The fabric’s nucleus for fibre. Fibre content was scattered across the sugar, whole-grain, and food-matrix pages; two meta-analyses now anchor it — Reynolds 2019 (the WHO-commissioned Lancet series, 185 prospective studies + 58 RCTs) for the outcomes, and Brown 1999 (67 controlled trials) for the LDL mechanism. Read together they give the honest shape: fibre is a real but modest lever, and the strongest evidence sits on the smallest effect.
(Reynolds et al., 2019) (Brown et al., 1999)
The bottom line
-
The big, impressive numbers are observational. Highest-vs-lowest fibre intake tracks a 15-30% lower risk of all-cause mortality (RR 0.85), CHD (0.76), type 2 diabetes (0.84) and colorectal cancer (0.84) — but this is prospective-cohort data on self-reported intake, so it carries the healthy-user confound and dietary measurement error (Measurement Error in Dietary Assessment). In absolute terms Reynolds puts all-cause mortality at «13 fewer deaths (95% CI eight to 18) … per 1000 participants over the duration of the studies».
-
The RCT-grade effect is real, causal, and small. Where fibre is tested as a dosable, blindable isolate in controlled trials, it moves surrogates: viscous/soluble fibre lowers LDL by -0.057 mmol/L per gram in the practical 2-10 g/d range (Brown), and higher fibre intake lowers bodyweight (-0.37 kg, GRADE High), total cholesterol and blood pressure (Reynolds RCTs). Brown’s own verdict: «The effect is small within the practical range of intake … 3 g soluble fiber from oats … can decrease total and LDL cholesterol by <0.13 mmol/L» and «can make only a small contribution to dietary therapy to lower cholesterol.» (Brown et al., 1999)
-
The per-gram slope is linear-then-plateau, not a first-gram premium
[2026-08-28, challenge #R47]. The -0.057/g figure is not a compression that hides a bigger first gram: Brown deliberately fit it within <=10 g «within the range in which the dose response appeared linear», with a zero intercept, so inside the practical range each gram pays about equally (no concave front-loading). The nonlinearity is at the top — Brown found «significant nonlinearity at higher doses… a biological maximum being reached» above ~10 g (Brown et al., 1999). So 10 g is the knee/plateau onset, not a mid-curve point: below it the slope is real and roughly constant, above it more viscous fibre buys progressively less. Quote the figure with its <=10 g boundary — bare, it reads as linearly extrapolable, which Brown’s own restriction forbids. -
So: eat enough fibre, don’t expect a miracle. It is a genuine supporting lever — worth reaching ~25-30 g/day — but its measured causal effect (on the surrogate we can trial) is modest, and the large mortality numbers should not be read as if they were RCT-proven.
The two legs are DIFFERENT quantities (BLOCKING parameter table — op-weave 2a)
The temptation is to stack Brown’s causal LDL effect under Reynolds’ 15-30% mortality reduction and call fibre powerfully proven. They are not the same claim:
| Parameter | Reynolds 2019 | Brown 1999 | Same quantity? |
|---|---|---|---|
| Design | prospective cohorts (185) + some RCTs | 67 controlled feeding RCTs | NO — observational-led vs RCT |
| Exposure | total dietary fibre (whole-food, self-reported) | soluble/viscous fibre isolate (oat/psyllium/pectin/guar), dosed | NO — whole-diet vs isolate |
| Endpoint | hard outcomes (mortality, T2D, CHD, cancer) | LDL/total cholesterol surrogate | NO — outcome vs surrogate |
| Dose metric | per 8 g/day total fibre | per 1 g/day soluble fibre | NO |
| Effect | all-cause mortality RR 0.85 (hi-vs-lo); 0.93 per 8 g | LDL -0.057 mmol/L per g (practical range) | NO — different scale + endpoint |
| Dose-response shape | «linear with no sign of a plateau» on outcomes | nonlinear, attenuates >8-10 g/d on LDL | NO — opposite shapes, different endpoints |
| Confounding risk | healthy-user + measurement error (observational) | low (isolate held constant vs control) | NO |
Defensible synthesis (type A + F): the RCT leg corroborates the direction of the observational leg and supplies a mechanism (viscous fibre -> lower LDL, independent of fat displacement), but the two do not sum into one big causal number. The composite claim is fibre is beneficial and at least partly causal, with a modest measured effect on the endpoints we can actually trial — the mortality magnitude stays observational.
Why the observational leg is more than just correlation — but not RCT-grade
Reynolds argues causality by triangulation: «The consistency between the trial and prospective study
results, together with the dose-response relationships, provide support that the effect on cardiometabolic
diseases is likely to be causal and not a consequence of confounding variables.»
(Reynolds et al., 2019) That is a legitimate
observational-upgrading move (Upgrading Observational Evidence) — a monotone dose-response plus a
concordant RCT surrogate arm raises confidence above bare correlation. But the hard-outcome arm is still
cohorts of self-reported eaters; symmetric standards forbid reading it as if a mortality RCT had run. Hence
confidence: medium, not high, on the outcome claim — and the causal mechanism is firmest exactly where
the effect is smallest (the LDL surrogate).
The trialled mechanism accounts for only a fraction of the association — the transmission bound [2026-08-17, challenge #R40]
The sharp form of the two-legs split (challenge r40): can the trialled viscous-fibre -> LDL effect transmit to the much larger observational hard-outcome association, or is agreement in direction being mistaken for same-hypothesis triangulation? Brown answers it directly, and the answer bounds the causal share downward. At a practical 3 g/day of soluble fibre, total cholesterol falls «<0.129 mmol/L (5 mg/dL), a <2% reduction. On the basis of estimates from clini-cal studies of cholesterol treatment (110), this could lower the incidence of coronary artery disease by <4%» (Brown et al., 1999) — Brown’s own verdict being that soluble fibre «may exert only a small effect on the risk of heart disease.»
- The order-of-magnitude read (not a same-quantity subtraction). Brown’s <4% is the CHD contribution of a specific practical dose of viscous fibre via LDL; Reynolds’ CHD RR 0.76 is a highest-vs-lowest total-fibre contrast (different exposure, dose and contrast — the parameter table’s all-NO column applies here too, so these are not subtracted). What the bound licenses is only the magnitude comparison: the pathway the wiki can actually trial is, at practical intake, an order of magnitude smaller than the association the cohorts report. Even granting the LDL effect its full cholesterol-treatment transmission, <4% is not ~24%.
- So the honest composite is a slice, not a proof. A small, causal, LDL-mediated slice sits inside a larger observational association whose remainder is unattributed — pattern/substitution, the other fibre fractions Brown’s LDL analysis does not capture (bulk, transit, fermentation), or residual confounding, none of them the trialled mechanism. This quantifies the «causal firmest where the effect is smallest» line, and it is the page’s strongest internal guard against crediting fibre-the-nutrient with the whole hard-outcome pattern. It does not demote fibre to a pure marker (challenge stop condition): the trialled slice is real and causal; what is unresolved is the independent share of the larger association.
The dose-response — a worked case for the CLAUDE.md prior
Fibre is the falsification prior’s cleanest fabric instance, now with a second independent estimate. Reynolds finds the fibre-outcome curves «many of which are linear with no sign of a plateau within the available data», with «the greatest benefits … for individuals consuming 25-29 g per day» and recommends «no less than 25-29 g per day with additional benefits likely to accrue with higher intakes.» (Reynolds et al., 2019)
- This confirms SACN’s reading (SACN’s 30 g marks where confidence intervals widen, not a knee) —
but as a type-F shared-literature corroboration, NOT independent backing: SACN’s dose-response
curves rest on the Aune 2011 / Threapleton 2013 cohort pools, the same cohort literature Reynolds
re-pools, so the MA-CONSTITUENCY test fails E exactly as this page’s own Veronese row rules
[demoted E->F at self-critique 2026-08-08; the ANALYTIC re-derivation is real, the evidence base is shared]. The decision default every reduction pays; the burden is on whoever claims a knee to locate it stands on the shared base, twice-derived. - But the shape is outcome-specific (the gate-6 refinement). Brown’s LDL-surrogate curve does attenuate above ~8-10 g/d — a located knee on a surrogate — while Reynolds’ hard-outcome curves stay linear. So fibre has no plateau is true of the outcomes and false of the LDL marker; the curve you get depends on the endpoint you pick. This is exactly why a plateau on a surrogate must not be read across to the outcome (Surrogate Outcomes).
- The monotone reading is the BENEFIT arm over the studied range, not the net curve. Reynolds’ “additional benefits likely to accrue with higher intakes” is a benefit-arm claim within the sampled data (a Route-2 open-topped direction -> The Underivable Optimum); it is not evidence that the net curve keeps rising indefinitely. At high intakes the net plausibly turns non-monotone through mechanism, not measured outcome: finite gut bulk displaces energy- and nutrient-dense foods, and phytate co-ingested in fibre-rich foods interferes with mineral absorption, with GI tolerance a further practical ceiling. The displacement bites hardest in the strata with least reserve (young children, the malnourished, very-high-phytate diets). Held as directional mechanism only — no held source dose-locates a net-harm threshold, and typical intakes sit far below where it would bind; the point is narrow: benefit-arm-monotone over the studied range must not be read as more-is-always- better on net.
Fibre type, and the food-vs-isolate question
- The viscous mechanism, not the source, carries the LDL effect — oat, psyllium, pectin and guar were each significant and type was not a predictor after adjusting for dose (Brown). So soluble/viscous fibre is the LDL-active fraction; insoluble/cereal fibre works through other routes (bulk, transit, fermentation) that Reynolds’ outcome data capture but Brown’s LDL analysis does not.
- The isolate out-grades the food by DESIGN, not by superiority — the LDL effect’s trial-grade standing (Brown: controlled-trial meta-analysis) exists because soluble fibre can be dosed and controlled; whole-food fibre is far harder to dose and blind, so most of its evidence is observational (though Reynolds’ pooled trials are food-based — the split is dominant, not absolute). This is the Is the Food Category Doing Any Work point in miniature: the better grade tracks the better-trialable form, and does not mean a psyllium supplement beats a bowl of beans for outcomes — the food carries fibre types and a matrix the isolate does not.
- Label grams are not the exposure — added/fortification fibre (challenge r40). A processed product can meet a «high fibre» claim with cheap added bran, cellulose or another legally countable isolate, supplying bulk without the viscosity, fermentability, or whole-food matrix of the cohort exposure — so a fortified product does not inherit Reynolds’ hard-outcome estimate. Do not assume 30 g of analytical/label fibre, 30 g intrinsic food fibre, and an effective dose of viscous or fermentable fibre are interchangeable; that transfer is unevidenced unless the same form is directly tested. This is the Is the Food Category Doing Any Work fortification case applied to fibre; the general label-equivalence gate (does a matching label gram of any nutrient reproduce the studied exposure?) is backlog #R41, not decided here.
- Whole grains likely work largely THROUGH their fibre — Reynolds notes the parallel whole-grain benefits and that fibre’s higher GRADE «could reflect the high fibre content of whole grains» -> Whole Grains Refined Grains and Pulses.
Fibre and IBD risk — a new outcome, and it is subtype-specific [2026-08-09]
A gold-tier SR + dose-response MA (Milajerdi 2021) adds inflammatory bowel disease to fibre’s outcome menu — and the finding is that fibre is not one effect across IBD: protective for Crohn disease but null for ulcerative colitis. Highest-vs-lowest fibre intake: CD «RR: 0.59; 95% CI: 0.46, 0.74; I2 = 0.0%» but UC «RR: 1.09; 95% CI: 0.88, 1.34; I2 = 0.0%» (Milajerdi et al., 2021) — both homogeneous, so the divergence is real, not a pooling artifact. Fibre->CD is nonlinear with «the highest risk reduction … for fiber intake >22 g/d» and a linear «14% reduction in CD risk» per 10 g/d (Milajerdi et al., 2021). Same caveats as the rest of this page — observational, FFQ-measured, healthy-user + food-matrix confounded, and here with reverse causation (preclinical IBD alters intake) acknowledged but unaddressed. So IBD is a low-confidence additional reason consistent with fibre’s other benefits, not a new big lever; the mechanism proposed is the gut-microbiota/SCFA route this page already carries (Valdes). The full facet, and the cross-disease synthesis with the RA risk factors, live on Autoimmune Disease and Modifiable Risk.
Fibre and COPD risk — a new outcome, and it is subtype-specific [2026-08-29]
A gold-tier SR + dose-response MA (Valisoltani 2023, 5 prospective cohorts, n≈213,912) adds chronic obstructive pulmonary disease to fibre’s outcome menu: highest-vs-lowest total fibre RR 0.72 (95% CI 0.64-0.80) for COPD, with cereal (0.76) and fruit (0.75) fibre significant but vegetable fibre null (0.95, CI 0.84-1.07) (Valisoltani et al., 2023). Same caveats as the rest of this page, and one more binding: NutriGrade credibility is LOW, so this is the softest outcome on the menu — a low-confidence additional reason consistent with fibre’s other benefits, not a new lever. It is also the component refinement of the pattern-level Dietary Patterns and COPD Risk signal (fibre isolated out of the umbrella “healthy pattern” bucket). The full estimate, the per-subtype dose-response, the vegetable-arm publication-bias flag, and the type-F parameter table vs the patterns MA live on Dietary Fibre and COPD Risk.
Decision relevance
- Target ~25-30 g/day of total fibre from food, mostly cereal/whole-grain + legumes + fruit/veg; more is fine and probably better (no plateau shown on outcomes). A realistic, sustained increase beats a supplement someone abandons (Free Sugars Intake: carbohydrate quality over quantity gimmicks).
- For LDL specifically, viscous fibre (oats, psyllium) is a legitimate adjunct but a small one — a few tenths of a mmol/L at practical doses, dwarfed by the saturated-fat-replacement and drug levers (Saturated Fat Intake and Replacement). Do not oversell it as cholesterol therapy.
- Rank honestly. Fibre is a real supporting lever, not a big rock; glycaemic index/load is a weaker
marker than fibre or whole-grain content and not worth optimizing for on this evidence.
[2026-08-13: Jenkins 2024 partially contests the "weaker" wording — see the GI/GL section below; the ranking conclusion survives via collinearity, but the honest phrasing is "GI adds little BEYOND fibre/whole grain," not "GI is weaker."]-> Glycaemic Index and Glycaemic Load and Chronic Disease
Certainty and gaps
confidence: medium— the LDL/surrogate leg is controlled-trial-grade (Brown 1999, a pre-GRADE meta-analysis of 67 controlled trials — no formal GRADE rating exists in the held text; an earlier “GRADE moderate-high” label here was unsupported and is RETRACTED); the hard-outcome leg is observational-with-dose-response (Reynolds, GRADE moderate) — upgraded above bare correlation but not to RCT status.- Gaps (G): no RCT of whole-food fibre on hard outcomes (structurally hard — can’t blind food); the
soluble-vs-insoluble and legume-vs-cereal outcome breakdown is too sparse in Reynolds to separate;
fibre’s colonic/transit mechanism is not covered by either source (the constipation / gut-transit
sub-question the Fibre deliverable still routes as unheld).
AWAITSa fibre-transit source. - The very-low/near-zero-fibre referent may not transport (G, challenge r40). Cohort «low fibre» is
low within a normal mixed diet — typically the refined/energy-dense end, so the low-fibre referent is
confounded with high junk-food intake. Whether fibre’s benefit gradient extends down to a minimally
processed near-zero-fibre pattern (a meat-based / carnivore diet) is untested: no held source
offers that like-for-like contrast, so the cohort estimate does not transport to that stratum. Held as
a named gap, not a claim either way — neither bowel-regularity nor anecdote settles long-term outcome
equivalence. The
microbiome mechanism is now partly held: fermentable fibre is the substrate bacteria turn into (Valdes et al., 2018)
short-chain fatty acids, and it is the dominant modifiable lever on microbial diversity — so
prebiotics are largely fermentable fibre by another name, and the prebiotic evidence reduces to the
fibre evidence on this page -> Gut Microbiome and Health
[2026-07-29, Valdes + WGO].
Self-critique [run 2026-07-29, before commit]
- Over-claim check. The page leads with the big numbers are observational and the causal effect is small — it does not let the 15-30% mortality reduction read as RCT-proven; the parameter table’s all-NO column is the guard. No superlative scoped to the vault’s holdings.
- Laundered-E check
[re-run 2026-08-08 — the earlier pass FAILED its own test]. The Reynolds-vs-SACN[E-independent]this section previously defended was itself laundered: SACN’s curves are Aune 2011 / Threapleton 2013 pools, the same cohort literature Reynolds re-pools — shared DATA defeats E regardless of separate analytic teams (the page’s own Veronese row applies the same rule). Demoted to F above; no[E-independent]token remains on this page. Brown and Reynolds are NOT called independent backing for one claim; they are different quantities (the table), classified A+F. - Not-joined / counter-passage. The apparent Reynolds-vs-Brown dose-response clash (linear vs attenuating) is resolved as different endpoints (outcome vs LDL surrogate), read from each source’s own figures — a distinction, not a tension.
- Symmetric standards. The observational leg is held to the confound caveat even though its direction is the conventional/expected one; the causal claim is the authors’, flagged as interpretation.
Appraising this observational evidence — the instrument [2026-07-31]
The mortality/T2D numbers here are observational (Reynolds 2019 cohorts). The instrument that would appraise them is ROBINS-I (Risk of Bias Assessment Tools): domain 1 (confounding — healthy-user) and domain 6 (measurement — Measurement Error in Dietary Assessment) are where the grade would most likely be capped below «comparable to a well-performed randomized trial». A per-domain read is a deliberate future Revisit, flagged there as a re-appraisal candidate, not done here. (inferred from Reynolds et al., 2019)
The umbrella review bounds the outcome breadth — Veronese 2018 [2026-07-31]
Reynolds is one WHO-commissioned SR/MA. Veronese 2018 is an umbrella review of 18 prior meta-analyses (298 prospective observational studies, 21 outcomes) that applied Ioannidis-style credibility diagnostics (excess-significance test, small-study effect, 95% prediction interval, heterogeneity) on top of the pooled estimates. It is a type-F refinement: it does not change the fibre magnitude, it grades which fibre-outcome associations survive bias screening.
The headline tempers the broad reading. «even though 85% of the associations were significant, a higher intake of dietary fibers was convincingly associated only with a decreased likelihood of early mortality and CVD». (Veronese et al., 2018) Of 21 outcomes only 3 reached class-I convincing, and «only CVD and all-cause mortality were based on prospective studies» (the third, pancreatic cancer, rested on 13 case-control + 1 prospective study). (Veronese et al., 2018)
- The robust core is mortality + CVD, not the long cancer list. T2D is class-III suggestive; most cancer associations are class III/IV and «largely based on case-control studies that suffer inherent limitations including recall bias and inability to examine temporal associations». (Veronese et al., 2018) Breast cancer and coronary artery disease carried outright excess-significance bias. Prostate cancer was non-significant.
- A located RCT-null on colorectal adenoma — the observational fibre-cancer signal is not confirmed where it has been trialled: «our findings are in agreement with a Cochrane review of randomized and quasi-randomized controlled trials in which the authors found that increased dietary fiber intake did not reduce the incidence or recurrence of adenomatous polyps in ∼5000 participants». (Veronese et al., 2018) So the colorectal arm is exactly the Surrogate Outcomes hazard: an observational association whose one RCT test on the precancerous lesion came back null.
The umbrella does NOT out-rank Reynolds (correcting the hierarchy intuition)
An umbrella review sits above a single SR in the tidy pyramid, but Veronese excluded RCTs — «meta-analyses including data from randomized controlled trials» were an exclusion criterion. (Veronese et al., 2018) So on the causal axis it is weaker than Reynolds (which carries 58 RCTs on the surrogates); its authors concede «future randomized controlled trials in large sample sizes are needed to confirm these observational findings». (Veronese et al., 2018) Its value is the bias-diagnostic + breadth layer, not a higher grade.
Parameter table — Veronese vs Reynolds (BLOCKING, op-weave 2a)
The temptation is to read Veronese’s near-identical mortality RR as an independent confirmation of Reynolds. It is not independent — both pool the same underlying prospective cohort literature (Veronese’s all-cause-mortality MA is Yang 2015; its CVD MA is Threapleton 2013 — the very cohorts Reynolds’ own meta-analysis re-pools). Shared primary studies = shared-evidence agreement, not type-E independent backing. (Reynolds et al., 2019; inferred from Veronese et al., 2018)
| Parameter | Veronese 2018 | Reynolds 2019 | Same quantity? |
|---|---|---|---|
| Design | umbrella of 18 observational MAs (298 cohorts), RCTs excluded | own SR/MA: 185 prospective cohorts + 58 RCTs | NO — observational-only vs SR-with-RCT arm |
| All-cause mortality | RR 0.835 (0.797-0.875), highest-vs-lowest | RR 0.85 highest-vs-lowest | ~SAME direction+magnitude, but shared cohorts |
| CVD / CHD | RR 0.913 (0.893-0.932) per 7 g/d | CHD RR 0.76 hi-vs-lo; 0.81 (0.73-0.90) per 8 g/d | NO — 7g vs 8g increment, different contrast, and the slopes visibly differ |
| Evidence appraisal | Ioannidis credibility class I-IV + excess-sig/small-study/PI | GRADE certainty | NO — different instruments |
| Causal stance | needs future RCTs to confirm (no RCT arm; authors’ own caveat) | likely-causal by trial + cohort triangulation | NO — Reynolds triangulates an RCT arm |
| Independence | pools same cohort literature Reynolds uses | same underlying cohorts | NOT independent -> F/shared, not E |
Defensible synthesis (type F): Veronese corroborates the mortality/CVD magnitude from the same
observational base and adds a bias-diagnostic grading that Reynolds did not run — confirming the
robust core (mortality, CVD) while flagging the cancer periphery as weak and case-control-driven. It
raises confidence in which outcomes to believe, not in the causal status (still observational,
still confidence: medium). It says nothing about dose-response shape (it used highest-vs-lowest
contrasts, not per-increment curves), so it neither confirms nor tempers the no-plateau reading.
A grade clash on fibre -> colorectal cancer — WCRF probable, Veronese weak [2026-08-05]
WCRF’s Third Expert Report grades dietary fibre a probable (strong-enough-to-recommend) protector against colorectal cancer — «Consuming dietary fibre helps protect against colorectal cancer» — and sets the plant-food recommendation’s fibre goal at «at least 30 grams per day … from food sources» (AOAC method), the same ~25-30 g target this page reaches on the mortality/CVD axis. (World Cancer Research Fund & American Institute for Cancer Research, 2018) That directly clashes with the Veronese reading above (colorectal-fibre signal weak, class III/IV, and the one RCT on colorectal adenoma came back null).
| Parameter | WCRF 2018 | Veronese 2018 | Same quantity? |
|---|---|---|---|
| Relationship | dietary fibre -> colorectal cancer | dietary fibre -> colorectal cancer (+ the adenoma RCT) | ~YES (adenoma is the precancerous lesion, not cancer) |
| Grade | probable (strong) — recommend | class III/IV weak; adenoma RCT null | NO — the crux |
| Instrument | WCRF/AICR (Bradford-Hill: cohort dose-response + mechanism) | Ioannidis credibility class + excess-significance, RCTs excluded | NO |
| Evidence base | prospective cohorts (CUP SLRs) | largely the same cohorts + case-control | ~shared |
This is the meat pattern again (Should Adults Reduce Red and Processed Meat): the same observational colorectal-cancer literature, graded strong-enough-to-act by a precautionary cancer-prevention body and weak under a strict credibility-class lens — a grading-and-standpoint disagreement, not a dispute about the associations -> Certainty of Evidence vs Strength of Recommendation, Diet Physical Activity and Cancer Prevention. The wiki’s read is unchanged: the fibre cancer arm is the softest of the outcome menu (the robust core is mortality + CVD, the causal firmest where the effect is smallest — the LDL surrogate), and WCRF’s “probable” does not upgrade it to causal here. (Veronese et al., 2018; inferred from World Cancer Research Fund & American Institute for Cancer Research, 2018)
Self-critique [run 2026-07-31, before commit — Veronese section]
- Over-claim. The section leads with the bounding reading (only mortality/CVD convincing) and explicitly denies the umbrella out-ranks Reynolds. No superlative scoped to the vault’s holdings.
- Laundered-E. Independence is denied, not claimed — the parameter table’s final row and the
shared-cohort note (Yang 2015 / Threapleton 2013) are the guard. No
[E-independent]token added. - Not-joined / counter-passage. No tension filed: Veronese concludes fibre intake should be promoted and «our results support dietary recommendations that promote a higher fiber intake» — same direction as Reynolds. The apparent friction is on certainty/breadth, resolved as an F-refinement, not an opposed claim (checked against Veronese’s own Discussion + Conclusion).
- Symmetric standards. The mortality corroboration (conventional direction) is held to the same shared-cohort / observational caveat as any contested claim.
GI/GL sits ALONGSIDE fibre as a collinear carb-quality marker — Jenkins 2024 [2026-08-13]
Jenkins 2024 (gold MA of 10 mega-cohorts) pooled GI/GL against the four main outcomes and compared them to fibre and whole grains in the same cohorts: «Associations between diets high in fibre and whole grains and the four main outcomes were similar to those for low GI diets.» (Jenkins et al., 2024) The authors read this as validating GI as an independent carbohydrate-quality predictor; the wiki reads it as collinearity — «foods high in carbohydrates also tend to have a high GI», and «only 12 (17%) of the 69 GI main outcome assessments controlled for fibre». (Jenkins et al., 2024) Similar magnitude among markers that ride together, in shared cohorts with minimal mutual adjustment, cannot attribute the effect to GI rather than the fibre/whole-grain pattern that carries it.
So this does not overturn this page’s ranking — it re-words it. GI/GL is not weaker than fibre on the numbers; it adds little beyond fibre/whole grain because it is a proxy for the same pattern. Steering by fibre and whole-grain content remains the honest lever. Full magnitudes, the guidance divergence (WHO 2019 dismissed GI; Jenkins rebuts), and the measurement-error caveat live on Glycaemic Index and Glycaemic Load and Chronic Disease. (inferred from Jenkins et al., 2024)
Fruit and vegetables are a fibre CARRIER, not clean fibre evidence [2026-08-13]
F&V benefit (Aune 2017: all-cause 0.90, CVD 0.92, CHD 0.92 per 200 g/day) is a recurring reason people credit fibre, but the source itself blocks that inference: F&V acts through «a myriad of nutrients and phytochemicals, including fibre, vitamin C, carotenoids, antioxidants, potassium, flavonoids and other unidentified compounds which are likely to act synergistically» (Aune et al., 2017) — fibre is one of several active fractions, not the isolated agent.
- The component-not-category trap, stated by the author. A whole-food F&V association cannot be attributed to any single constituent; Aune cites the fibre-CVD MAs (Threapleton 2013) only as part of the mechanism, alongside potassium/BP, flavonoid/vascular and antioxidant/DNA pathways. So F&V outcome data corroborate the direction of the fibre story without isolating it — the observed-healthy-population rule -> Fruit and Vegetable Intake and Health.
- Where fibre is isolable (dosed, blindable isolates — Brown’s cholesterol MA; Reynolds’ intake gradients) the evidence is cleaner than any F&V contrast can be; that design advantage, not the food, is why isolate evidence out-grades F&V-carrier evidence -> Is the Food Category Doing Any Work.
Self-critique [run 2026-08-17 — challenge #R40 weave]
- Adjudication was covered-with-mechanism, not manufactured divergence. r40’s central charge (does the fabric over-attribute the hard-outcome pattern to fibre-the-nutrient?) was already answered by this page before the challenge — the all-NO parameter table, the carrier-not-component Aune section, the isolate-out-grades-by-design point, the colorectal-adenoma RCT null, and the study-edge (not knee) target. The honest finding is the fabric already resists the over-attribution; the challenge is logged upheld-but-largely-already-held, not as a refutation of the page. Reporting the convergence is the symmetric-standards move (a confirmed-convention result is as reportable as a divergence).
- Only source-supported increments were woven. The transmission bound is Brown’s own <4% CHD figure (cite.py-emitted, cold-audited); the order-of-magnitude comparison to Reynolds’ ~24% is flagged and explicitly not a same-quantity subtraction (guarding the exact error the parameter table exists to prevent). The fortification/label point applies held logic and is marked with the general gate routed to r41 (not decided here — no scope creep into r41). The near-zero-fibre item asserts an absence, not an effect.
- No over-claim, no laundered-E. No superlative scoped to the vault; no independence claimed (Brown
and Reynolds stay different quantities, A+F). Confidence stays
medium— the weave sharpens why the causal share is bounded, it does not add certainty to the outcome leg.