Nucleus of the inflammation-cvd cluster. Chronic vascular inflammation is a causal lever for
cardiovascular events — not merely a marker that travels with them. Two randomized trials of
different anti-inflammatory drugs cut cardiovascular events in statin-treated coronary patients without
changing lipids, which is the direct test the observational literature could never supply. But the
lever has three sharp boundaries the same sources force: the target is the inflammatory pathway, not
the CRP molecule that indexes it; the demonstrated benefit is on events, not all-cause mortality;
and the dietary version of the lever is, on current evidence, an observational proxy for diet patterns
already known to be healthy, not a shown independent lever. ((CCGC) et al., 2011; Nidorf et al., 2020; inferred from Ridker et al., 2017)
The causal proof — two RCTs, lipids unchanged, events down
The observational CRP-vs-disease literature (below) could show only association. CANTOS supplied the first randomized test by isolating inflammation from lipids: canakinumab, a monoclonal antibody against interleukin-1-beta, in 10,061 post-MI patients with hsCRP >= 2 mg/L, already on statins (91%).
- hsCRP fell dose-dependently (26/37/41 percentage points more than placebo at 50/150/300 mg); LDL and HDL were unchanged and triglycerides rose 4-5% (Ridker et al., 2017).
- Primary endpoint (nonfatal MI, nonfatal stroke, CV death) at 150 mg: HR 0.85 (95% CI 0.74-0.98, P=0.021) — 3.86 vs 4.50 events/100 person-years; 300 mg HR 0.86 (0.75-0.99) did not clear the multiplicity-adjusted threshold; 50 mg null (Ridker et al., 2017).
- The trial’s own verdict: canakinumab «led to a significantly lower rate of recur- rent cardiovascular events than placebo, independent of lipid-level lowering» (Ridker et al., 2017). Independent of lipids is the load-bearing phrase — it is what upgrades inflammation from marker to lever.
LoDoCo2 generalizes the finding with a cheap generic drug and a broad mechanism, in a different population and by an unrelated team. Colchicine 0.5 mg/day in 5522 chronic-coronary-disease patients (94% on statins): primary composite (CV death, spontaneous MI, ischemic stroke, ischemia-driven revascularization) HR 0.69 (95% CI 0.57-0.83, P<0.001; 6.8% vs 9.6%) (Nidorf et al., 2020). Colchicine’s mechanism is deliberately unlike canakinumab’s: «In contrast to selective inhibition of interleukin-1β by canakinumab, colchicine has broad cellular effects that include inhibition of tubulin polymerization and alteration of leu- kocyte responsiveness» (Nidorf et al., 2020).
Why the two together are stronger than either alone (convergent backing). Two different
anti-inflammatory agents (a narrow anti-IL-1-beta antibody; a broad-acting alkaloid), two different
teams and funders (Novartis-sponsored vs publicly funded Australian/Dutch), two different populations
(post-MI + high-CRP vs chronic stable coronary), both cutting events on top of statins — the shared
factor is dampening inflammation, not any one drug’s off-target profile. LoDoCo2 reads its own result
this way: «The CANTOS trial provided evidence suggest- ing that inflammation plays a causal role in the
pathogenesis of cardiovascular disease… Our results with colchicine are consistent with those obtained
in the first LoDoCo trial and the COLCOT trial» (Nidorf et al., 2020).
Not marked [E-independent], deliberately: LoDoCo2 was designed to test the hypothesis CANTOS
raised and cites CANTOS as antecedent, and both are secondary-prevention, on-statin, coronary
populations — so the two share a support-factor context and an explicit lineage of hypothesis. The
convergence is real and by independent interventions, but it is corroboration of one hypothesis, not two
blind routes arriving separately. (Nidorf et al., 2020; inferred from Ridker et al., 2017)
The target is the pathway, NOT the CRP molecule — the marker/lever crux
The single most important refinement the cluster forces: CRP predicts cardiovascular disease but does not cause it. Getting this wrong turns a good prognostic marker into a false treatment target -> Surrogate Outcomes. Two large individual-participant meta-analyses settle it, and — critically — they do not conflict. The BLOCKING parameter table:
| Parameter | ERFC 2010 (observational) | CCGC 2011 (Mendelian randomization) | Same quantity? |
|---|---|---|---|
| Design | IPD meta-analysis, 54 prospective studies, 160,309 people | MR IPD meta-analysis, 47 studies, 194,418 people, 4 CRP-gene SNPs | NO — measured association vs genetic instrument |
| Exposure contrast | circulating (measured) CRP, per 1-SD higher usual loge CRP (~3-fold) | genetically-raised CRP, per 1-SD higher genetically-predicted ln CRP | NO — same numeric scale, different source of the variation (measured level vs inherited genotype) |
| Outcome | CHD (first MI or fatal CHD) | CHD (fatal CHD / nonfatal MI / >50% stenosis) | ~yes |
| Circulating CRP -> CHD (risk-factor-adjusted) | RR 1.37 (95% CI 1.27-1.48) (Kaptoge et al., 2010) | RR 1.33 (1.23 to 1.43) ((CCGC) et al., 2011) | YES — both observational circulating-CRP -> CHD -> they AGREE |
| Genetically-raised CRP -> CHD | not estimated (no genetic arm) | RR 1.00 (0.90 to 1.13) per 1-SD ((CCGC) et al., 2011) | N/A — a G-gap on ERFC’s side, not a contrast |
Classification: refinement + disambiguation (F/B), NOT a tension (D). Where the two measure the same quantity (circulating CRP -> CHD) they agree (1.37 vs 1.33, overlapping intervals). CCGC then adds a genetic arm (1.00) that is discordant with its own observational arm — the Mendelian- randomization logic showing the observational association is confounded rather than causal. So the sources do not clash on a shared quantity; CCGC bounds ERFC (type F — the composite predicts-but-does-not-cause beats either alone) and disambiguates one term into two objects (type B — CRP-the-predictor vs CRP-the-cause). CCGC’s own conclusion: «Human genetic data indicate that C reactive protein concentration itself is unlikely to be even a modest causal factor in coronary heart disease» ((CCGC) et al., 2011).
Counter-passage read (CCGC’s own statement of what its MR does and does not show). CCGC explicitly denies the two readings that would manufacture a tension: it does not claim inflammation is non-causal, only that CRP-the-molecule is — «Irrespective of the causal relevance of C reactive protein itself to coronary heart disease, however, there is considerable evidence that persistent inflammation might contribute to cor- onary heart disease… studies seeking to test the inflammation hypothesis in coronary heart dis- ease should examine inflammatory mediators other than C reactive protein» ((CCGC) et al., 2011). And it explicitly preserves CRP’s predictive value (ERFC’s claim) — «Our findings also do not address the separate issue of the value of measurement of circulating C reactive protein in prediction of long term vascular risk» ((CCGC) et al., 2011). The not-joined checks fire: the two answer different questions (prediction vs causation) and do not disagree on the one they share. This is a distinction, not a joined issue.
These two are NOT independent backing — do not read their agreement as robustness. ERFC and CCGC
share the same Cambridge coordinating centre (Kaptoge, Di Angelantonio, S.G. Thompson, Danesh sit on
both), draw on overlapping cohorts, and CCGC cites ERFC 2010 as antecedent. Their agreement on the
observational number is shared-data agreement, not a type-E convergence — no [E-independent].
((CCGC) et al., 2011; inferred from Kaptoge et al., 2010)
The reconciliation, and why it is decision-relevant. ERFC (CRP predicts), CCGC (CRP does not cause but inflammation may), and CANTOS/LoDoCo2 (dampening inflammation cuts events) fit together with no contradiction: CRP is a prognostic marker for placing someone in a higher-risk stratum, while the lever is the inflammatory pathway itself, reached by a drug that acts upstream of CRP (IL-1-beta / IL-6 signalling; the NLRP3 inflammasome), not by “lowering CRP.” CANTOS names the resulting stratum: «statin-treated patients with residual inflammatory risk as assessed by means of a high-sensitivity C-reactive protein level of 2 mg or more per liter at baseline have future event rates that are at least as high as, if not higher than, those among statin-treated patients with a residual risk due to LDL cholesterol level» (Ridker et al., 2017). CRP selects who benefits; it is not itself the thing being fixed.
Causal for EVENTS is not a mortality lever — the net-effect boundary
Both trials cut cardiovascular events and neither reduced all-cause mortality — the textbook Net Effect vs Intended Effect case, where a real targeted benefit is offset elsewhere.
- CANTOS: «There was no significant difference between the canakinumab groups and the pla- cebo group in all-cause mortality» (HR 0.94, 95% CI 0.83-1.06) — and «signifi- cantly more deaths were attributed to infection or sepsis in the pooled canakinumab groups than in the placebo group» (0.31 vs 0.18 events/100 person-years, P=0.02) (Ridker et al., 2017). Suppressing innate immunity trades vascular events for fatal infection. (A partly offsetting reduction in cancer mortality was also seen — net all-cause mortality was flat.)
- LoDoCo2: «Colchicine did not result in a lower in- cidence of death from any cause than placebo» (0.9 vs 0.8/100 person-years, HR 1.21, 95% CI 0.86-1.71), with noncardiovascular death numerically higher (HR 1.51, 95% CI 0.99-2.31 — a CI that barely includes 1, flagged by the authors as a potential concern) (Nidorf et al., 2020).
Decision consequence: the inflammatory lever is validated for cardiovascular events (a patient-important outcome), not for length of life. A person weighting all-cause survival above event-avoidance gets a different answer than the event endpoint alone implies — the weighting is theirs (layer 3), but the wiki must not let the event benefit stand in for a mortality benefit it does not have. (Nidorf et al., 2020; inferred from Ridker et al., 2017)
The “anti-inflammatory diet” is a proxy for known-healthy patterns, not a shown independent lever
The drugs above are the evidenced form of the lever, and they are pharmacotherapy — out of reach as a lifestyle change for most people. The tempting inference is that an anti-inflammatory diet pulls the same lever. The best summary of that literature is the Dietary Inflammatory Index (DII) meta-analysis: 14 observational studies, 161,337 participants; highest-vs-lowest DII «showed a 36% increased risk of CVD incidence and mortality, with moderate evidence of heterogeneity (relative risk (RR) = 1.36, 95% confidence interval (CI): 1.19, 1.57; heterogeneity index I2 = 69%, p < 0.001)» (Shivappa et al., 2018).
Three reasons this does not establish an independent dietary lever [inferred from @shivappa2018dii]:
- It is association, and the authors frame it as prediction, not intervention — «These results further substantiate the utility of DII as tool to characterize the inflammatory potential of diet and to predict CVD incidence and mortality» (Shivappa et al., 2018). Predict, echoing the CRP marker leg — not a demonstrated causal handle.
- The DII is collinear with diet patterns already known to be healthy. It is an FFQ-derived score built from foods (fruit/vegetables, whole grains, polyphenols anti-inflammatory; red/processed meat pro-inflammatory), so a high-vs-low-DII contrast largely re-measures the Mediterranean / high-fibre pattern — several included cohorts are MedDiet cohorts (PREDIMED, SUN). The open question is whether “anti-inflammatory” adds anything beyond the pattern it re-labels -> Is the Food Category Doing Any Work.
- The index is validated against CRP and other markers — but CRP is the non-causal marker CCGC demoted, so the “inflammation” the DII scores is doubly removed from any evidenced causal target.
So the dietary leg belongs in the insufficient-evidence state for an independent lever: reasonable by mechanism and consistent with the pattern literature, but not shown to move events by acting on inflammation as opposed to by being a healthy diet. (inferred from Shivappa et al., 2018)
The same DII scores lower depression incidence in a separate prospective meta-analysis — low-vs-high DII «OR = 0.81, 95% CI = 0.71 to 0.92» with low heterogeneity (Molendijk et al., 2018) — which extends the same collinearity caveat to a second outcome: the DII re-labels a healthy-diet pattern, so a depression association does not establish an inflammation-specific pathway either, and the diet->depression link is itself suspected to run through the cardiometabolic route -> Depression and Modifiable Exposures.
Cross-outcome corroboration (cancer, a different field). The same caveat surfaces independently in
WCRF/AICR’s 2025 pattern report on colorectal cancer: the Empirical Dietary Inflammatory Pattern
(EDIP) and its insulin-anchored siblings were «graded as ‘strong-probable’ for their likelihood of
causality for an increased colorectal cancer risk» (World Cancer Research Fund International & American Institute for Cancer Research, 2025) — but the report itself flags the index does not cleanly map
its named mechanism: «alcohol consumption is negatively scored in these patterns but has strong evidence
of an increased risk for colorectal cancer. In non-drinkers, associations between the Empirical Dietary
Inflammatory Pattern and colorectal cancer risk were stronger than in those who consumed alcohol.»
(World Cancer Research Fund International & American Institute for Cancer Research, 2025) That is
the same the-index-relabels-a-diet-pattern, it does not isolate an inflammation-specific target point
the DII leg makes for CVD, now reached in cancer epidemiology by a different body — corroboration of the
caveat, not a confidence upgrade (different outcome, same-family limitation, no [E-independent]).
(inferred from World Cancer Research Fund International & American Institute for Cancer Research, 2025)
Yin’s 2025 umbrella review adds the DII->CRC direction from the observational-MA side: «The present UR also found a harmful association between the DII score and the risk of CRC comparing the highest score to the lowest, as well as a 1-unit increment in the score» (Yin et al., 2025) — but at the bottom of the credibility ladder, «classified as very low quality in terms of GRADE, mainly owing to the presence of heterogeneity and publication bias» (Yin et al., 2025). So the pro-inflammatory- diet -> CRC direction now shows up across three routes (WCRF’s EDIP grade, Yin’s DII re-grade, the DII CVD leg), all carrying the same the-index-relabels-a-diet-pattern caveat and none clearing a high evidence bar — reinforcing the insufficient-evidence verdict on an independent dietary-inflammation lever, not lifting it. (inferred from Yin et al., 2025)
Where it sits — residual risk, reverse causation, and the open gaps
- A second causal axis beside lipids. CANTOS is the cleanest evidence for residual inflammatory risk — event reduction on top of well-controlled LDL, by a non-lipid pathway. This bounds the lipid-centric model: apoB is a causal axis, not the only one -> LDL ApoB and Cumulative Exposure.
- CRP is partly a consequence, which is why it is a marker. Circulating CRP rises with adiposity, smoking, and occult illness, so a raised level sits partly downstream of the things one fears — the reverse-causation reading that makes it a predictor rather than a target -> The U-Shaped Association Artifact, Surrogate Outcomes.
- Open gaps (
G). (i) No anti-inflammatory therapy has shown an all-cause mortality benefit — the net-effect ceiling. (ii) The modifiable-by-lifestyle version of the lever is unproven at the outcome level (no RCT of an anti-inflammatory diet on hard CV events isolating inflammation from the diet pattern). (iii) The evidenced levers are drugs with real harms (fatal infection; GI/myotoxic effects), so the lever’s reachability for a general, lower-risk person is itself a limit, not a detail.
Synthesis
The cluster earns a nucleus because it resolves a question the individual sources cannot: inflammation is a validated causal lever for cardiovascular events, but only the pathway is the lever — CRP is the dial, not the mechanism; the benefit is on events, not survival; and the dietary version is, so far, a relabelling of diets already known to be healthy. The move present in no single source is the reconciliation of the marker leg (ERFC), the causal-molecule refutation (CCGC), the pathway-level RCTs (CANTOS + LoDoCo2), and the diet proxy (DII) into one coherent, bounded lever. Configurative, not aggregative: the sources answer adjacent questions (predicts / causes / treatable / dietary) that only together specify what to do and where the lever stops. ((CCGC) et al., 2011; Kaptoge et al., 2010; Nidorf et al., 2020; inferred from Ridker et al., 2017; Shivappa et al., 2018)
The loop is open. Every claim here is coherence- and source-graded, not outcome-validated; the mortality ceiling and the diet gap are exactly where a confident-sounding event benefit could mislead.
Self-critique [run 2026-08-08, before commit; author's — parent re-critiques independently]
- Overclaim: the headline is scoped to events, and the two boundaries (no all-cause-mortality
benefit; diet-proxy is not an independent lever) are stated as prominently as the benefit. Confidence is
medium, not high — it survives stripping the strongest single claim (the causal-for-events verdict still rests on two RCTs even if the diet and marker legs are removed). - Laundered independence: explicitly refused twice. ERFC+CCGC are flagged NON-independent (shared
Cambridge coordinating centre, overlapping cohorts, CCGC cites ERFC) — their agreement on the
observational RR is shared-data, not type-E. CANTOS+LoDoCo2 are convergent by different agents/teams
but NOT marked
[E-independent]because LoDoCo2 was designed to test the hypothesis CANTOS raised and cites it. No[E-independent]token anywhere on the page. - Fake tension: the ERFC-vs-CCGC crux is explicitly classified a refinement/disambiguation, NOT a
tension — the parameter table shows they AGREE where they measure the same quantity, and CCGC’s own
counter-passages (inflammation-still-open; prediction-not-addressed) close the not-joined checks. No
tensionfiled. - Same-quantity discipline: the crux parameter table carries a filled “same quantity?” column; circulating-CRP (both sources, agree) is kept distinct from genetically-raised-CRP (CCGC only, null).
- Selection bias (declared): these five sources were curated to build one argument; that is weak
evidence about the world, and the diet leg in particular is held at
insufficient-evidence, not asserted.
Addendum [run 2026-08-29, WCRF-2025 cross-outcome line]. The added EDIP->CRC paragraph is scoped
as corroboration of the existing caveat, not a confidence upgrade: it is a different outcome (CRC, not
CV events) from a different body (WCRF/AICR), so it is deliberately not marked [E-independent] and
confidence: stays medium. Overclaim guard: the paragraph asserts only that an empirical inflammatory
index re-labels a diet pattern (the same point the DII leg already makes), evidenced by WCRF’s own
alcohol-miscoring observation — it does not claim inflammation is or is not causal for CRC, and it does
not touch the page’s CVD headline. No new tension filed.
Addendum [run 2026-08-29, Yin-2025 DII->CRC line]. The Yin sentence is scoped as a third route
carrying the same caveat, explicitly at very-low GRADE, and marked not [E-independent]:
Yin pools the same observational-cohort DII literature the CVD leg draws on (shared-source, not an
independent route), so it reinforces the insufficient-evidence verdict rather than lifting it.
confidence: stays medium (the CVD-events headline is untouched). Overclaim guard: no causal claim
about inflammation and CRC is made — only that the pro-inflammatory-diet -> CRC direction recurs on
weak evidence.