A FormBlends network publication
The Science

How this site grades mechanism evidence, and how to read a mechanism claim

The A to D scale used across The Science and the mechanism map, why animal findings get their own grade, what to check before believing any claim about how a drug works, and worked examples from this site.

By FormBlends editorial teamUpdated September 4, 2026Educational, not medical advice

Most explanations of how GLP-1 drugs work mix four kinds of evidence into one confident paragraph. This site separates them, and this page is the key.

The scale

A. Randomised human trial, or an FDA label statement derived from one. SELECT's 20% reduction in major cardiovascular events. The Wegovy label's 44% nausea rate. These are effects measured in people under randomisation. They tell you what happens; they rarely tell you why.

B. Human physiology, imaging, pharmacokinetics or large cohort. Nauck's clamp studies showing glucose-dependent insulin release. Blundell's ad libitum meal test. Pasternak's thyroid cancer cohort. Human data, but observational or small, or measuring an intermediate rather than an outcome.

C. Animal or cell evidence only. Secher's arcuate nucleus tracing. Huang's dissociable hindbrain circuits. Bjerre Knudsen's rodent C-cell tumours. This is where most mechanism knowledge lives. It explains; it does not prove in people.

D. Hypothesis or actively debated. The GIP agonist-antagonist paradox. Weight-independent cardiovascular benefit. Direct anti-inflammatory action. Serious proposals with serious people on more than one side.

A grade describes the kind of evidence, not its quality or the size of the effect. A well-run mouse study is still grade C. A grade A trial can have a small effect.

Evidence grades A to D with examples and the checklist for reading a mechanism claimLeft: an inverted pyramid of four bands. The narrow top band A, randomised human trial, example SELECT MACE reduction. Band B, human physiology, example Nauck clamp studies. Wide band C, animal or cell, example arcuate nucleus tracing in mice. Bottom band D, hypothesis or debated, example GIP agonist versus antagonist paradox. Right: a checklist of five questions: is the receptor present in that human tissue; is the evidence in people or animals; is the effect at clinical doses; is it an outcome or an intermediate marker; does anyone credible disagree.A. Randomised human trial or labelB. Human physiology or cohortC. Animal or cell onlyD. Hypothesis, debatede.g. SELECT: MACE down 20%(what happens, not why)e.g. clamp studies of insulin release(in people, intermediate measures)e.g. arcuate nucleus tracing in mice(explains; does not prove in people)e.g. GIP agonist vs antagonistBand widths are illustrative of where mechanism papers sit, not measured.Before believing a mechanism claim1. Is the receptor in that human tissue?2. People or animals?3. Clinical doses, or pharmacological?4. Outcome, or intermediate marker?5. Does anyone credible disagree?Validated mapping removed hepatocytes andadipocytes from the GLP-1 receptor list.Animal effect sizes often shrink or vanishin human trials (Perel 2007).Glucagon raises expenditure at infusiondoses; drug doses are a separate question.CRP fell in STEP-HFpEF; whether thatmediates the benefit is unproven.If yes, this site grades it D and says so.
Figure 14. The four evidence grades used on this site, with one example of each from these pages, and the questions to ask of any mechanism claim. The width of each band is illustrative of how much of the mechanism literature sits at each level, not a measured proportion.

Why animal evidence gets its own grade

It is tempting to treat a beautiful mouse experiment as an explanation of a human trial. The record says to be careful. Hackam and Redelmeier reviewed highly cited animal studies and found that only about a third translated into human randomised trials that reproduced the finding (PubMed 17032985). Perel and colleagues compared treatment effects across animal experiments and matched clinical trials for six interventions and found agreement in some and discordance in others, with methodological weaknesses in the animal work common (PubMed 17175568).

GLP-1 pharmacology has one of the better translation records in medicine: the incretin effect, glucose-dependent insulin release and appetite suppression all carried over from animals to people. That is a reason to take the animal work seriously, not a reason to skip the grade.

Five questions

Before accepting any statement of the form "GLP-1 drugs work by X", the checklist in the figure is the minimum. Is the receptor actually in that human tissue? Validated mapping in 2014 removed the liver and fat cells from the list, which retired a decade of claims (PubMed 24467746). Is the evidence from people or animals? Is the dose the one patients take? Is the measured thing an outcome or a marker? And is there a credible dissent?

Worked example: SELECT

SELECT randomised 17,604 adults and found a 20% relative reduction in major cardiovascular events on semaglutide 2.4 mg (PubMed 37952131). Grade A for the effect. The mechanism has several candidates: weight loss, blood pressure, lipids, glucose, inflammation and direct vascular effects. The event curves separated within months, before most weight was lost, and the benefit looked similar across baseline BMI, which argues against weight alone. Nothing in the trial design can distinguish the candidates. Grade D for the mechanism. Both grades are right, and a page that gives only one of them is misleading.

Where to go next

Questions people ask

Why not use an existing grading system like GRADE?

GRADE rates confidence in an effect estimate for a clinical decision. This site is grading a different thing: what kind of evidence supports a statement about mechanism. A rodent tracing study can be excellent science and still tell you nothing certain about people. The A to D scale is a label for the type of study, not a verdict on its quality.

Does grade A mean the mechanism is proven?

No. Grade A means a randomised human trial established the effect. SELECT proved that semaglutide reduces cardiovascular events; it did not prove how. On this site the outcome is grade A and the mechanism is grade D, and both grades appear on the same page.

Sources

  1. Hackam DG, Redelmeier DA. Translation of research evidence from animals to humans. JAMA 2006. PubMed 17032985 Accessed September 4, 2026.
  2. Perel P, Roberts I, Sena E, et al. Comparison of treatment effects between animal experiments and clinical trials: systematic review. BMJ 2007. PubMed 17175568 Accessed September 4, 2026.
  3. Pyke C, Heller RS, Kirk RK, et al. GLP-1 receptor localization in monkey and human tissue: novel distribution revealed with extensively validated monoclonal antibody. Endocrinology 2014. PubMed 24467746 Accessed September 4, 2026.
  4. Lincoff AM, Brown-Frandsen K, Colhoun HM, et al. Semaglutide and cardiovascular outcomes in obesity without diabetes (SELECT). N Engl J Med 2023. PubMed 37952131 Accessed September 4, 2026.

Canonical URL: https://formblendsscience.com/mechanisms/how-we-grade-mechanism-evidence. Written by the FormBlends editorial team. This page is educational and is not medical advice; see the medical disclaimer.