The fabric’s microbiome nucleus. The anchors are not the same tier — read each accordingly: WGO 2023, a Global Guideline (gold), carries the probiotics/prebiotics evidence by indication; Minkoff 2023, a Cochrane systematic review + meta-analysis of RCTs (gold), now carries the one clean causal win and its pooled magnitude, with van Nood 2013, the landmark FMT-for-recurrent-C. difficile RCT (high), demoted to the largest-effect trial within that pool (superseded 2026-09-16 — the FMT leg previously rested on the single van Nood RCT); Valdes 2018, a BMJ narrative review / evidence-map (moderate tier), supplies the causation problem and the field overview. Valdes is the weaker witness — a non-systematic overview, not a pooled or graded synthesis — so its role here is deliberately confined to a methodological caution (composition-shift is not an outcome) and to the field bound on FMT (not yet for other pathologies), never to carry the FMT-CDI magnitude, which now rests on the held RCT. The through-line: the microbiome is genuinely modifiable and matters in specific places, but “it altered the microbiome” is a surrogate, and most popular interventions are either strain-and-indication-specific or not-yet-evidenced.
The load-bearing epistemics: composition-shift is not an outcome, and causation is the central confound
Almost all microbiome research reports a composition change (which bacteria, how diverse). Valdes states the limit plainly: «observational studies can show cross-sectional associations between microbes and health traits but are limited by the inability to measure causal relations. The strongest level of evidence is obtained from interventional clinical studies — in particular, randomised controlled trials.» (Valdes et al., 2018) So low diversity is a marker of dysbiosis, not a demonstrated cause of disease, and the direction of the arrow (does dysbiosis cause the illness, or the illness/diet cause the dysbiosis?) is usually unresolved. Treat a composition-shift as a surrogate that earns belief only when followed through to an outcome (Surrogate Outcomes). The template done right is Non-Sugar Sweeteners (Suez 2022): it did not stop at a microbiome shift — it followed through to glucose tolerance and proved cause by transplanting the bacteria into germ-free mice.
What actually moves the microbiome: diet, and mostly fibre
The dominant modifiable lever is diet, and within diet, fermentable fibre: gut bacteria ferment it into
short-chain fatty acids (acetate, propionate, butyrate). The diet-diversity link is real but NOT a clean
“more fibre -> more diversity”: Valdes reports that long-term weight gain «correlates with low microbiota
diversity, and this association is exacerbated by low dietary fibre intake» — i.e. low fibre worsens the
weight-gain/low-diversity association, not that fibre raises diversity directly [the earlier reading of this quote swapped its subject — corrected 2026-08-08]. Valdes is in fact two-sided on the intervention
direction: «Many short term feeding trials with purified dietary fibres or even whole plant based diets
either have no effect on microbiota diversity or reduce it, but can still have clinical benefits.»
(Valdes et al., 2018) So the microbiome story still
points to a fibre-rich, plant-diverse diet (Dietary Fibre and Health) — but via SCFA/clinical benefit,
not via a simple diversity gradient. It also reframes prebiotics — WGO defines a prebiotic as
«Prebiotic A selectively fermented ingredient that results in specific changes in the composition and/or
activity of the gastrointestinal microbiota, thus conferring benefit(s) upon host health»
(World Gastroenterology Organisation et al., 2023) — mostly fermentable fibre by
another name, so the prebiotic evidence largely is the fibre evidence. (The «selectively utilized by host
microorganisms that confers a health benefit» wording used earlier is WGO’s synbiotic definition, not the
prebiotic one — corrected 2026-08-08.)
Refinement — fermented foods moved diversity where fibre did NOT, head-to-head [2026-08-05, Wastyk]
Valdes’s claim that microbial diversity «tracks fibre intake» is a long-term observational association. The Wastyk 2021 RCT Fermented Foods and Health supplies a short-term interventional test that refines it: over a 10-week head-to-head, the high-fermented-foods arm steadily raised alpha diversity while in the high-fibre arm alpha diversity «did not change cohort-wide over the course of the intervention» (Wastyk et al., 2021) (corrected 2026-10-06: non-verbatim span re-lifted via cite.py).
This is not a contradiction of Valdes (different scope/horizon — a long-term association vs a 10-week intervention; the not-joined check fires on horizon, so it is a distinction). Wastyk’s own reading is that fibre’s diversity effect likely needs longer than 6 weeks of high intake plus a source of new microbes, not that fibre fails. What it does bound is the reduction “the microbiome story reduces to the fibre story”: on the diversity marker over an actionable timescale, fermented foods were the more reliable lever — an indirect remodeling of the resident community, not colonization by eaten microbes. The prebiotic (fibre-feeds-resident) and probiotic (fermented-adds-microbes) routes are genuinely distinct, and both are surrogate-level here. (inferred from Wastyk et al., 2021)
Probiotics: strain- and indication-specific, NOT a general tonic
The WGO guideline’s central discipline is that «the effects of probiotics are strain-specific and dose-specific» and «the most robust approach to probiotic evidence is to link benefits … to specific strains or strain combinations of probiotics at the effective dose». (World Gastroenterology Organisation et al., 2023) So “take a probiotic for gut health” is not a claim the evidence supports; the evidence attaches to specific strain x indication x dose triples (WGO Tables 8-9). Where the evidence is real:
| Indication | WGO evidence status | Note |
|---|---|---|
| Necrotizing enterocolitis (preterm neonates) | benefit — reduced NEC + reduced death, NNT 20 | «not all probiotic preparations tested are effective»; moderate certainty |
| Antibiotic-associated diarrhoea | benefit (specific strains) | a real prevention win |
| C. difficile-associated diarrhoea (prevention) | benefit (specific strains) | prevention, distinct from FMT treatment below |
| Acute infectious diarrhoea | benefit (specific strains) | modest, strain-specific |
| IBS | symptom benefit for certain strains | «strain-specific effects … on IBS symptoms» |
| Pouchitis | benefit (specific preparations) | — |
| H. pylori eradication (adjunct) | reduces therapy side-effects | «quality of the evidence was weak» |
- The NEC finding is the most striking patient-important outcome: «The number needed to treat to prevent one death from all causes by treatment with probiotics is 20» in preterm neonates — but «not all probiotic preparations tested are effective», so it is a strain-quality claim, not a class claim. (World Gastroenterology Organisation et al., 2023)
- Meta-analysis caveat (a method point that travels): WGO warns that pooling across different strains without a shared-mechanism rationale «should be avoided» — a strain-blind probiotic meta-analysis can manufacture or wash out an effect. This is the Is the Food Category Doing Any Work problem at the organism level: the strain is the exposure, not the word “probiotic”.
- General health / immunity / mood — NOT unstudied: both held sources report evidence-bearing states. WGO finds «suggestive evidence that several probiotic strains and the prebiotic oligofructose are useful in improving the immune response» (World Gastroenterology Organisation et al., 2023); Valdes reports meta-analytic benefit for URTI, eczema prevention and depressive symptoms. The honest state is weak-to-suggestive benefit, heavily discounted for strain-blind pooling and heterogeneity (Valdes: the studies «were not homogeneous … which limits precise recommendations») — a demotion the wiki makes,, NOT the insufficient-evidence state (the earlier “insufficient evidence, not a finding” mis-stated the evidence-state and cut in the deflationary direction — corrected 2026-08-08). And swallowed probiotics are typically transient colonisers.
FMT: the field’s strongest win, and only for one thing so far
Faecal microbiota transplant is the clearest demonstration that the microbiome is causal in humans — and the fabric now holds the pooled gold evidence, not a single trial. The Minkoff 2023 Cochrane systematic review + meta-analysis pooled six RCTs (320 immunocompetent adults with recurrent CDI) against control (mainly antibiotics — vancomycin/fidaxomicin — or placebo): FMT «likely leads to a large increase in resolution of rCDI in the FMT group compared to control (risk ratio (RR) 1.92, 95% confidence interval (CI) 1.36 to 2.71; P = 0.02, I2 = 63%; 6 studies, 320 participants; number needed to treat for an additional beneficial outcome (NNTB) 3; moderate-certainty evidence)». (Minkoff et al., 2023) In absolute terms that is resolution rising from 401 per 1000 (~40%) with control to 770 per 1000 (~77%, 95% CI 545 to 1000) with FMT — a ~37 percentage-point absolute gain. (Minkoff et al., 2023) Safety pooled toward FMT but imprecisely: serious adverse events RR 0.73 (95% CI 0.38 to 1.41; moderate certainty), all-cause mortality RR 0.57 (95% CI 0.22 to 1.45; low certainty — few events, an increased risk «could not be ruled out»); no trial reported colectomy. (Minkoff et al., 2023) Certainty is moderate, not high (heterogeneity I2 = 63%, risk of bias) — so the decision-relevant claim is large, consistent-in-direction benefit at moderate certainty, not a precise effect size.
The landmark trial sits inside this pool, and the pool bounds its over-estimate. In the van Nood 2013 open-label RCT — one of Minkoff’s six included studies — donor-feces infusion cured 13/16 (81%) on the first infusion and 15/16 (94%) overall, versus 4/13 (31%) for standard 14-day vancomycin and 3/13 (23%) for vancomycin + bowel lavage; cure rate ratio 3.05-4.05, so large that «The study was stopped after an interim analysis» under the Haybittle-Peto rule. (van Nood et al., 2013) The wiki flagged the tiny early-stopped n (99.9% CI upper bounds ~290) as carrying the usual early-stopping upward bias — and that is exactly what the meta-analysis now shows: the pooled RR of 1.92 is materially below van Nood’s within-trial ratio, so the single trial did over-state the magnitude while getting the direction right. This is the F (quality-upgrade) move: Minkoff pools van Nood among the six, so their agreement is shared-source, not independent (F + re-nucleus, never type-E), and the composite is the better anchor. (inferred from Minkoff et al., 2023; van Nood et al., 2013)
Why this is the surrogate done right, not a bare composition claim: the same trial measured the microbiota shift AND followed it to the clinical outcome — Simpson’s diversity rose from ~57 to ~179 (into the donor range) as patients were cured, and «The mechanism underlying the efficacy of donor-feces infusion is probably the reestablishment of the normal microbiota as a host defense against C. difficile.» (van Nood et al., 2013)
The scope boundary is the decision-relevant part. CDI is the case where the pathology is the dysbiosis — antibiotics collapse diversity and FMT directly restores it — which is exactly why it is both the cleanest causal win and a poor template for generalization. It is not evidence that manipulating the microbiome helps general health, weight, or metabolic disease, where no established dysbiosis-as-cause exists for a transplant to reverse. Valdes states the field bound directly: «For other pathologies, faecal transplants are not yet clinical practice but have been explored.» (Valdes et al., 2018) So FMT-for-recurrent-C. difficile is at once the anchor of what real microbiome evidence looks like and a warning against reading the one clean win as a general microbiome mandate. (Valdes et al., 2018; inferred from van Nood et al., 2013)
Decision relevance
- The big microbiome lever is diet — specifically fibre/plant diversity — and it is already the fibre recommendation (Dietary Fibre and Health); you do not need a supplement aisle to act on it.
- The Low-FODMAP Diet for IBS runs this lever in reverse — a deliberate, time-limited cost. Restricting fermentable carbohydrates (a fibre subset) relieves IBS symptoms short-term but reduces bifidobacteria/actinobacteria by starving them of substrate — the same fibre-feeds-resident mechanism, removed on purpose. This is why full long-term FODMAP restriction is not a neutral diet: it sacrifices the prebiotic benefit.
- Probiotics: match a specific strain to a specific indication, or don’t bother. They are a real tool for AAD, C. difficile prevention, NEC in preterm infants, pouchitis, acute infectious diarrhoea, and some IBS — at named strains and doses. As a general “gut health” tonic they are unsupported.
- FMT is medicine for recurrent C. difficile (gold Cochrane MA of 6 RCTs, pooled RR 1.92 / NNTB 3 / moderate certainty), not a wellness intervention — the one clean microbiome win, and bounded to the indication where the pathology is itself the dysbiosis.
- Rank the topic LOW relative to its attention. The microbiome is discussed far out of proportion to its established, outcome-level effect sizes — the telos’s attention-is-an-anti-signal rule (Layer 1 - Ranking Interventions for a Stratum). The real levers here are the ones already ranked elsewhere (fibre, diet quality).
Certainty and gaps
confidence: medium— the grade rests on WGO (gold guideline) for the probiotics-by-indication claims (citing RCT/meta-analytic evidence) plus the Minkoff 2023 Cochrane MA (gold) for the FMT-CDI leg, which this ingest upgraded from the single early-stopped van Nood 2013 RCT (high) to the pooled six-trial estimate; it is NOT carried by Valdes (a moderate-tier BMJ narrative review), whose role is confined to the surrogate caution and the field bound. The FMT leg is now the fabric’s best-evidenced microbiome-causation claim, but it is pooled at moderate certainty (I2 = 63%), and the general microbiome-health story remains mostly observational (Valdes’s own caution) — somedium, not higher, and it survives stripping the Valdes leg.- Independence note: Valdes (BMJ review) and WGO (guideline) agree on strain-specificity and the FMT-CDI
win, but both partly rest on the same primary literature — consistent corroboration, not strict
[E-independent]backing. And Minkoff pools van Nood among its six RCTs, so the two agree by shared data and shared authors within the pool, not independently — Minkoff is a refinement / quality-upgrade (F) superseding the single-trial anchor, never independent (E) backing of it. - Gaps (G): the FMT-CDI leg is now anchored on the held Minkoff 2023 Cochrane MA — this ingest
closed the previously-named “unheld Cochrane FMT-CDI SR” gap (it pooled the effect the single
early-stopped van Nood trial over-estimated, confirming the wiki’s early-stopping-bias caveat). Still
unheld: the AGA probiotics guideline, and Camilleri 2019 (the intestinal-permeability / “leaky gut”
review that would let the fabric sort real permeability from the syndrome).
AWAITSthose. Direction-of-causation for dysbiosis-disease associations is the field’s open confound.
Self-critique [run 2026-07-29, before commit]
- Composition-vs-outcome discipline held. The page leads with the surrogate caveat and never credits a bare composition-shift with a health benefit; the probiotics wins are tied to named indications with WGO’s own certainty language, not asserted as class effects.
- Over-claim check. NEC NNT 20 is paired with «not all preparations … effective»; the H. pylori win is paired with «evidence was weak»; general-health probiotic claims are placed in insufficient-evidence, not benefit. No superlative scoped to the vault.
- Independence not laundered — Valdes+WGO corroboration is flagged as sharing primary literature, so no
[E-independent]is claimed. - Demarcation deferred honestly — “leaky gut” / candida sorting needs Camilleri (unheld); named as a gap rather than adjudicated from the overview.
Self-critique [re-run 2026-07-31, tilt-audit relabel, before commit]
- Tier mislabel fixed (the tilt). The header previously called both anchors gold-standard,
which over-graded Valdes — a moderate-tier BMJ narrative review / evidence-map, not a pooled or
GRADE-graded synthesis. Corrected to tier each source explicitly (WGO gold; Valdes moderate) and to
confine Valdes’s load-bearing role to a methodological caution (composition-shift is not an outcome)
and to established clinical facts (FMT-for-recurrent-C. difficile), never to a contested magnitude.
Verified against the registry tier (
valdes2018=moderate/narrative-review). - Confidence re-checked, held at
medium. The decision-relevant probiotics/FMT claims rest on WGO (gold) and on FMT-CDI being routine practice — not on Valdes — somediumis earned by the gold leg and survives stripping the Valdes leg. Not demoted (the fix was the label, not the underlying grade); not inflated (the general microbiome-health framing stays hedged as mostly observational). - No new source-attributed claim. This pass changed the wiki’s own tier characterization, not any
[EXTRACTED]quote or attribution; the Valdes/WGO quotes and their loci are untouched.
Self-critique [re-run 2026-08-05, van Nood FMT-RCT rewove, before commit]
- FMT leg de-borrowed. The FMT-CDI magnitude now rests on the held van Nood 2013 RCT, not on Valdes’s narrative-review assertion; blind cold-audit returned 10/10 SOUND on the numbers and both quotes. The early-stopping upward bias is stated (not buried), and the ~63pp/NNT~2 figures are marked arithmetic, not attributed to the paper.
- F, not E. van Nood is the primary source under Valdes’s FMT summary — a refinement replacing a
secondhand claim, not independent backing. No
[E-independent]claimed; the Independence note says so. - Scope boundary sharpened, not overclaimed. The one clean win is explicitly bounded to CDI (pathology is the dysbiosis) and flagged as a non-template for general-health/metabolic FMT — reinforcing, not contradicting, the page’s surrogate-vs-outcome thesis (the trial measured diversity AND cure together).
- Confidence held at
medium. A high-tier RCT strengthened one leg, but the general microbiome-health story stays mostly observational; the grade survives stripping the van Nood/WGO legs (Valdes cannot carry it).
Appraising this observational evidence — the instrument [2026-07-31]
The composition-shift → outcome cohorts here are observational; ROBINS-I (Risk of Bias Assessment Tools) is the appraisal instrument, with domain 1 (confounding / reverse causation) and domain 6 (measurement of a surrogate) the likely caps. Named as a re-appraisal candidate on the RoB-tools page; not re-graded here.