Type-A synthesis (a structure induced across sources, present as a general claim in none). Two dietary exposures the wiki holds — saturated fat and sodium — share one shape: the guideline is rated most certain about the surrogate and least certain about the patient-important outcome the surrogate stands for. Each source states its own split; neither states the regularity, and neither holds the counter-instance that reveals what actually drives it.
The pattern — matched on within-exposure certainty
The compared parameter is GRADE certainty on the surrogate vs certainty on the patient-important outcome, within the same guideline’s own evidence profile — both grades are the body’s own, on the same exposure, so this is like-for-like.
| Exposure | Surrogate — certainty | Patient-important outcome — certainty | Gap | Carrier page |
|---|---|---|---|---|
| Reduce saturated fat | LDL cholesterol — HIGH | CVD events — Moderate; all-cause mortality — Moderate (null) | 1 level | Saturated Fat Intake and Replacement |
| Reduce sodium | blood pressure — HIGH | direct cohort hard outcomes — VERY LOW (fatal stroke Low; one 2-RCT CVD estimate Moderate) | up to 3 levels | Sodium Intake and Blood Pressure |
In both cases the best-known quantity is the marker, not the outcome. WHO’s sodium guideline makes the move explicit — because the direct outcome evidence is very low, «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» — the outcome recommendation is carried by the surrogate, transferred down one level (WHO Sodium Annex 8, on Sodium Intake and Blood Pressure). The sodium gap quoted above is floor-to-ceiling; the effective gap after that transfer is narrower.
The counter-instance, and what it reveals about the cause
The inversion is NOT a law, and physical activity breaks it: it is rated HIGH certainty on all-cause mortality itself — a patient-important outcome, no surrogate needed (total activity HR 0.34, 0.27-0.43, HIGH; on Physical Activity Dose and Mortality). There is no gap to invert.
Attempting this contradiction relocates the cause — and the first candidate cause fails. The obvious story is self-reported diet flattens the outcome signal. Sodium refutes it: sodium exposure is objectively measured — urinary sodium is a recovery biomarker, absolute not self-reported (Measurement Error in Dietary Assessment) — yet sodium carries the widest outcome-certainty gap of any exposure here. If self-report were the driver, the best-measured exposure would not be the worst-graded on outcomes.
So the driver is the availability of credible hard-outcome evidence, not how the exposure is measured:
- sodium’s hard-outcome evidence is cohort (very low), because almost no trials randomised people to a sodium intake and followed them to events; its blood-pressure evidence is 36 RCTs (high);
- physical activity’s HIGH grade rests on eight harmonized cohorts, all device-measured, with a mortality endpoint — the outcome itself has been studied at scale;
- self-report is a real aggravator on the diet side but a secondary one — it is why SFA’s observational mortality estimate is graded Very low, but it is not why sodium’s is.
The inversion is therefore a symptom of thin hard-outcome trial evidence meeting a well-studied
marker — wherever the outcome has been tried (or measured in large objective cohorts), certainty can
attach to the outcome directly.
[INFERRED (WHO - Sodium Intake 2012; WHO - Saturated and Trans Fatty Acid Intake 2023; WHO - Physical Activity Web Annex Evidence Profiles 2020) — each certainty grade and the sodium biomarker classification are the sources'; the cross-exposure regularity, the counter-instance, and the trial-availability mechanism are this page's]
Why carbohydrate restriction is NOT a third instance
It looks like one — HbA1c is graded High at 6 months while medication-free remission is Low (Carbohydrate Restriction and Type 2 Diabetes Remission) — but the structure differs. HbA1c does not stand for remission; remission is defined as an HbA1c threshold, so this is a marker against a stricter cut of the same marker (plus medication status), not a surrogate standing for a distinct patient-important outcome (events, death). Its lower certainty comes from few, short trials and medication withdrawal, not from a surrogate-to-outcome transmission gap. Recorded as a distinction, not folded in — including it would stretch the pattern to fit.
Why it is a decision-change, not a curiosity
When a dietary guideline sounds confident, check what the confidence is attached to. For these exposures it attaches to a marker (LDL, blood pressure), and the patient-important outcome sits one to three levels lower — so:
- Discount confident surrogate-based dietary advice toward the outcome’s certainty, not the marker’s -> Surrogate Outcomes. A strong recommendation resting on high LDL evidence is not a strong claim that fewer people die — the SFA guideline is the worked case (strong recommendation, HIGH on LDL, all-cause mortality a Moderate-certainty null).
- What would raise the certainty is a hard-outcome trial, not another marker study — the gap does not close by measuring LDL or blood pressure more precisely.
Limits
- Two instances and one counter-instance — a pattern with a boundary, not a proven law.
- Not independent corroboration. Both inverted instances are WHO guidelines applying the same GRADE
machinery, so their shared shape is partly a shared method, not two independent witnesses — which is
itself part of the finding (the inversion may be as much a property of how GRADE meets thin outcome
evidence as of the evidence). No
[E-independent]. - Coherence, not validity (method-risks R1): the pattern says where guidance is certain, not where it is right. A high-certainty surrogate may transmit to the outcome exactly as claimed; the inversion is about warrant, not truth.
Self-critique [run 2026-07-28, before commit — two defects caught and fixed]
- The first mechanism was self-refuting and was replaced. The draft blamed self-reported diet and listed sodium among the self-reported exposures — but sodium is objectively measured (recovery biomarker) and has the widest gap, which falsifies that mechanism. Corrected to hard-outcome trial sparsity, with sodium as the proof rather than an example of self-report.
- A stretched instance was removed. Carbohydrate restriction was dropped from the pattern (HbA1c is the remission-defining marker, not a surrogate for a distinct outcome) and re-filed as a distinction — the not-joined discipline over instance-count.
- Parameter-table discipline: the matched quantity is within-exposure certainty (surrogate vs outcome), both grades from the same body’s own profile — not LDL compared to blood pressure across exposures.
- Independence checked and denied: both instances are WHO/GRADE — recorded as a limit, no E-claim.