The wiki’s red-meat -> CVD evidence is otherwise entirely observational cohort and holds no lean-red-meat feeding-trial lipid data at all. Two controlled-feeding RCTs fill that empty cell. Together they answer a question neither answers alone: within a low-SFA background, the atherogenic-lipoprotein signal is set by the background SFA load and by whether the protein is meat or plant — not by how much lean beef is on the plate, and not by whether the meat is red or white. Every outcome here is a surrogate (LDL-C, apoB, LDL particles); neither trial measured a clinical CVD event, so read the whole page at surrogate level -> Surrogate Outcomes. (Bergeron et al., 2019; inferred from Roussell et al., 2012) A gold pooled MA of 36 substitution RCTs (Guasch-Ferré 2019, below) now sits over the two feeding trials: it makes the comparator-dependence quantitative across the whole trial literature — the sign of red meat’s lipid effect flips with what replaces it — and it pools Roussell’s own BOLD data, so it is a pooled-trial refinement of this page, not an independent route. It too measures only surrogates. (inferred from Guasch-Ferré et al., 2019)
Why controlled feeding is the right instrument here
The observational red-meat fabric is confounded and carries large dietary-measurement error -> Measurement Error in Dietary Assessment. These are metabolic-kitchen feeding trials — foods provided, weight held stable, high compliance — so they estimate the direct lipid effect of substituting one protein for another with the rest of the diet matched. That buys internal validity on a surrogate at the cost of a short horizon and no hard endpoint. Bergeron framed the specific hole: «there has to date been no systematic evaluation of the potential interaction of dietary protein source and SFA content on concentrations of LDL cholesterol and related atherogenic lipoprotein measures, including levels of lipoprotein particles». (Bergeron et al., 2019)
Roussell BOLD 2012 — amount of lean beef, within a low-SFA pattern
effect_measure: LDL-C fell vs the high-SFA control (HAD) by 5.5% (DASH), 4.7% (BOLD),
4.4% (BOLD+); all P<0.05, and NO difference among the three test diets
(P>0.1). Absolute LDL-C decrease from baseline: DASH -0.37, BOLD -0.35,
BOLD+ -0.345 mmol/L (vs HAD's own -0.14 decrease). apoB fell significantly
only on BOLD+ (88.6 vs HAD
92.8 mg/dL); BOLD 91.1 and DASH 91.0 were not significant vs HAD.
population_and_comparator: 36 hypercholesterolemic adults (baseline LDL-C >2.8 mmol/L), 4-period
randomized crossover controlled feeding, 5 wk/diet. Arms: HAD (33% fat,
12% SFA, 20 g beef/d), DASH (27% fat, 6% SFA, 28 g beef/d), BOLD (28% fat,
6% SFA, 113 g beef/d), BOLD+ (28% fat, 6% SFA, 27% protein, 153 g beef/d).
outcome: LDL-C, TC, HDL-C, apoB, apoA-I, apoC-III (all SURROGATE).
dose_response_shape: flat across beef dose once SFA is fixed — 20 -> 153 g beef/d did not
separate the test diets on LDL-C.
uncertainty: n=36, 5-wk, surrogate-only, no events; industry funded (Beef Checkoff).
effect_modifiers: baseline CRP: in the high-CRP (>1 mg/L) subgroup, BOLD/BOLD+ (but not DASH)
significantly lowered TC/LDL-C from baseline — a secondary, hypothesis-
generating finding.
certainty: author conclusion: lean beef in a low-SFA pattern gives favorable lipid
effects comparable to DASH.
confidence: medium
The design point: BOLD swapped a large amount of lean beef (113-153 g/d) into a DASH-like pattern and the LDL-C reduction stayed equal to DASH’s. Roussell reads this as the protein source not mattering once the fat is controlled — «the protein source [with the exception of soy protein (21)] does not appear to modify the TC or LDL cholesterol response to a cholesterol-lowering diet». (Roussell et al., 2012) The LDL-C drop tracked the SFA cut (12% -> 6% E), not the beef. The conclusion: «Low-SFA, heart-healthy dietary patterns that contain lean beef elicit favorable effects on cardiovascular disease (CVD) lipid and lipoprotein risk factors that are comparable to those elicited by a DASH dietary pattern». (Roussell et al., 2012)
The blind spot Roussell cannot see: it has no plant/nonmeat arm. Every comparator is another animal-protein or mixed diet, and the control is high-SFA, so lean beef can only look good. The trial shows beef is not worse than DASH; it cannot show whether beef is as good as removing the meat. (inferred from Roussell et al., 2012) -> The Comparator Problem
Bergeron APPROACH 2019 — protein source x SFA level, factorially
effect_measure: LDL-C and apoB were HIGHER on red and white meat than on nonmeat,
independent of SFA level (P<0.0001, except apoB red-vs-nonmeat P=0.0004),
driven by LARGE LDL particles; small+medium LDL and total/HDL-C were
unaffected by protein source (P=0.10, P=0.51). Red vs white meat did NOT
differ on any primary outcome. Independent of protein source, high vs low
SFA raised LDL-C (P=0.0003), apoB (P=0.0002), and large LDL (P=0.0002);
effects were ADDITIVE (no protein x SFA interaction).
population_and_comparator: 113 generally-healthy adults (per-protocol; 62 high-SFA, 51 low-SFA arm),
2 parallel SFA arms (high \~14% E vs low \~7% E) x 3-period crossover of
red meat / white meat / nonmeat protein (\~12% E each), 4 wk/diet.
outcome: primary: LDL-C, apoB, small+medium LDL, total/HDL-C (all SURROGATE).
dose_response_shape: SFA effect on LDL-C/apoB confined to LARGE LDL, not small/medium.
uncertainty: surrogate-only, no events; 4-wk; lean cuts only, no fish, grain-finished
beef; SFA supplied mainly from dairy; industry-linked (Krauss/Bergeron
hold a Dairy Management Inc grant [not for this work]; Krauss holds an
ion-mobility patent).
effect_modifiers: none material — no protein x SFA interaction on the lipoprotein biomarkers.
certainty: author conclusion: no lipid basis to choose white over red meat; plant
substitution + unsaturated-for-SFA would yield benefit.
confidence: medium
Bergeron adds the two contrasts Roussell lacks. First, the plant arm. Both meats — red and white — raised LDL-C and apoB relative to nonmeat protein, even at low SFA: «LDL cholesterol and apoB were higher with red and white meat than with nonmeat, independent of SFA content». (Bergeron et al., 2019) So the comparator was doing the work in Roussell: against a plant referent, lean meat is not neutral. Second, red vs white head-to-head — no difference: the trial does «not provide evidence for choosing white over red meat for reducing CVD risk on the basis of plasma lipid and lipoprotein effects». (Bergeron et al., 2019) Meat color is not the lever; meat-vs-plant and background SFA are.
Bergeron’s surrogate caveat, on its own data. The LDL-C/apoB rise from meat and from SFA landed in large LDL particles, which are more weakly CVD-associated than small/dense LDL — so the surrogate may overstate the hard-outcome effect: «the impact of high intakes of red and white meat, as well as SFA from dairy sources, which selectively raised large LDL subfractions, may be overestimated by reliance on LDL cholesterol». (Bergeron et al., 2019) This is a rare instance of a trial flagging that its own primary surrogate over-reads the endpoint that matters -> Surrogate Outcomes, LDL ApoB and Cumulative Exposure.
Guasch-Ferré 2019 — the comparator sets the SIGN, pooled across 36 trials
Where the two feeding trials fix one comparator each, this gold MA pools 36 RCTs (1803 participants) of red-meat diets versus diets that replace the red meat with a specified food, and stratifies the effect by what the replacement was. The result is the substitution principle made quantitative: the sign of red meat’s lipid effect flips with the comparator, so does red meat harm lipids? is ill-posed until the substitute is named. Positive WMD below = red meat lowered the lipid LESS than the comparator (comparator better); negative = red meat lowered it MORE (red meat better on that analyte). All mmol/L, all SURROGATE.
effect_measure: ALL comparators combined (n=36): NO differential effect on total-C, LDL-C,
HDL-C, apoA1, apoB, or BP; red meat gave lesser TG decrease (WMD +0.065;
95% CI 0.000-0.129). vs HIGH-QUALITY PLANT protein (legumes/soy/nuts, n=8):
red meat lesser decrease in total-C (WMD +0.264; 95% CI 0.144-0.383;
P<0.001) and LDL-C (WMD +0.198; 95% CI 0.065-0.330; P=0.003) -- plant is
better. vs FISH only (n=10): red meat GREATER decrease in LDL-C (WMD -0.173;
95% CI -0.260 to -0.086; P<0.001) and HDL-C (WMD -0.065; 95% CI -0.109 to
-0.020) -- fish RAISED LDL (and HDL). vs POULTRY: no differential lipid
effect. vs CARBOHYDRATE (n=3): red meat GREATER TG decrease (WMD -0.181;
95% CI -0.349 to -0.013) but lesser HDL-C decrease (WMD +0.139; 95% CI
0.004-0.275).
population_and_comparator: 36 RCTs, 1803 adults, diets prescribing differing red-meat amounts (46.5-
500 g/d intervention vs 0-266 g/d comparator), >=2 wk; PubMed to July 2017.
Comparators STRATIFIED (plant / fish / poultry / poultry+fish / mixed animal
/ carbohydrate / usual diet). 20 crossover, 16 parallel; none double-blinded.
outcome: total-C, LDL-C, HDL-C, TG, apoA1, apoB, BP (all SURROGATE; no CVD event).
dose_response_shape: NULL over the studied range 0-500 g/d red meat (continuous g/d): total-C
P=0.73, LDL-C P=0.49, HDL-C P=0.57, TG P=0.05. No knee located; monotonicity
not even shown -- a flat pooled slope.
uncertainty: surrogates only; most trials small (n 8-191); magnitude of differences small;
no blinding; heterogeneity present but reduced by comparator-stratification.
Funding-source subgroup (red-meat-industry vs other) showed no differential.
effect_modifiers: the COMPARATOR is the modifier (route not (b)-on-one-food but the whole
substitution frame); lean vs nonlean (below); SFA gap between arms.
certainty: author conclusion: substituting red meat with high-quality plant protein --
but not fish or low-quality carbohydrate -- gives more favorable lipids.
confidence: medium
The sign-flip is the decision content. Against plant protein red meat looks worse on LDL-C (+0.198 mmol/L); against fish it looks better on LDL-C (-0.173 mmol/L, because oily-fish comparators raised LDL); against refined carbohydrate it looks better on triglycerides (-0.181 mmol/L). Same food, opposite-signed surrogate effects — the clean pooled-trial instance of The Comparator Problem and of framing every recommendation as a substitution. (Guasch-Ferré et al., 2019) Guasch-Ferré states the principle directly: «Specification of an explicit comparison is the cornerstone of nutritional substitution analysis … Analyses that do not specify a comparison implicitly compare the food(s) under study with the mixture of all other calorie-bearing foods in the diet, making interpretations and dietary recommendations difficult.» (Guasch-Ferré et al., 2019) The all-combined near-null is itself an artifact of not naming the substitute — averaging opposite-signed effects to roughly zero: «Inconsistencies regarding the effects of red meat on cardiovascular disease risk factors are attributable, in part, to the composition of the comparison diet.» (Guasch-Ferré et al., 2019)
Two refinements of this page’s earlier claims. (i) The dose-response null over 0-500 g/d pools the very BOLD data the page holds — Roussell 2012 and Hill 2015 are among the 36 — lifting «flat across beef dose» from one trial to a 36-trial pooled null, and stating it correctly bounded to the studied range. (ii) Lean red meat vs all comparators gave greater decreases in total-C (WMD -0.05; 95% CI -0.12 to -0.02) and LDL-C (WMD -0.08; 95% CI -0.15 to -0.02), but nonlean did not — pooled support for the page’s lean-cuts-only caveat. (Guasch-Ferré et al., 2019)
Surrogate discipline holds here too. No CVD event was measured, and the authors say the transmission is not theirs to make: «we cannot directly extrapolate CVD risk from intermediate biomarkers such as lipids, apolipoproteins, and blood pressure». (Guasch-Ferré et al., 2019) -> Surrogate Outcomes, LDL ApoB and Cumulative Exposure
Synthesis — the parameter table (same-quantity check) BEFORE the cross-source claim
The two trials share analytes but test different contrasts, so this is a type-A/F composite, not a head-to-head. The analyte-level comparison is valid; the contrast-level one is not.
| Matched quantity | Roussell (BOLD) | Bergeron (APPROACH) | Same quantity? |
|---|---|---|---|
| LDL-C | Friedewald-calculated, mmol/L | Friedewald-calculated, mmol/L | Yes — same measure |
| apoB | mass, mg/dL (immunoturbidimetric) | mass, g/L (immunoturbidimetric) | Yes — same analyte, different units (92.8 mg/dL = 0.928 g/L) |
| LDL particles | not measured | ion-mobility subfractions (large / small+medium) | No — Bergeron only |
| The CONTRAST tested | amount of lean beef (20->153 g/d), background SFA cut 12->6% E | source of protein (red/white/nonmeat) x SFA level (7 vs 14% E) | No — different independent variables |
Because the analytes match but the contrasts do not, the synthesis is compositional, not a convergence of the same estimate. What each supplies:
- Type-A (composite claim, in neither source alone): within a low-SFA background the atherogenic- lipoprotein signal is governed by the background SFA load and by meat-vs-plant, not by beef amount or red-vs-white color. Roussell supplies beef amount is irrelevant once SFA is fixed (flat across 20->153 g/d); Bergeron supplies SFA is a main effect and red=white (additive SFA effect, no color effect). Neither states the joint claim. (Bergeron et al., 2019; inferred from Roussell et al., 2012)
- Type-F (Bergeron bounds/refines Roussell): Roussell’s lean beef is fine rests on a high-SFA, no-plant comparator; Bergeron adds the plant arm (meat > nonmeat on LDL-C/apoB) and the particle data (the rise is large-LDL, weakly atherogenic), so the composite is more informative than Roussell’s reading alone. (inferred from Bergeron et al., 2019)
- NOT scored type-E. The two do share one compatible sub-finding — lean red meat within low SFA does not raise LDL-C versus a lean animal comparator (Roussell: BOLD=DASH; Bergeron: red=white) — but they are different labs (Kris-Etherton/Penn State vs Krauss/CHORI, no shared authors) testing different referents, so this is compositional agreement, not the same estimate reached by two independent routes. Do not launder it as independent-backing robustness. (Bergeron et al., 2019; inferred from Roussell et al., 2012)
- Guasch-Ferré 2019 is type-F + type-C on this page, and explicitly NOT type-E of it. Type-F —
the MA POOLS Roussell (BOLD) and Hill 2015 among its 36 trials, so its 0-500 g/d dose-response null
refines the page’s single-trial «flat across beef dose» to a pooled null, and its lean-vs-nonlean
subgroup refines the lean-cuts caveat; a source that contains the incumbent’s own data is a
refinement, never an independent route. Type-C — it names the substitution frame as the reason the
red-meat lipid question is ill-posed unbounded (the all-combined near-null is opposite-signed effects
averaged away). The compatible meat-worse-than-plant finding it shares with Bergeron is NOT stamped
[E-independent]: Guasch-Ferré is Harvard/Hu-Willett, Bergeron is Krauss/CHORI (different labs), but the estimates are not identical and the meat>plant claim is broadly established, so this is shared-conclusion, not two non-obvious independent routes converging — the independence bar is not met. (inferred from Guasch-Ferré et al., 2019)
Is the food category doing any work here?
This is a clean worked instance for Is the Food Category Doing Any Work: once SFA and total protein are matched, red meat as a lipid category dissolves — the beef amount does not separate diets (Roussell), and red does not separate from white (Bergeron). What does separate is (i) the background SFA load and (ii) meat versus plant protein. The category carrying the lipid signal is not red meat; it is animal protein plus its saturated-fat travelling companions. This bounds Saturated Fat Intake and Replacement on the specific case of lean red meat and refines the leaner-cut reasoning by attaching an actual apoB/LDL-C number to it. (Bergeron et al., 2019; inferred from Roussell et al., 2012)
What this does NOT license, and the gaps (type-G)
- Surrogate only — no hard-endpoint claim. Neither trial measured ASCVD events; apoB’s transmission
to ASCVD is an evidenced claim held elsewhere (LDL ApoB and Cumulative Exposure,
Surrogate Outcomes) — link it, do not assert the event benefit here. Bergeron itself calls for
«clinical CVD outcomes in individuals with hyperlipidemia».
(Bergeron et al., 2019)
G-gap: no lean-red-meat feeding trial with a patient-important CVD endpoint or a
morbidity-trajectory outcome.
type-G - Lean cuts only. Both trials used lean cuts matched for SFA, so nothing here transports to
higher-fat red meat — Bergeron: «we cannot extrapolate our findings to the lipid and lipoprotein
effects of higher-fat red meat products».
(Bergeron et al., 2019)
type-G - How much of the plant-arm benefit is the plant, versus the removal of meat, is unresolved
(Bergeron flags plant phytochemicals/fibre vs meat removal as unseparated).
type-G - Industry funding on both sides (Beef Checkoff; a Dairy Management grant + an ion-mobility patent) — a symmetric-standards bias watch, not a dismissal; the controlled-feeding designs and the fact that the two funders’ interests point in opposite directions yet the lipid findings agree, blunts the concern.
Net decision-change: for a reasonably healthy or hypercholesterolemic person already eating a
low-SFA pattern, swapping in lean red meat (or choosing red vs white) is close to lipid-neutral at the
surrogate level; the levers that move LDL-C/apoB are cutting background SFA and shifting protein from
meat toward plant sources — and even those move mostly large LDL, so the surrogate likely overstates
the hard-outcome size. Guasch-Ferré sharpens the substitution framing: the surrogate benefit of
reducing red meat exists chiefly when the replacement is high-quality plant protein (LDL-C WMD +0.198
mmol/L vs plant); replacing it with fish or refined carbohydrate does not improve — and on some analytes
worsens — the lipid profile, so eat less red meat carries a decision only once the substitute is named.
The magnitude is small and event evidence is absent, hence confidence: medium.
(inferred from Guasch-Ferré et al., 2019)