Nucleus of the white-meat cluster. The poultry -> patient-important-outcome cell, newly opened.
Until now the fabric held poultry only as the substitute arm inside red-meat substitution analyses;
this is its first first-hand exposure->outcome evidence. Three gold meta-analyses of observational
cohorts now held — Lupoli (all-cause + CV mortality/events), Kim (stroke incidence + mortality), and
Ramel (CVD mortality + T2D incidence, NNR2023 SR group). Across all three the picture is the same:
poultry is neither a protective nor a harmful food in its own right — a small favourable signal on
aggregate endpoints (all-cause mortality, stroke incidence) that is inseparable from red-meat
displacement, and NULL on the cause-specific CV and T2D endpoints. Ramel closes the CVD-mortality and
T2D cells and, distinctively, attaches a formal WCRF certainty grade to the null (below). A fourth
source — one large US cohort (Etemadi 2017, NIH-AARP) — adds the piece the three MAs structurally lack:
a within-cohort substitution estimate (red -> white), the swap the intake-only MAs could not compute.
(this lead is the wiki’s cross-source framing; each finding is attributed on its own section
below)
What “white meat” is here — and why the exclusion of fish matters (the B-distinction)
The exposure is poultry (chicken, turkey, duck, goose) and rabbit — EXCLUDING fish. Lupoli
«included studies defining “white meat” as poultry (chicken, turkey, duck and goose) and rabbit» and
deliberately dropped a fish-combined study (Sinha) and a fried-poultry-only study (Sun), because it
«better defined white meat by excluding studies on fish, due to the different health impact of fish and
poultry consumption, which may have confounded prior analyses». This is a decision-relevant
terminological cut: an earlier MA (Abete 2014) mixed fish into “white meat,” and fish carries its own
distinct effect profile -> Fish and Seafood Consumption. A finding attached to the label “white
meat” does not transport unless the label excludes fish. type-B
(Lupoli et al., 2021)
The effect estimate
effect_measure: all-cause mortality OR 0.94 (95% CI 0.90, 0.97; p<0.001), highest vs lowest
intake (a 6% lower rate). CV mortality OR 0.95 (0.89, 1.01; p=0.13) — null.
Non-fatal CV events OR 0.99 (0.95, 1.02; p=0.48) — null.
population_and_comparator: general adult cohorts, 22 prospective studies, 3,132,149 subjects;
highest vs LOWEST self-reported white-meat category (not a defined dose).
outcome: all-cause mortality (patient-important); CV mortality + non-fatal CV events.
dose_response_shape: NONE estimated — categorical highest-vs-lowest only; exposure metrics
heterogeneous across studies (servings/day, servings/week, g/day). No knee,
plateau, or threshold locatable.
uncertainty: very high heterogeneity (all-cause I2 95.6%, not reduced by leave-one-out);
observational only; residual confounding acknowledged.
effect_modifiers: none found — meta-regression showed no impact of age, male gender, BMI,
hypertension, diabetes, prior CV events, smoking, or follow-up on the
all-cause estimate (a negative route-(b) result).
certainty: the source gives no GRADE rating; observational + high-heterogeneity +
residual-confounding put this at LOW certainty on the wiki's reading.
confidence: low
- The all-cause signal is small and, in absolute terms, unquantified here — a 6% relative reduction on a highest vs lowest contrast whose exposure separation differs study to study, so it cannot be read as a per-serving effect or converted to an absolute risk difference without the baseline and the contrast width. State it as relative, on an undefined contrast, not as a dose.
- The two CV endpoints are null, and shape is outcome-specific: an all-cause benefit sits beside a neutral CV mortality and a neutral CV-events result in the same analysis. That the all-cause effect is not carried by a matching CV-mortality effect is itself a caution about the 6%. (Lupoli et al., 2021)
The comparator problem is the load-bearing caveat
The all-cause association is inseparable from what poultry displaces. Lupoli states it plainly: «The interpretation of the effects of white meat consumption on health is a difficult task, as subjects consuming more white meat are, at the same time, consuming less red meat. On the other hand, people with a low red meat intake may prefer others protein sources, such as proteins from vegetable origin that could have per se a beneficial impact on cardiovascular health.» So a high-poultry diet is also a low-red-meat diet and often a more-plant-protein diet, and the 6% could be crediting poultry for the removal of red meat or the addition of plant protein -> The Comparator Problem. The author’s own reframe — «white meat… is a source of high-quality proteins and may therefore fully substitute red meat» — is a substitution claim, not evidence that poultry per se lowers mortality. (Lupoli et al., 2021)
Mechanism — directional, not an outcome finding
(inferred from Lupoli et al., 2021) Marked as mechanism, human-corroborated but discounted: poultry vs red meat carries lower fat, a more favourable (mostly unsaturated) fatty-acid profile, and lower heme iron, with SFA and heme iron named as atherosclerosis-promoters; processed-meat preservatives (sodium, nitrates) promote hypertension, insulin resistance, and endothelial dysfunction. This gives a direction (poultry less harmful than red/processed meat on a CV pathway) but the CV endpoints here were null, so the mechanism is not confirmed by this outcome evidence — hold it as directional support for the substitution reading, not as a demonstrated poultry benefit.
Robustness and what does not move it
- Excluding low-quality studies (NOS < 6) confirmed all three estimates (all-cause 0.95 [0.91, 0.99]).
- No publication bias for all-cause/CV mortality (Egger p = 0.71, 0.85); significant bias for CV events (Egger p < 0.001) but trim-and-fill held the null.
- Geographic subgroups for all-cause were directionally consistent but the Asian subset was both significant and low-heterogeneity (0.94 [0.90, 0.98], I2 12.9%) while America carried near-total heterogeneity (I2 98.5%) and Europe was non-significant. (Lupoli et al., 2021)
The stroke endpoint (Kim 2017) — white meat favourable, on a thin base
Kim pooled prospective cohorts on meat -> stroke, keeping the SAME fish-excluding definition of white meat («white meat: poultry meat only (fish excluded)»), so the exposure is comparable to Lupoli’s. For white-meat -> stroke incidence the pooled RR is 0.87 (95% CI 0.78-0.96), I2=0.00 — a ~13% relative reduction — but on a much thinner base than Lupoli’s all-cause estimate: the white-meat pool rests on «4 comparisons (138 761 participants)» from just 2 articles, versus Lupoli’s 22 studies / 3.1M subjects. No heterogeneity (I2=0), but only two source articles and a highest-vs-lowest categorical contrast — no dose-response, no absolute risk difference, same categorical-exposure limitation as Lupoli. (Kim et al., 2017)
- Stroke mortality for white meat is UNESTIMABLE here — Kim «were not able to perform analyses on the associations between processed and white meat consumption and stroke mortality because of limitation of data». So the favourable signal is on stroke incidence only; the cause-specific fatal endpoint is a data gap, not a null. (Kim et al., 2017)
- The measurement-error caveat is stated at source: self-reported meat consumption «may have led to a misclassification of the different types of meat caused by a measurement error», and the high-vs-low cutpoints — «quantity of meat intake dividing high versus low consumption groups were not entirely consistent in the studies included in this meta-analysis» — so the categorical contrast is the same self-report attenuation that binds every dietary dose-response -> Measurement Error in Dietary Assessment. (Kim et al., 2017)
The within-Kim contrast is the decision-relevant move (a substitution signal)
In the SAME analysis, same design and pooling, the meat types split by direction:
| Meat type (Kim, highest vs lowest) | Stroke incidence RR (95% CI) | I2 | Direction |
|---|---|---|---|
| Total meat | 1.18 (1.09-1.28) | 0.00 | increase |
| Red meat | 1.11 (1.03-1.20) | 0.00 | increase |
| Processed meat | 1.17 (1.08-1.25) | 0.00 | increase |
| White meat | 0.87 (0.78-0.96) | 0.00 | decrease |
Kim states the split directly — total, red and processed «meat intake is associated with an increase of stroke incidence, whereas white meat (RR, 0.87; 95% CI, 0.78–0.96 [I2=0.00]) consumption is related to a reduction of stroke incidence.» (Total and red meat -> stroke mortality were null — 0.97 [0.85-1.11] and 0.87 [0.64-1.18] — so the incidence signal does not carry to fatal stroke for the red/total arms.) (Kim et al., 2017)
Kim draws the substitution recommendation explicitly: «Individuals who are at a higher risk of stroke who habitually consume red and processed meats should consider substituting a source of their protein intake to white meat.» This is the substitution frame the whole cluster turns on — the poultry benefit is a contrast against red/processed meat within one analysis, not proof that poultry per se protects, exactly as the comparator caveat above warns -> The Comparator Problem, Should Adults Reduce Red and Processed Meat. (Kim et al., 2017)
The CVD-mortality and T2D endpoints (Ramel 2023) — null, and WCRF-graded
Ramel is the NNR2023 SR group’s SR+MA of white meat -> CVD and T2D: 26 studies (23 prospective cohorts + 3 RCTs), same fish-excluding poultry exposure. It fills the two cells the cluster still had open — CVD mortality and T2D incidence — and both land NULL:
effect_measure: CVD mortality RR 0.95 (95% CI 0.87-1.02, P=0.23), I2=25% (low heterogeneity),
6 cohorts pooled, highest-vs-lowest categorical intake.
T2D incidence RR 0.98 (95% CI 0.87-1.11, P=0.81), I2=82% (HIGH heterogeneity),
7 of 9 available cohorts pooled (2 dropped for not reporting OR by extremes).
population_and_comparator: general adult cohorts (Europe/Asia/USA), 4,304-511,781 subjects,
4.6-26 yr follow-up; highest vs lowest self-reported white-meat category.
outcome: CVD mortality; incident T2D (both patient-important).
dose_response_shape: NONE — categorical highest-vs-lowest; heterogeneous exposure metrics.
certainty (WCRF): "substantial effects unlikely" for BOTH CVD mortality and T2D; "limited -
no conclusion" for incident CHD, incident stroke, and incident CVD (too few studies).
confidence: low (the CV-mortality null is firm; the T2D null carries I2=82%).
- Both meta-analysed endpoints are null. CVD mortality: white meat «indicating no significant associations between intake of white meat and risk of CVD mortality (RR: 0.95, 95%CI: 0.87–1.02, P = 0.23) with low heterogeneity (I2 = 25%)». T2D: «no significant associations between high versus low intake of white meat and risk of T2D were found (RR: 0.98, 95%CI: 0.87–1.11, P = 0.81) with high heterogeneity (I2 = 82%)». The overall verdict: the evidence «does not indicate a role, either beneficial or detrimental, of white meat consumption for these diseases.» (Ramel et al., 2023)
- The T2D null is weaker than the CVD-mortality null — carry the I2=82%. High between-study heterogeneity means the pooled RR near 1.0 averages over studies pulling in both directions (Talaei and EPIC-InterAct-women showed higher T2D risk; Montonen and Villegas lower). The source names this: the «main uncertainty concerning the grading was the heterogeneity observed in the meta-analysis». It still graded «substantial effects unlikely» because it «was deemed unlikely that studies in the near future would affect the conclusion» — a certainty judgment, not a homogeneity claim. (Ramel et al., 2023)
- The evidence-state upgrade is the decision-relevant move (type-C / four-states). Lupoli and Kim reported bare null point estimates; Ramel attaches a WCRF certainty grade that separates two of the four evidence states cleanly: CVD mortality and T2D are “no meaningful effect” (graded «substantial effects unlikely»), while incident CHD / stroke / CVD are “insufficient evidence” (graded «limited – no conclusion» because the studies «were few and displayed somewhat mixed findings»). So the poultry->CVD-mortality and poultry->T2D cells move from unrated null to graded no-effect — a genuine certainty change, not a restated number. (Ramel et al., 2023)
The RCT arm — null, but the design suppressed the expected mechanism (a fat-matching artifact)
Three short RCTs (4-5 weeks; Bergeron, Mateo-Gallego, Scott) compared white meat to red meat on cardiometabolic risk factors and found nothing: «All included intervention studies matched fat content of intervention and control diets, and none of them showed any significant effects on the selected outcomes of white meat when compared to red meat.» But the reason they were null is design-mechanical, and the source flags it: «all the included trials matched the dietary fat intake of the different study arms and thus do not necessarily reflect real-world conditions.» (Ramel et al., 2023)
- Why this matters for the substitution reading. The proposed mechanism for red meat’s CV harm is its fat (SFA -> LDL/TC), and white meat’s real-world advantage is that it usually contains less fat. Matching fat between arms removes exactly the channel through which swapping red for white would act — so a null fat-matched RCT is not evidence that the swap is useless in practice; it is evidence that at equal fat, the meat type per se does little. This is the blinding/matrix problem in miniature (a design that isolates the protein by equalizing the fat answers a different question than the one the eater faces) -> The Comparator Problem, Saturated Fat Intake and Replacement. (inferred from Ramel et al., 2023)
Processed vs unprocessed white meat — the live, decision-relevant split (type-G gap)
Ramel surfaces the same processed-vs-unprocessed distinction that dominates the red-meat literature, now for poultry — but the data are almost absent: only two included studies isolated unprocessed white meat, and only one differentiated processed from unprocessed. The two that did split by direction: Steinbrecher found «processed poultry was associated with an increased risk of T2D in both men and women, whereas the intake of unprocessed poultry was not», and Villegas «showed a lower risk for unprocessed white meat». Ramel reports both only in its narrative synthesis — no processed- or unprocessed-poultry point estimate is extractable from the review for either (the direction is stated, the magnitude is not). (Ramel et al., 2023)
- Decision consequence: the aggregate white-meat-is-null-for-T2D reading may be masking a processed-poultry harm offsetting an unprocessed-poultry neutral-to-benefit — the same structure as red vs processed red meat -> Red and Processed Meat and Cancer, Should Adults Reduce Red and Processed Meat. This is a named gap, not a finding: two studies cannot carry it, and Ramel calls the differentiation «necessary in future studies». Hold it as the leading open question on the T2D cell. (inferred from Ramel et al., 2023)
Substitution was NOT modelled — the load-bearing limitation Ramel shares with Lupoli
Ramel states plainly that it «did not consider substitution of red meat with white meat but only intake of white meat», calling this a limitation because «food items are usually not consumed in addition to other foods but will replace them in the diet». It notes the substitution literature separately: «Replacing red meat/processed red meat with poultry has been associated with lower total mortality, while associations with CVD endpoints or T2D have been unclear.» So the decision that matters — swap red for white? — is answered by neither this SR’s intake analysis nor Lupoli’s: all three held sources measure intake of poultry, not the substitution, and the substitution signal (lower total mortality, unclear CVD/T2D) is imported, not estimated here. (Ramel et al., 2023)
The within-cohort SUBSTITUTION model (Etemadi 2017) — the swap the three MAs did not estimate [2026-09-02]
The gap Ramel names above — no held source estimates the red -> white substitution, only intake — is partly cashed by an independent US mega-cohort (NIH-AARP, 536 969 adults aged 50-71, 16-year follow-up, 128 524 deaths) that models the swap directly. Its main model holds total meat constant, so a rise in one meat type is a fall in the others: «This model was adjusted for total meat intake, so that increases in the meat variable of interest reflected reductions in other meat types and the total meat intake remained constant» (Etemadi et al., 2017).
- Substituting toward white meat is associated with lower all-cause mortality. «When the total meat intake was constant, the highest fifth of white meat intake was associated with a 25% reduction in risk of all cause mortality compared with the lowest intake level» (Etemadi et al., 2017), and almost all causes of death showed the inverse association. Per-unit, each 20 g/1000 kcal increase carried all-cause HR 0.93 (0.92-0.94) for poultry and 0.95 (0.94-0.96) for fish (Etemadi et al., 2017); the reduction was largest for unprocessed white meat (conclusion: reduced risks «particularly unprocessed white meat»).
- The B-distinction runs the OTHER way here — Etemadi’s white meat INCLUDES fish. Where Lupoli / Kim / Ramel define white meat as poultry with fish excluded, Etemadi’s aggregate white meat is «poultry and fish» — but it reports poultry (0.93) and fish (0.95) separately, so the poultry-only arm stays comparable to the fish-excluding MAs, and it is that arm, not the fish-inclusive aggregate, that matches this cluster’s exposure. (Etemadi et al., 2017)
- What it is and is not. This is a statistical substitution inside one FFQ-measured observational cohort — not a real swap and not a pooled substitution-SR. Total-meat-constant modelling infers the swap, it does not randomize it, and residual confounding by the whole healthier-diet pattern that accompanies white-meat preference is exactly the comparator problem the cluster turns on -> The Comparator Problem. So it narrows the substitution gap (a first within-cohort estimate of the swap — direction and rough magnitude) without closing it (one cohort, modelled not randomized).
Cross-source composite — what Kim adds to Lupoli (type-F, endpoint extension)
The two sources do not measure the same quantity, so this is a claim-refinement (F), not an independent-backing convergence (E). The parameter table shows why — every same quantity? cell is NO:
| Parameter | Lupoli 2021 | Kim 2017 | Same quantity? |
|---|---|---|---|
| Exposure | white meat, fish excluded | white meat = poultry only, fish excluded | YES — same B-distinction |
| Contrast | highest vs lowest, categorical | highest vs lowest, categorical | YES — both undefined dose |
| Endpoint | all-cause mortality; CV mortality; non-fatal CV events | stroke incidence; stroke mortality | NO — different outcomes |
| White-meat estimate | all-cause OR 0.94 (0.90-0.97); CV null | stroke incidence RR 0.87 (0.78-0.96); stroke mortality unestimable | NO — different endpoints |
| Evidence base | 22 studies / 3.1M | 2 articles / 4 comparisons / 138 761 | NO — Kim far thinner for white meat |
No cell pairs an identical quantity across the two, so there is no E and no D — the honest move is F: Kim extends the composite with a stroke-incidence endpoint Lupoli did not isolate. The composite the two build together (mark):
- White meat reads favourable on the two aggregate/incidence endpoints — all-cause mortality (Lupoli 0.94) and stroke incidence (Kim 0.87) — but null or unestimable on the cause-specific cardiovascular endpoints (Lupoli CV mortality 0.95 null and non-fatal CV events 0.99 null; Kim white-meat stroke mortality unestimable). The one place the two endpoints come closest — Kim’s favourable stroke incidence (0.87) beside Lupoli’s null non-fatal CV events (0.99, which would include stroke) — mildly diverges, and Kim’s base is only two articles, so the composite reading is a small favourable signal concentrated on aggregate endpoints, not a demonstrated cardiovascular mechanism. This is consistent with, and reinforces, the single-source caution above that the all-cause 6% is not carried by a matching CV-mortality effect.
- Both are substitution-confounded and both say so — Kim’s within-analysis red/processed-vs-white split and Lupoli’s red-meat-displacement caveat are the same structural point reached on two endpoints: what looks like a poultry benefit is inseparable from the red/processed meat it replaces -> The Comparator Problem.
Ramel vs Lupoli — the SAME CVD-mortality quantity, but shared cohorts (type-F, NOT E)
Ramel and Lupoli both pool prospective cohorts of poultry -> CVD/CV mortality, highest-vs-lowest, and
reach an almost identical null. This is the shape that most tempts an [E-independent] stamp — and the
parameter table shows it must be resisted, because the two rest on a largely overlapping evidence
base and Ramel explicitly cites Lupoli:
| Parameter | Lupoli 2021 | Ramel 2023 | Same quantity? |
|---|---|---|---|
| Exposure | white meat = poultry + rabbit, fish excluded | white meat = poultry, fish excluded | YES — same B-distinction |
| Endpoint (matched arm) | CV mortality | CVD mortality | YES — same endpoint |
| Contrast | highest vs lowest, categorical | highest vs lowest, categorical | YES — both undefined dose |
| Estimate (matched arm) | CV mortality OR 0.95 (0.89-1.01), null | CVD mortality RR 0.95 (0.87-1.02), null | YES — same null, ~same point |
| Evidence base | 22 prospective cohorts | 23 prospective cohorts; cites Lupoli (ref 15), shares cohorts (Bernstein, Farvid, Nagao, Rohrmann, Takata, van den Brandt, Key, Lee…) | NO — overlapping, not independent |
| Distinctive addition | all-cause mortality (Ramel did NOT do) | T2D incidence + WCRF certainty grade + fat-matched RCT arm (Lupoli did NOT do) | — |
Verdict: type-F (endpoint extension + certainty refinement), NOT type-E. The CVD-mortality
agreement is the same claim reached over the same underlying cohorts — shared-source corroboration,
which the laundered-E rule bars from an independence stamp (two reviews re-pooling the same trials, one
citing the other, is not independent backing). Ramel’s genuine contribution is F: it extends the
composite with the T2D cell Lupoli never measured, adds a formal WCRF grade, and adds the fat-matched
RCT arm; Lupoli conversely holds the all-cause endpoint Ramel omits. No [E-independent] token applied.
(inferred from Lupoli et al., 2021; Ramel et al., 2023)
- Ramel confirms the outcome-specific-shape reading. Ramel itself notes Lupoli found white meat «related to neither lower CVD incidence nor lower CVD mortality, although it was associated with a lower total mortality, an outcome that we did not investigate in the present analysis» — i.e. the two agree that the aggregate (all-cause) signal is not carried by the cause-specific CV endpoints, the exact caution the single-source Lupoli section already lodged. (Ramel et al., 2023)
Symmetric-standards / halo note — an independent government-funded SR reaching the null
Ramel is funded by the Nordic Council of Ministers and the Nordic governments’ food/health authorities,
with no declared conflicts. It reaches the SAME null on the poultry->T2D cell as the parked
National-Chicken-Council-funded conference abstracts (Connolly 2024, Baker 2026, held in inbox/_hold
as abstracts-only awaiting full papers). Two symmetric-standards points: (i) the null is not an
artifact of industry funding — an independent gold SR lands it too; and (ii) the fabric therefore
cites Ramel (independent, full paper, WCRF-graded) as the authority on this cell and does not need
to lean on an industry-funded abstract for the poultry->T2D question. Recorded as provenance, not as a
refutation of any source — symmetric standards apply to the funding note in both directions.
(inferred from Ramel et al., 2023)
Open gaps (type-G)
- Confounding by red-meat displacement is unresolved — the central caveat above; no analysis in any of the three held sources isolates poultry from what it replaces.
- Stroke mortality for white meat — unestimable in Kim (data limitation); still an open cell, distinct from a null. Stroke incidence is now held (Kim 0.87, thin 2-article base, above).
- CVD mortality + type-2-diabetes — now CLOSED by Ramel 2023 (both null, WCRF «substantial effects unlikely»; CVD-mortality null firm, T2D null carries I2=82%). The incident CV endpoints (CHD, stroke, CVD incidence) remain graded «limited – no conclusion» = insufficient evidence, an open cell distinct from the mortality null.
- Processed vs unprocessed white meat (the leading open question on T2D). Only 2 of Ramel’s cohorts isolated unprocessed poultry and only 1 split processed from unprocessed — processed poultry -> higher T2D (Steinbrecher), unprocessed -> neutral-to-lower (Villegas 0.79). The aggregate null may mask a processed-harm / unprocessed-neutral split, exactly as in red meat. Cannot be resolved on 2 studies — a named gap. (inferred from Ramel et al., 2023)
- The substitution question (red -> white) is now PARTLY estimated — by one cohort, not the MAs. The
three MAs all measure intake of poultry, not the swap; Etemadi 2017 (NIH-AARP) is the first held
source to model the substitution directly (total-meat-constant), landing a 25% lower all-cause mortality
for highest-vs-lowest white meat and per-20 g/1000 kcal HRs 0.93 (poultry) / 0.95 (fish) — see the
substitution-model section above. This narrows the gap to direction + rough magnitude but does not
close it: it is one FFQ-measured observational cohort with a modelled swap, so a pooled
substitution-analysis SR is still owed -> The Comparator Problem.
G (needs aggregation) - No dose-response on any endpoint — all three held sources use categorical highest-vs-lowest contrasts; a per-gram poultry curve, knee, or absolute risk difference remains uncomputable (needs aggregation).
Self-critique [run 2026-08-28 after attaching Kim's stroke endpoint — the cross-source step]
- The type is F, and the temptation was E — resisted. Two independent groups (Naples, Seoul)
reaching white meat looks favourable is exactly the shape that invites an
[E-independent]stamp. The parameter table blocks it: no cell pairs an identical quantity (Lupoli all-cause / CV mortality vs Kim stroke incidence / stroke mortality), so the agreement is on the substitution framing, not on a common endpoint. No[E-independent]token was applied; the composite is, not sold as convergent backing. - Not a manufactured tension either. Kim’s favourable stroke incidence (0.87) beside Lupoli’s null non-fatal CV events (0.99) could be dressed as a D-clash. It is not filed as one — they are different quantities on a thin base (Kim = 2 articles), so it is recorded as a mild divergence within the composite, a caution, not a joined issue.
- The thinness is stated, not buried. The white-meat stroke estimate rests on 2 articles / 4 comparisons; that limitation leads the section rather than trailing it, and the favourable number is not allowed to read as robust.
- Symmetric standards held. The within-Kim red/processed/white split is the decision-relevant move, but it is a substitution contrast, not proof poultry protects — the same comparator caveat applied to Lupoli is applied to Kim, so cashing the AWAITS did not tilt the page toward a poultry is protective pole. The stroke-mortality UNESTIMABLE arm is held as a gap, never as a reassuring null.
Self-critique [run 2026-09-01 after attaching Ramel's CVD-mortality + T2D endpoints]
- The Ramel/Lupoli CVD-mortality agreement is the classic laundered-E trap — resisted. Both pool
poultry -> CV mortality highest-vs-lowest and land RR/OR 0.95 (null); the temptation to stamp
[E-independent]is maximal because the point estimates coincide. The parameter table blocks it on the evidence-base row: Ramel cites Lupoli (ref 15) and re-pools largely the same cohorts, so the agreement is shared-source, not independent backing. Classified F, not E; no[E-independent]token. - The T2D null was not over-sold. I2=82% is stated at the point of the estimate, in the effect block, and in the gap list — the high heterogeneity leads, not trails, so the T2D “no meaningful effect” is not allowed to read as firm as the CVD-mortality one. The WCRF grade («substantial effects unlikely») is reported as a certainty judgment the source made, not re-derived by the wiki.
- The fat-matched-RCT claim is marked, not dressed as a finding. The observation that matching fat between arms suppresses the very mechanism the real-world swap would act through is my reasoning built on the source’s stated limitation, not the source’s conclusion — tagged accordingly.
- The symmetric-standards / industry-funding note cuts both ways. I did not claim the NCC-funded abstracts are wrong (they land the same null); the point is that the independent SR makes leaning on them unnecessary. The Barilla-funded Lupoli and government-funded Ramel are both recorded as funding notes, neither as a refutation — the same standard applied in both directions.
- No manufactured tension. Ramel agrees with Lupoli and Kim; nothing here is filed as a D-clash. The processed-vs-unprocessed divergence (Steinbrecher up, Villegas down) is held as a gap on 2 studies, not a joined issue.
Self-critique [run 2026-09-02 after attaching Etemadi's substitution model]
- Not sold as closing the gap. Etemadi is one FFQ cohort with a modelled (total-meat-constant) swap, not a randomized substitution and not a pooled substitution-SR. The section and the G-gap both say narrows, not closes, and the residual-confounding / comparator caveat is carried at point of use — the 25% is not allowed to read as a demonstrated causal swap effect.
- Not type-E. NIH-AARP/Sinha overlaps the cohort base the three MAs draw on, and the estimand
(highest-vs-lowest mortality under constant total meat) differs from the MAs’ intake contrasts, so no
[E-independent]stamp — Etemadi’s value is F/G (it extends the cluster with a substitution estimand none of the three computed), not independent convergence. - The B-distinction reversal is stated, not buried. Etemadi’s aggregate white meat includes fish, opposite to the cluster’s fish-excluding definition; the page flags this and uses the poultry-only arm (0.93) for comparability rather than the fish-inclusive aggregate — the exposure is matched, not assumed.
- No manufactured tension. Etemadi’s favourable white-meat substitution signal agrees with the cluster’s aggregate-endpoint direction; nothing is filed as a D-clash.
Provenance / independence note
Three gold MAs, two of them from separate groups and one (Ramel) from the NNR2023 SR group:
Lupoli from Naples (Riccardi/Vaccaro), Kim from Seoul National University (Park/Lee), Ramel from the
Nordic NNR2023 committee (Iceland/Sweden/Norway/Finland). Lupoli and Kim measure different endpoints
(parameter table), so their agreement is on framing, not a common quantity — NOT type-E. Ramel and
Lupoli do share the CVD-mortality quantity, but on overlapping cohorts with Ramel citing Lupoli —
shared-source, also NOT type-E. So the whole cluster is F/shared-source corroboration, with no verified
independent convergence and no [E-independent] stamp anywhere. Funding is recorded, not weighed:
Lupoli — Barilla Center (no stated role); Ramel — Nordic Council of Ministers + Nordic governments (no
declared conflicts). Lupoli also invokes environmental sustainability (poultry’s lower ecological
footprint) inside its recommendation, and Ramel’s intro likewise notes poultry’s lower greenhouse-gas
emissions — a named non-health axis the fabric records but does not price
-> Which Objective Moved This Recommendation.
(Lupoli et al., 2021)
(Kim et al., 2017)
(Ramel et al., 2023)