine-fortified and most specialty salts are not, so switching away from iodized salt removes a population deficiency-correction vehicle and can lower iodine intake -> Excess Iodine Intake and Thyroid Disease (iodized salt is held there as the structural, population-level iodine-deficiency lever).
(inferred from World Health Organization, 2012)
Sodium is one BP lever among several — the DASH pattern lowers BP independently of it [2026-08-07, Siervo]
Sodium reduction is not the only dietary BP lever. The DASH pattern lowers SBP by -5.2 mmHg (-7.0, -3.4) vs control in a 20-RCT MA — and Siervo’s meta-regression finds the between-arm sodium difference does not predict that BP change (SBP P=0.67), so DASH’s effect is carried by the whole pattern (K, Ca, Mg, fibre, nitrate, low SFA), not by its incidental sodium contrast -> DASH Diet and Blood Pressure. (Siervo et al., 2014)
Decision consequence — do NOT double-count. DASH’s -5.2 mmHg and this page’s sodium-reduction effect (He -4.18, WHO -3.39) are different exposures against different comparators, not two independent additive channels, so they cannot be summed. They do stack when deliberately combined (the DASH-Sodium factorial shows salt restriction adds to DASH), but that is a stacking of two applied interventions, not an attribution of DASH’s own effect to sodium. (inferred from Siervo et al., 2014)
Refinement (2026-10-06, Filippou): a possible DASH x background-sodium interaction — a between-trial hypothesis, not a stacking rule. A 30-RCT DASH pooling (SBP -3.2 mmHg overall, on attained rather than change-from-baseline BP) splits trials by sodium-intake level: SBP fell -4.5 mmHg (-6.1, -3.0; 9 trials) where intake exceeded 2400 mg/d and -2.1 mmHg (-2.5, -1.8; 14 trials) at or below it (P=0.003; DBP difference n.s., P=0.39). (Filippou et al., 2020) The authors point to DASH-Sodium:
«Indeed, Sacks et al. (46) suggested that the BP-lowering effect of the DASH diet compared with the control diet was almost 3-fold higher for those at higher than for those at lower sodium intake.» (Filippou et al., 2020)
That trial (ref 46) contributes arms to both sodium subgroups, so the Sacks figure is not separate support for the split; it is also reported second-hand through Filippou rather than read from the trial (read from the trial itself 2026-10-06, below). And the split is confounded with age: 7 of the 9 high-sodium trials are also in the mean-age <50 subgroup, and the low-sodium, age >=50 and drug-treated subgroups all pool to about -2 mmHg with near-identical narrow intervals. (Filippou et al., 2020) That pattern suggests the same large trials (the largest, Naseem n=1492, is in all three) drive the low-effect subgroups. (inferred from Filippou et al., 2020) Filippou’s own meta-regression on the change in 24-h urinary sodium was null, consistent with Siervo’s null on the between-arm dietary sodium difference (related measures, not the same one).
This is not the same variable as Siervo’s null. Siervo tested the between-arm sodium difference (does DASH’s own sodium contrast explain its effect? no); Filippou’s split is by the trial’s reported sodium-intake level (which arm the figure describes is not stated; read here as background intake). Both can hold. Decision consequence: none yet. The no-double-count and stacking statements above stand; whether DASH adds less once sodium is already low is an open hypothesis, owed a within-trial factorial reading (DASH-Sodium itself), not a number to quote. (Cashed 2026-10-06 in the next subsection.) (inferred from Filippou et al., 2020; Siervo et al., 2014)
The factorial itself: background diet halves the sodium effect, and the two levers stack sub-additively [2026-10-06, Sacks 2001]
DASH-Sodium (NEJM 2001, n=412, SBP 120-159 / DBP 80-95, 41% stage-1 hypertensive, drug-treated excluded) randomized diet (DASH vs typical-US control) and fed each person three sodium levels for 30 days each in random order, with weight held constant. Achieved urinary sodium was ~141-144 / ~106-107 / ~64-67 mmol/d against targets of 150 / 100 / 50. (Sacks et al., 2001) The trial is inside He 2013’s pool (its ref 37), Siervo’s (its ref 28) and Filippou’s (ref 46), so it is a primary trial refining the pooled estimates, not a new witness to them. (Filippou et al., 2020; He et al., 2013; inferred from Sacks et al., 2001; Siervo et al., 2014)
| Sodium contrast (SBP, mmHg, 95% CI) | Control diet | DASH diet |
|---|---|---|
| high -> intermediate | -2.1 (-3.4, -0.8) | -1.3 (-2.6, 0.0) |
| intermediate -> low | -4.6 (-5.9, -3.2) | -1.7 (-3.0, -0.4) |
| high -> low | -6.7 (-5.4, -8.0) | -3.0 (-1.7, -4.3) |
1. The sodium effect depends on what else is eaten (route b, positive within-trial evidence).
«The level of dietary sodium had approximately twice as great an effect on blood pressure with the control diet as it did with the DASH diet (P<0.001 for the interaction).» (Sacks et al., 2001)
This is a randomized interaction between two randomized exposures, which is the evidence route (b)
asks for. It is one trial, 30-day periods, BP only. The authors’ explanations (potassium/calcium in DASH
blunting the low-sodium effect, or low sodium blunting potassium’s) are offered with «perhaps» and not
tested -> Potassium Intake and Blood Pressure (which now holds a between-trial baseline-sodium
subgroup from a potassium dose-response meta-analysis pointing the same way, still a hypothesis
(inferred from Filippini et al., 2020) [2026-10-22]). Decision consequence: a person already eating a
DASH-like diet should expect roughly half the BP fall from the same sodium cut that a typical-US eater
gets (in a 30-day fed trial, over ~140 -> ~65 mmol/d; free-living DASH-like diets untested); the sodium
cut still lowered BP on DASH. The halved sodium step, the sub-additive combination (point 2) and DASH’s
smaller effect at low sodium are one diet x sodium interaction seen from each lever
(6.7 - 3.0 = 12.6 - 8.9 = 5.9 - 2.2 = 3.7 mmHg SBP, using the point-2 figures below), not three findings. (inferred from Sacks et al., 2001)
2. The they stack statement above, refined: they stack sub-additively. DASH plus low sodium against control plus high sodium lowered SBP -8.9 mmHg (-6.7, -11.1), more than either alone, and «The reductions in blood pressure caused by the combination of dietary interventions were smaller than they would have been if the effects of each dietary intervention were strictly additive (P<0.001 for the interaction).» (Sacks et al., 2001) The single effects from the same start (DASH -5.9, sodium -6.7) sum to about -12.6. So the no-double-count rule now has a number behind it: estimate a combined DASH + sodium effect from the factorial cell, not by adding the two. (inferred from Sacks et al., 2001) This also answers the open question above in its within-trial form: DASH vs control SBP was -5.9 at high sodium and -2.2 (-4.4, -0.1) at low, so DASH adds less once sodium is already low. (Sacks et al., 2001) Whether the low-effect subgroups in Filippou’s pooling are explained by this or by age stays open: Sacks reports no age subgroup. (Filippou et al., 2020; inferred from Sacks et al., 2001)
3. Shape over the studied range: no plateau; on the control diet, steeper at the lower step.
«In the control diet, a reduction in the sodium intake of about 40 mmol per day from the intermediate sodium level lowered blood pressure more than a similar reduction in the sodium intake from the high level (P=0.03 for systolic blood pressure, P=0.045 for diastolic blood pressure).» (Sacks et al., 2001)
The achieved steps were 35 and 42 mmol/d, so the step sizes do not explain the difference. On DASH the two steps were -1.3 and -1.7 with no step comparison reported. Scoped claim: over achieved urinary sodium ~64-144 mmol/d, in 30-day feeding periods, three points show no knee or plateau, and on a typical-US diet the curve steepens toward the lower end. Nothing is shown below ~64 mmol/d. (inferred from Sacks et al., 2001)
Parameter table (op-weave 2a) — does this within-trial shape meet Huang’s null baseline-intake modifier?
| Parameter | Huang 2020 | Sacks 2001 | Same quantity? |
|---|---|---|---|
| What varies | trial’s baseline sodium intake (crossover trials: the highest-sodium period) | the fed sodium level at which a ~40 mmol step starts, same people | PARTLY — both ask whether the response depends on the starting level |
| Design | between trials, «no association for baseline sodium intake (P trend=0.20)» | within person, randomized order, P=0.03 SBP | NO — between-trial association vs within-trial contrast |
| Response | SBP reduction by baseline-intake group | SBP fall per step on the control diet | PARTLY |
Huang column: (Huang et al., 2020). Sacks column: (Sacks et al., 2001)
Not joined (scope check ii): a non-significant between-trial association and a significant within-trial step contrast can both hold, and the Huang null has the power caveat this page already attaches to WHO’s by-intake test. Recorded as a distinction, not a tension. Both are compatible with this page’s every-reduction-pays default over the studied range. (inferred from Huang et al., 2020; Sacks et al., 2001)
4. Subgroups — the hypertensive gradient again, from inside the pools. The sodium effect was present in every prespecified subgroup and larger in hypertensives:
«The effects of sodium were greater in participants with hypertension than in those without hypertension (interaction, P=0.01 on the control diet; P=0.003 on the DASH diet), in blacks on the control diet than in participants of other races or ethnic groups on that diet (P=0.007), and in women on the DASH diet than in men on that diet (P=0.04).» (Sacks et al., 2001)
Subgroup P values were not adjusted for multiple comparisons, by protocol. (Sacks et al., 2001) The hypertensive > normotensive direction matches He, WHO and Huang, but this trial is in He’s pool and referred to by Huang through DASH-Sodium secondary reports (e.g. Akita 2003) (WHO’s study list is unpublished), so it is not corroboration; it does not independently strengthen the page’s existing baseline-SBP modifier (route b, above). For race and sex, which are single-trial, unadjusted, and diet-specific (each reported for one diet only), route (a) stays the default. (He et al., 2013; Huang et al., 2020; inferred from Sacks et al., 2001)
Tolerability, and the limits. Headache was reported by 47% on high-sodium control vs 39% on low-sodium control and 36% on low-sodium DASH; non-completion was 7 / 7 / 8 across sodium levels. (Sacks et al., 2001) Controlled feeding measures efficacy; the authors’ own words are that they «measured only true biologic effects» and that «long-term health benefits remain to be demonstrated». The periods were 30 days, the endpoint is BP, and people on BP drugs were excluded. (Sacks et al., 2001) The authors’ statement that the trial «should settle the controversy» over sodium in people without hypertension is their assertion about a BP surrogate; it does not reach the hard-outcome J-curve dispute -> Should Sodium Reduction Be Population-Wide or Targeted. (inferred from Sacks et al., 2001)
Limits
- 2012; self-dated for review “by the end of 2017”. The post-2011 sodium controversy is by construction outside it.
- Three of the four underpinning systematic reviews are unpublished WHO in-house documents with no journal or URL, so study lists, risk-of-bias assessments and heterogeneity statistics are not verifiable from this document alone.
- The guideline’s evidence-state vocabulary has three states (conclusive-of-benefit/harm, conclusive-of-no-effect, inconclusive) and the middle one requires the threshold it never specifies. WHO does exercise the conclusive-of-no-effect state — the tempting reading that it never does is wrong: total cholesterol and plasma noradrenaline are rated high and explicitly “did not cross threshold of relevance of benefit or harm”, with Annex 8 carrying “High-quality evidence that decreasing sodium has no harmful effect on blood lipids, catecholamine levels, renal function or any minor side effects (e.g. headache and dizziness) in adults.” It is used on the adverse-effect outcomes, and scoped to adults. What the guideline does not do is apply it to the efficacy outcomes — those stay “inconclusive” in the body and reappear as “no association” in the Executive summary, which is the four-states error the telos names.
(inferred from World Health Organization, 2012)
The exposure measure here is in a different class from the rest of the corpus [2026-07-28, Willett ch.8]
Almost every dietary exposure the wiki holds is measured by food-frequency questionnaire. Sodium is not. Van Dam & Hunter classify urinary sodium as a recovery biomarker — one of only four they name, alongside urinary nitrogen, urinary potassium and doubly-labelled-water energy:
«In addition to urinary nitrogen, urinary potassium, urinary sodium, and energy intake estimated using the doubly labeled water technique are recovery biomarkers.» (Willett, 2012)
A recovery biomarker has «a quantitative relationship… between values for the biomarker and dietary intake in a specific time period», so it yields absolute intake rather than a ranking. Full taxonomy and its consequences: Measurement Error in Dietary Assessment.
What this does and does not buy.
- It removes one error source, not the others. Design, confounding and outcome ascertainment are untouched. This page’s certainty split — high on blood pressure, very low on hard outcomes — is unaffected, and nothing here upgrades it.
- It does not solve day-to-day variation. A 24-hour urine measures one day, and a single day is a poor estimate of habitual intake for any nutrient. Recovery fixes calibration; it does not fix within-person variability, and the two are independent problems.
- What it does buy is that a sodium threshold is auditable in a way most are not. Chapter 4’s division — questionnaires for ranking, records or recalls for absolute magnitude when «comparing nutrient intakes with specific dietary recommendations» — is the general obstacle to checking a person against a cutpoint. Sodium escapes it, because an absolute intake can be measured directly. WHO’s «<2 g/day sodium» is therefore one of the few guidance numbers in this corpus a person could actually be measured against. (inferred from Willett, 2012; World Health Organization, 2012)
A second guidance family on the same stratification — and it is NOT a second witness [2026-07-28, ESC]
ESC reports the same hypertensive/normotensive split this page already holds from WHO:
«A reduction in sodium intake may reduce SBP by, on average, 5.8 mmHg in hypertensive, and 1.9 mmHg in normotensive patients.» (European Society of Cardiology, 2021)
Parameter table (op-weave 2a) — built because the surface agreement is striking:
| Parameter | WHO 2012 | ESC 2021 | Same quantity? |
|---|---|---|---|
| Hypertensive SBP reduction | 4.06 mmHg (2.96-5.15), 24 studies | 5.8 mmHg | NO — ESC states no dose |
| Normotensive SBP reduction | 1.38 mmHg (0.02-2.74), 6 studies | 1.9 mmHg | NO — same reason |
| Ratio hypertensive : normotensive | 2.94 | 3.05 | comparable — both are internal ratios, so the missing dose cancels |
| Sodium reduction assumed | specified in WHO’s analysis | not stated in this sentence | — |
| Uncertainty | intervals given | none given | no |
The magnitudes are not comparable and the ratio is. ESC’s sentence attaches no dose, so its 5.8 and 1.9 cannot be set against WHO’s 4.06 and 1.38 — a larger assumed reduction would produce larger numbers with no disagreement at all. What survives the dose problem is the internal ratio, because the unstated dose divides out of it: 2.94 against 3.05. Two guidance families, a decade apart, on a roughly 3:1 hypertensive-to-normotensive gradient.
That agreement is NOT independent backing, and the check is what shows it
Both trace to the same investigators. ESC’s cited source for this sentence is He FJ, Tan M, Ma Y, MacGregor GA (JACC 2020), and He FJ / MacGregor GA studies sit inside WHO’s own cited evidence base. So the convergence runs through one research group’s primary work reaching two guideline committees.
He 2013 is that lineage’s own primary meta-analysis, and it gives a THIRD estimate of the same ratio: 2.23 (SBP -5.39 hypertensive / -2.42 normotensive) — against WHO’s 2.94 and ESC’s 3.05. So the corpus now holds three ~2-3:1 hypertensive-to-normotensive gradients, all from the same He/MacGregor lineage — which is why none is an independent witness to another. Three numbers on one gradient is a robustness of that group’s finding, not cross-group convergence; the spread (2.23-3.05) reflects different pooled sets and unstated doses, not three independent confirmations. (inferred from He et al., 2013) (Not the stronger claim that both bodies pooled the identical study set — WHO’s actual pooled lists are in unpublished in-house reviews this page’s Limits calls unverifiable; see the width-of-claim note below.)
No [E-independent] is claimable here. Every surface marker points the wrong way — different body,
different continent, different decade, near-identical ratio — and the substantive marker is absent.
This is the laundering trap in its textbook form, and it was caught by checking the reference rather
than by the numbers looking suspicious.
(inferred from European Society of Cardiology, 2021; World Health Organization, 2012)
Stated at the width the evidence supports: what is established is shared primary-study lineage, not that the two pooled estimates were computed from identical study sets. The wiki has not established that, and the differing magnitudes suggest they were not.
What ESC adds that WHO does not have
- A hard-outcome magnitude for a stated dose: «In a meta-analysis, salt reduction of 2.5 g/day resulted in a 20% reduction of ASCVD events (RR 0.80).» This page’s central gap is that hard outcomes sit at very low certainty, and ESC supplies a figure — but from the same He/MacGregor source, so it does not raise certainty, and ESC attaches no certainty rating to it.
- The gap between intake and target, quantified: «In most Western countries, salt intake is high (9-10 g/day), whereas the recommended maximum intake is 5 g/day.» A halving, population-wide — which is an adherence fact, not an efficacy one, and belongs in any realistic reading of what the recommendation asks. (European Society of Cardiology, 2021)
A blood-pressure number is not a blood-pressure number [2026-07-28, ESC chunk 05]
This page’s outcome is SBP, and it already carries a puzzle: WHO’s pooled resting SBP effect is 3.39 mmHg while its ambulatory SBP effect is 5.51 mmHg — the ambulatory figure is larger than the resting one, on the same exposure. ESC supplies the mechanism.
White-coat hypertension is common and carries intermediate risk:
«It occurs in up to 30-40% of patients. The risk associated with white-coat hyperten- sion is lower than sustained hypertension but may be higher than normotension.» (European Society of Cardiology, 2021)
So office/resting measurement misclassifies a large minority, in a direction that adds variance uncorrelated with the exposure. An intervention effect measured against a noisier, upward-contaminated baseline attenuates, which is the expected direction of the WHO resting-vs-ambulatory gap. (inferred from European Society of Cardiology, 2021; World Health Organization, 2012)
ESC applies a measurement correction to SPRINT’s headline number
«the even lower SBP in the intensively treated group in SPRINT (Systolic Blood Pressure Intervention Trial) (mean 124 mmHg) probably reflects a conventional office SBP range of 130-139 mmHg» (European Society of Cardiology, 2021)
A landmark trial’s most-quoted number is restated as 130-139 in the units a clinic actually uses — a shift of roughly 6-15 mmHg, larger than the entire sodium effect this page reports. The trial is not wrong; the number is method-specific, and reading it against a conventional office reading is a same-quantity error of the exact kind the parameter-table rule exists to catch.
The consequence for this page, stated at the right width: any BP figure — WHO’s 3.39, ESC’s 5.8/1.9, a personal reading — carries an implicit measurement method, and the methods are not interchangeable. This does not impeach any figure held here. It means a threshold and a measured value must share a method before they can be compared, and none of the sources this page holds states its method in the same breath as its number.
AWAITS a source specifying the BP measurement protocol behind WHO’s pooled estimates — that is
what would let this page say whether its central figure is an office, resting-standardised or
ambulatory quantity. The wiki cannot currently say which.
A third guidance family sets a slightly laxer chronic-disease limit — NNR2023 [2026-08-27, NNR revisit]
NNR2023 is a third guidance family on the sodium -> BP -> CVD chain (after WHO and ESC), and it lands on a modestly higher chronic-disease-reduction intake than WHO. It carries two distinct sodium numbers, on two different curves:
- AI (adequate intake) 1.5 g/day sodium — the deficiency-side population reference value (~3.75 g salt): «(EFSA, 2019b; NASEM, 2019), the AI in NNR2023 is set to 1.5 g sodium per day» … «(females and males), which corresponds to 3.75 g salt per day.» (Nordic Council of Ministers, 2023)
- CDRR (chronic-disease-risk-reduction) limit 2.3 g/day sodium (~5.75 g salt), adapting NASEM: «(CDRR) of 2.3 g/d are expected to reduce chronic disease risk within the general population. NNR2023 thus adapts the reasoning from NASEM to recommend limiting intake to 2.3 g/d (about 5.75 g salt).» (Nordic Council of Ministers, 2023)
Parameter table (op-weave 2a) — WHO target vs NNR CDRR, matched:
| Parameter | WHO 2012 | NNR2023 | Same quantity? |
|---|---|---|---|
| Chronic-disease-reduction sodium ceiling | <2 g/day (5 g salt), strong | <=2.3 g/day (5.75 g salt), CDRR | YES — both a chronic-disease-reduction intake ceiling |
| Deficiency-side reference | not set (the guideline is a reduction target) | AI 1.5 g/day (3.75 g salt) | NO — different construct |
The divergence is real but small, and it is a threshold-pick difference on a shared evidence base, not a disagreement about the curve. NNR endorses the same linear BP dose-response WHO does — «Sodium restriction down to a sodium intake level of less than 2 g/d decreases blood pressure linearly by a dose-response manner» and «Interventional studies confirm the efficiency and safety of reducing sodium intake to a level of less than 2 g/d (Jula, 2023).» (Nordic Council of Ministers, 2023) — yet sets its recommended CDRR ceiling at 2.3 g/d following NASEM rather than at WHO’s <2 g/d. So the ~0.3 g/d gap is where each body drew the policy line, not a difference in the underlying slope.
This is NOT independent backing. NNR2023 is a guideline synthesis resting on the qSR/NASEM/EFSA
evidence base (its CDRR is explicitly adapted from NASEM), which overlaps the same primary evidence
WHO distilled — so it is a guidance-family confirmation of direction while choosing a slightly laxer
number (telos divergence class 1: a population-standpoint threshold pick), not a second independent
witness. No [E-independent]; the surrogate-high / hard-outcome-very-low reading is unchanged.
(inferred from Nordic Council of Ministers, 2023)
Sodium is pleiotropic — a second outcome channel: kidney-stone recurrence
Sodium reduction pays on a patient-important outcome beyond blood pressure. In the stone-former stratum, high sodium intake raises urinary calcium (reduced tubular reabsorption), lowers urinary citrate (bicarbonate loss), and raises sodium-urate crystal risk — so «Calcium stone formation can be reduced by restricting sodium and animal protein», and restricted salt is beneficial specifically «if there is high urinary sodium excretion» (LE 1b, Strong) (European Association of Urology et al., 2026). Caveat on the causal weight of sodium alone: the positive sodium-risk correlation was «confirmed only in women» and «There have been no prospective clinical trials on the role of sodium restriction as an independent variable in reducing the risk of stone formation» (European Association of Urology et al., 2026).
The incidence channel now carries a gold magnitude [2026-09-23, Lin 2020]. Beyond the EAU
recurrence lever, a gold SR+MA of incident nephrolithiasis in the general population pools high vs low
dietary sodium at RR 1.38 (1.21-1.56) — «high in-take of dietary sodium increased the risk by 38%»,
via hypercalciuria (Lin et al., 2020). This is a
type-F magnitude upgrade over the guideline: the EAU statement gave the direction for recurrence, Lin
supplies a pooled RR for incidence — but it is not an independent second witness (Lin’s cohorts
likely overlap the Curhan/NHS data EAU rests on, and its sodium contrast is highest-vs-lowest
observational with no dose-response), so it sizes the direction, not a threshold, and [E-independent]
is not claimed. Incidence (Lin) and recurrence (EAU) are a stratum distinction, not a tension
-> Kidney Stone Recurrence Prevention.
Layer-1 consequence — the same lever, two rocks. Because sodium reduction moves both BP and stone risk, its marginal rank rises for anyone in both strata: a stone former already motivated on BP gets stone prevention as a co-benefit at no extra adherence cost, and vice versa. This is the pleiotropy guard on the drug-substitution rule — a BP drug that captures the blood-pressure benefit does not shrink the sodium lever’s stone-channel value (substitution is outcome-specific). -> Kidney Stone Recurrence Prevention (inferred from European Association of Urology et al., 2026; Lin et al., 2020)