Does lowering inflammation slow aging?

CANTOS supports cardiovascular event reduction through IL-1β inhibition in a selected population with previous myocardial infarction. CIRT did not demonstrate cardiovascular benefit from low-dose methotrexate in its target population. Neither trial identifies a general slowing of aging simply through a change in inflammatory markers. Post-treatment CRP groups do not preserve the original randomized comparison, and differences in inflammatory profiles between populations do not identify industrialized lifestyles as their cause.

Correction, 19 September 2026: We obtained the full CANTOS primary report, added dose-specific infection and mortality data, corrected person-time units in the absolute-effect figure, and withdrew the NNT extrapolated from median follow-up. We also withdraw claims of exact zero efficacy, a universal requirement for marker reduction, and a demonstrated nonmonotonic dose response. The IL6R paper’s title and consortium attribution were incorrect; its PMID, PMC record and reported numbers were consistent. Both texts, all four figures and the reproduction package have been updated. The original author’s sign-off is retained as history.

01Four questions do not establish a necessary causal chain

A drug may affect a pathway and several biomarkers. Seeing a biomarker and event rate change together does not establish that the biomarker mediates the benefit. Conversely, a trial without demonstrated marker or event reduction cannot establish that all benefits must pass through that marker. CRP should be distinguished from the upstream pathways it reflects. Conceptual background

02CANTOS: event benefit does not identify dose-response shape or mediation

CANTOS randomized 10,061 people with prior myocardial infarction and hsCRP of at least 2 mg/L to three canakinumab doses or placebo. Median follow-up was 3.7 years. The primary endpoint was the first post-randomization occurrence of nonfatal MI, nonfatal stroke or cardiovascular death. This was a trial of further events in people with established disease, rather than general aging in healthy adults. Methods and Table 2 Original design

At 48 months, the median percentage reduction in hsCRP from baseline was 26, 37 and 41 percentage points greater than placebo in the three dose groups. These are differences in percentage change, not absolute CRP concentrations or a change experienced by every participant.

The 150mg group passed the prespecified primary and key-secondary testing procedure. Primary thresholds were 0.02115 for 50/150mg and 0.01058 for 300mg; nominal P <0.05 alone is insufficient. We rely on the authors’ reported formal testing result and did not obtain individual data to rerun the stratified Cox models. Statistical methods and Table 2 footnotes

However, a rate of 3.90 at 300mg versus 3.86 at 150mg does not demonstrate nonmonotonicity or superiority of 150mg. An illustrative Poisson approximation from published counts and rounded rates gives a 300mg/150mg rate ratio of about 1.01, with a 95% interval of approximately 0.87–1.18. This is not a prespecified adjusted dose comparison. A tiny difference between point estimates cannot decide the mechanism.

03Benefits and harms cannot be subtracted as a net score

The primary-event rate difference for 150mg is −0.64 per 100 person-years, or 6.4 fewer participants with a first primary event per 1,000 person-years. This incidence-density difference is not an observed reduction of 24 events per 1,000 people treated for 3.7 years. Median follow-up is not an identical observation period for everyone, and a first-event rate cannot directly provide cumulative risk at a common time. The earlier NNT of approximately 42 is withdrawn as a trial conclusion.

The full report distinguishes dose groups and analysis windows:

The reported increase in fatal infection, nominal P=0.02, compares all doses pooled with placebo. It cannot be reassigned to 150mg alone. The 150mg point estimate is higher than placebo, with limited precision. Serious infection and all serious adverse events are also different outcomes from fatal infection; the pooled comparison for serious infection did not show a clear difference. Tables 2–3 and footnotes

Clinical endpoints use intention-to-treat follow-up, while Table 3 reports safety during treatment. A common person-time unit does not remove differences in observation windows or outcome severity. Subtracting excess infections from prevented cardiovascular events would not produce a validated net-benefit score. Mortality intervals do not precisely establish a survival benefit, but also do not prove exact cancellation of all benefits and harms.

04CIRT: no demonstrated benefit does not mean exact zero effect

CIRT randomized 4,786 people with prior MI or multivessel coronary disease plus diabetes or metabolic syndrome. Before randomization, participants underwent a five-to-eight-week methotrexate run-in. Those developing relevant adverse effects or laboratory abnormalities did not proceed. Safety findings therefore apply to selected tolerators, not every person starting treatment. CIRT methods

The trial did not demonstrate reductions in IL-1β, IL-6 or hsCRP. Median baseline hsCRP was about 1.5 mg/L, below the CANTOS entry threshold, but that was one of several differences. Drug, pathway, eligibility and background characteristics changed together. A comparison between trials cannot isolate whether the lack of benefit arose from the drug, inflammation level or another feature, nor establish that methotrexate had no relevant pathway activity whatsoever.

At a median 2.3 years, the final primary composite occurred in 201 versus 207 participants, HR 0.96 (0.79–1.16). The original MACE endpoint occurred in 170 versus 167, HR 1.01 (0.82–1.25). The final definition added unstable-angina hospitalization requiring urgent revascularization. The report states that the relevant decision-makers were unaware of unblinded data when the endpoint was expanded. The trial subsequently stopped after crossing a futility boundary and showing no evidence of CRP reduction. Both endpoint definitions should be retained. Endpoints, statistical analysis and Table 2

The 12 displayed endpoints did not demonstrate clear benefit, but they overlap and include nested composites. Their consistency is not equivalent to 12 independent trials. The primary interval still permits some benefit or harm. Secondary and tertiary intervals were not multiplicity-adjusted; a collection of nonsignificant results does not set the effect to exactly zero.

Harms also require distinct definitions. Any adverse-event rates were 62.4 versus 56.0 per 100 person-years, and any-infection rates 16.5 versus 14.4. Serious infection rates were 2.24 versus 2.47, without the same clear direction of increase. Non-basal-cell skin cancer counts were 31 versus 10 on visit case-report forms, and 33 versus 12 using the broader MedDRA/event-report ascertainment. These definitions must not be counted twice. Signals for liver-enzyme elevations, leukopenia and oral symptoms remain relevant. Table 3

05Responders and subgroups generate hypotheses

In the CANTOS post-treatment hsCRP<2 mg/L group, the published adjusted MACE HR was 0.75 (0.66–0.85), compared with 0.90 (0.79–1.02) among those not reaching that level. The lower-CRP group had an all-cause mortality HR of 0.69. The authors reported baseline adjustment and multiple sensitivity and causal-inference analyses. Nevertheless, directly grouping by a post-treatment variable is not the original randomized comparison: whether the same individual would respond under both treatment and placebo is not simultaneously observable. Response analysis

These findings support predictive or mechanistic hypotheses, rather than establishing from the strata alone that lowering CRP itself causes the benefit. This review re-read the abstract but did not fully audit or reproduce that report’s causal models; its sensitivity analyses are not treated as demonstrated elimination of selection bias.

Other exploratory CANTOS results remain worth examining. The MACE HR among TET2-CHIP carriers was 0.38 (0.15–0.96), but the interaction P=0.14 does not establish greater treatment benefit than in others. TET2 analysis The lung-cancer analysis also reported favorable signals while explicitly calling for independent confirmation. Exploratory cancer results These publications are not new independent randomized trials.

Post-treatment IL-6 and IL-18 still predicted subsequent risk, and inflammatory associations were also observed in participants with impaired kidney function. Residual risk does not establish that adding another pathway inhibitor will improve outcomes. A nonsignificant lipid association in a selected, statin-treated subgroup likewise does not show that lipids have become irrelevant. IL-6/IL-18 Kidney-function substudy

06Genetics and population comparisons address different questions

The IL6R source used here is the IL6R Mendelian Randomisation Analysis Consortium paper, The interleukin-6 receptor as a target for prevention of coronary heart disease: a mendelian randomisation analysis, PMID 22421340, PMC3316968. The previous title and consortium attribution belonged to a different collaborative report.

The functional variant tagged by rs7529229 was associated per allele with approximately 9.45% higher IL-6 and 8.35% lower CRP; the coronary-disease OR was 0.95 (0.93–0.97), using 25,458 cases and 100,740 controls. These numbers belong to the corrected paper identity. Higher circulating IL-6 with lower downstream CRP is compatible with attenuated receptor signaling; a lower ligand concentration is not a universal definition of pathway inhibition. Abstract and discussion

The finding supports IL6R involvement under genetic-instrument and extrapolation assumptions. Lifelong genetic variation is not the same as taking a small drug dose from embryonic life; the authors also discuss discordant lipid effects between genotype and medication. This study does not establish that CRP itself causes disease or that lowering CRP necessarily produces benefit.

The 2025 population study tested a particular axis constructed from circulating cytokines, comparing Italian InCHIANTI, Singaporean SLAS, Bolivian Tsimane and Malaysian Orang Asli data. The groups did not all measure the same 19 markers: overlap with the Italian panel was 16, eight and eight markers, respectively. Assay platforms and available disease outcomes also differed. Results, Figure 1 and methods

Italian and Singaporean structures were similar but not identical, and each age or disease association need not replicate. The different structures and associations in the two nonindustrialized populations limit transportability of this measurement. The authors propose industrialization-related exposures and other explanations, but explicitly acknowledge that four populations cannot directly test those drivers. Infection, assay differences, outcome definitions and mortality selection remain relevant. These data neither identify industrialized lifestyles as the cause of all inflammaging nor show that inflammation cannot be an aging biomarker in any population.

07What the evidence supports

Randomized evidence supports cardiovascular event reduction from a particular inflammatory-pathway intervention in a selected population, alongside safety concerns. Genetics supplies another line of support for pathway involvement without establishing CRP as the mediator. The negative CIRT result constrains its drug, population and comparison, rather than every anti-inflammatory strategy.

Inflammatory measurements can carry age- and disease-related information in specific settings; their meaning across environments requires validation. Disease prevention is itself an important health outcome. What these results do not identify is a unified slowing of aging across function, cognition, multiple organ systems and lifespan. The primary reports reviewed here did not establish that connection; failure to establish it is not a demonstration that it is false.

We did not obtain individual data from either trial or the four population cohorts. Our computations check arithmetic and approximate intervals from published counts and rounded incidence rates. Formal HRs, prespecified testing and subgroup estimates remain the original authors’ results. Approximate Poisson intervals do not reproduce adjusted models and are not multiplicity-corrected. Inputs, code, outputs and four figures are provided in the reproduction package. Source checking extends to 19 September 2026. Codex performed the revision and bilingual self-review; this is not independent clinical or human professional review and provides no personal medication advice.