GLP-1 Evidence

Not medical advice. A record of published research and our assessments of it. The limits

GLP-1 receptor agonist treatment reduces systemic inflammatory markers beyond what can be explained by treatment-associated weight loss.

Current status
Plausible unprovenEvidence weakening · Inflammation
Why
Biomarker reductions in CRP and hs-CRP are consistent and large across trials in obesity, type 2 diabetes, heart failure with preserved ejection fraction and obstructive sleep apnoea. Whether any of that reduction survives accounting for weight loss is a separate question, and a documented search plus two-pass appraisal of every study that attempts to answer it leaves the question open. The two designs that separate drug from weight most cleanly - a weight-matched comparator, and the same model run with and without adjustment for weight change - are both null. Three results point toward weight-independence, each with a limit: exenatide versus insulin glargine publishes only the weight-adjusted hs-CRP estimate, so the attenuation cannot be checked; a tirzepatide post hoc analysis reports only the share of hsCRP variance explained by weight, not an adjusted contrast; and a small exenatide trial without weight loss lowered IL-6 within the treated arm but reports no between-arm contrast. The largest mediation analysis reports the mediated proportion and omits the residual direct effect. Earlier support for this claim rested on SELECT's early-timepoint and non-loser analyses, which are at high risk of bias because they are not locatable in the trial's protocol or analysis plan, and on one small MCP-1 result against a diet arm that also lost weight (equal weight loss is not reported), which two later comparisons contradict. No study exists in normal-weight people, and CRP is a biomarker, not a clinical outcome.
What would change it
A randomized trial in normal-weight adults with inflammatory biomarkers AND a clinical or functional endpoint.

A weight-matched comparison (drug vs diet-induced equal weight loss) with hs-CRP and IL-6 as prespecified outcomes, adequately powered.

Evidence that biomarker reductions translate to fewer inflammatory disease events in people without obesity.

A published residual direct effect, with a confidence interval, from a mediation analysis of hs-CRP on weight change in a trial powered for it - the statistic 36056351 omitted.

A weight-matched comparison adequately powered for hs-CRP. Every existing one is under n=40.
Linked studies
15 · 3 supports · 3 contradicts · 3 mixed · 2 mechanism only · 4 consistent, but doesn't test the claim
Last updated
Oct 19, 2026

The status is the collection's own judgment on its assessment scale, not a GRADE certainty rating and not medical advice.

01The evidence

The evidence, sorted by what it shows

Supports (3)

Contradicts (3)

Mixed (3)

Mechanism only (2)

Consistent, but doesn't test the claim (4)

02Certainty

Certainty by outcome

A claim can be broken into separate outcomes, each with its own body of evidence. Certainty is rated per outcome, never for the claim as a whole, and every rating shows the reasoning for each domain.

O4 — Inflammatory marker change not explained by weight change

Field Assembly assessment: Untested (not a GRADE rating — what our scale means)

Rated against: null: certainty that a non-zero reduction remains after accounting for weight

No between-arm, weight-separated estimate exists in the body. Six designs capable of producing one; none reports it.

DomainJudgment and reasoning
Risk of biasvery serious · Every O4 result in the body is High, and both independent passes agreed on all six records (Q-025, twelve passes). The drivers recur: the weight-separating contrast is not itself randomization-protected in five of the six, the decisive analyses are post hoc or not prespecified, and two of the weight-matched designs achieve their matching by conditioning on success — 30513818 withdrew and replaced the ten of 62 participants who never reached the -7% target, analysing 35.
Inconsistencyserious · The body splits in direction, and the split tracks design quality rather than chance: the adjusted-model results point toward a weight-independent effect while the weight-matched comparators and the with-and-without-adjustment contrast do not. With no pooled estimate and no interval on most contributing results, the inconsistency cannot be quantified, only described.
Indirectnessserious · The outcome is a surrogate biomarker, which GRADE 8 treats as costing one or two levels, and no contributing study measures a clinical outcome. The studied populations are obesity, type 2 diabetes, obesity with HFpEF or OSA, and single instances of HIV on ART and PCOS — none is the healthy normal-weight population the collection's readers ask about, which is handled separately as an applicability judgment per grade-method-decision.md section 3. Two contributing contrasts use an active comparator.
Imprecisionvery serious · No meta-analysis and no pooled estimate, so this is written out in prose per section 4. The decisive point is stronger than imprecision: for the O4 target there is no between-arm, weight-separated estimate anywhere in the body to be imprecise about. 20424219 publishes only weight-adjusted figures with the crude contrast referenced but never printed; 31246368 reports its adjusted result narratively; 36056351 gives no point estimate, CI or p-value for the residual direct effect in any of its four trials; 30513818 reports 'ns'; 40514652 has no between-arm test for hs-CRP at all; 22013105 reports within-arm change only. Contributing sample sizes run from 24 to 2482.
Reporting and publication biasserious · The documented search is PubMed-only (data/searches/c001-2026-09-13.md), so per section 4 this domain cannot be judged confidently and is recorded as a limitation rather than rated not serious. Separately and more concretely, three of the most informative analyses omit the specific statistic that would answer their own question. That is selective within-study non-reporting, counted once at D5 in the study assessments, but it is publication-bias-shaped at body level and invisible to funnel-plot methods. The absence of a downgrade here should be read as ignorance, not reassurance.
Why untestedSix records use designs capable of answering the O4 question: two weight-matched comparators (30513818, 40514652), a model run with and without adjustment for weight change (31246368), a formal mediation across four phase 3 trials (36056351), an adjusted-model contrast (20424219), and a trial in which no weight loss occurred (22013105). None reports the between-arm, weight-separated estimate that would answer it. The question has been asked and the answers have not been published. Per-record reasons: data/assessments/Q-025-O4-ROB2-2026-09-14.md.

Population: Adults with obesity, type 2 diabetes, obesity with HFpEF or OSA, and single instances of HIV on ART and PCOS. Not normal-weight adults aged 55-75.
Comparator: For O4: a weight-matched comparator, statistical adjustment or mediation for weight change, or a design in which weight did not change.
Assessed by software, two independent passes · search: data/searches/c001-2026-09-13.md · method: c001-outcomes v1.4; rob2-guide v1.3; fa-certainty-scale v1

9 studies in this body. 9 further records were considered and left out:

03History

How this assessment has changed

  • Plausible unprovenOct 19, 2026 · Evidence weakening

    Owner decision 2026-10-18 on IN-061 (claims audit): recommended options accepted (IN-061.22). Status unchanged; rationale corrected to match the audited links.

  • Plausible unprovenSep 14, 2026 · Evidence weakening

    Trend changed from strengthening to weakening after Q-020 to Q-026: a documented search, two-pass appraisal of all 15 O4-capable records, and the finding that no study reports a between-arm weight-separated estimate. Status held at PLAUSIBLE_UNPROVEN because the nulls are small and unbounded and the question has not been properly asked. Owner-approved in chat.

  • Plausible unprovenSep 13, 2026 · Evidence strengthening

    METHOD_CHANGE: statement reworded to the falsifiable C-001 form (docs/c001-outcomes.md). Status unchanged.

  • Plausible unprovenSep 13, 2026 · Evidence strengthening

    initial curated status