Skip to main content

Welcome to a Grasp lesson

Create a free account to keep studying and create your own study path.

Lesson 3

Evaluating Causal Evidence for GLP-1 Medicines

Recommended resources

Hello. The previous lesson explained how GLP-1 receptor agonists activate a receptor and remain available longer than the natural hormone. That biological account tells us how a drug could act. It does not, by itself, tell us whether the drug prevents a heart attack, changes drinking behavior, or causes a rare adverse effect in people.

This lesson is about matching the strength of a causal claim to the study that supports it. You will compare controlled animal experiments, observational studies of people, and randomized clinical trials. Plan on about 35–40 minutes, including the short videos and readings.


Start with the claim, not the study label

Consider three statements:

  • “Activating GLP-1 receptors changed feeding behavior in mice.”
  • “People prescribed a GLP-1 medicine had a lower rate of a particular diagnosis.”
  • “In a trial, people assigned a GLP-1 medicine lost more weight than people assigned placebo.”

They sound similar because each links a medicine to an outcome. But they refer to different populations, comparisons, and kinds of evidence. A causal claim says that an outcome would have been different because of the treatment, compared with what would have happened without it. We cannot observe both possibilities for the same individual at the same time, so researchers construct comparison groups.

A useful way to read any claim is to pin down four details: which drug or intervention, in whom, compared with what, and over what period. “GLP-1 medicines improve health” leaves all four unclear. “In a 68-week trial of adults meeting specified weight-related criteria, participants assigned semaglutide lost more weight on average than those assigned placebo” is narrower—and much easier to assess.

The evidence pyramid below is a familiar reminder that study designs answer questions with different degrees of control. Treat it as a starting point, not a rule that every study near the top is better for every question. A short trial may be poorly equipped to detect an exceptionally rare harm that a large, long-running observational study can investigate.

The left panel arranges study designs in an evidence pyramid; the right panel contrasts observational studies with randomized trials. The diagram is a guide to how studies are designed, not a substitute for judging the population, outcome, follow-up, and biases of an individual study.


Animal experiments: a causal result in an animal is not yet a human benefit

In an animal experiment, researchers can assign an intervention and tightly control features of the setting. If otherwise comparable mice receive a GLP-1 receptor agonist or a control treatment, and the treated mice subsequently eat less, that design can support a causal conclusion about eating behavior in those mice under those conditions. Additional experiments might test whether activity at a particular receptor or brain circuit is necessary for that effect.

That is valuable mechanistic evidence. It helps researchers develop explanations and decide which human outcomes are worth testing. Its boundary is equally important: species can differ in physiology, drug exposure, and behavior. Eating less in an experimental setting does not establish that a medicine treats a human addiction or prevents a human neurological disease. Nor does showing that a brain pathway responds to a drug establish that the pathway accounts for a clinical benefit.

The appropriate language preserves that boundary: “An animal experiment supports this mechanism in the tested animals”, rather than “the medicine has been shown to produce this clinical benefit in people.” We will return to that distinction when the course reaches research on reward pathways.



Observational studies: useful human evidence with an unresolved comparison

In an observational study, researchers do not randomly assign the medicine. They might follow people who were prescribed it and compare their outcomes with those of people prescribed something else. This can include large numbers of people in routine care and, sometimes, follow them long enough to investigate uncommon outcomes.

The difficulty is that the two groups may differ before treatment starts. Their underlying conditions, access to care, other medicines, or likelihood of receiving a diagnosis may differ. In drug research, confounding by indication is particularly important: the reason a clinician prescribes one treatment rather than another may itself predict the outcome. Statistical adjustment can make groups more comparable on recorded characteristics, but it cannot reliably remove differences that were not measured or were measured poorly.

Evidence‐based medicine—When observational studies are better than randomized controlled trials

Read the article’s explanation of observational designs to see why a human association is informative without automatically being causal.

In Section 3, “OBSERVATIONAL ANALYTIC STUDIES,” begin in the opening paragraph where the authors explain why an observational design may be preferred. Read through the limitations, ending just before “BOX 1.” Skim the illustrative kidney-disease examples; focus on how a cohort is assembled, how a case-control study begins with an outcome, and why baseline differences can persist after adjustment.

This does not make observational studies disposable. They may be the practical way to examine long-term outcomes or rare events that trials were too small or short to resolve. A carefully chosen comparison group, reliable outcome records, and results that hold up under different analyses all strengthen an observational finding. The wording should still reflect its design: “use was associated with a lower rate” is not the same claim as “the drug caused the rate to fall.”

Observational evidence is especially important for possible unintended effects. A trial built to measure weight loss might record a serious event, but have too few participants or too little follow-up to estimate its risk precisely. Post-treatment monitoring and large healthcare datasets can identify a concern worth investigating. Even then, an apparent safety signal must be checked against differences in who received the drug and how often they were examined.

Evidence‐based medicine—When observational studies are better than randomized controlled trials

Continue in the same article for the distinction between intended benefits and unintended effects.

In Section 4, “STUDY OF UNINTENDED EFFECTS,” read the first paragraph, including its qualifications about confounding. Then read the paragraph beginning “Once a treatment has been introduced into clinical practice” through the surveillance discussion. Focus on why a large observational study may reveal a rare concern that a trial misses, without assuming that every reported concern is caused by the drug.



Randomized trials: a stronger test of the assigned treatment

A randomized controlled trial assigns participants to a treatment or comparison group by chance. This makes pre-existing reasons for receiving the treatment less likely to systematically favor one group. Randomization does not make the groups perfectly identical; chance differences remain, especially in smaller trials. But it gives researchers a stronger basis for attributing a difference in measured outcomes to the assigned intervention.

Blinding, when feasible, tackles a different problem. If participants or investigators do not know the assignments, expectations are less likely to influence reported symptoms, behavior, or assessment. Neither randomization nor blinding repairs an outcome that was measured poorly.

Randomized control trial (RCT) explained

Watch “Randomized control trial (RCT) explained” from Henrik’s Lab for a visual account of assignment and blinding.

Watch random assignment for the comparison-group idea, then blinding for how expectations can affect measurement. The video describes groups as closely matched; retain the qualification that randomization reduces systematic imbalance but does not guarantee identical groups.

The article puts that advantage alongside the limits of trials. In particular, a result applies most directly to the participants, intervention, comparator, outcome, and follow-up actually studied.

Evidence‐based medicine—When observational studies are better than randomized controlled trials

Return to the article for its account of what randomization accomplishes—and what it does not.

In the discussion immediately before subsection 2.1, “Limitations of RCT,” read the randomization explanation. In subsection 2.1, read the trial limitations. Concentrate on confounding by indication, eligibility criteria, sample size, and follow-up; the intervening example uses a different class of medicines.

A weight-loss trial makes the scope of a trial result concrete. In the STEP 1 trial discussed in the next video, participants without type 2 diabetes were randomly assigned semaglutide or placebo, and both groups received behavioral support. At 68 weeks, average weight loss was greater in the semaglutide group. That supports a causal claim about the effect of assignment to that studied regimen, compared with placebo, on weight over that period in the trial population. It does not establish that every participant responded, that weight loss persists after treatment stops, or that the drug prevents dementia.

Role of GLP-1 Receptor Agonists for Weight Loss

Watch the STEP 1 portion of “Role of GLP-1 Receptor Agonists for Weight Loss,” from Harvard Medical School Continuing Education, as a concrete example of a randomized GLP-1 trial.

Watch the STEP 1 account. Note the eligibility criteria, placebo comparison, behavioral counseling in both groups, 68-week endpoint, and group-average results. Use those particulars to keep the conclusion no broader than the trial.



Put the evidence types side by side

Imagine an article claiming that a GLP-1 medicine “reduces the risk of a rare cancer.” Each of the following findings would contribute something different:

Finding What it can support What it cannot establish alone
A controlled rodent experiment finds a biological response A response under the tested animal conditions; a possible mechanism to investigate That the same cancer outcome occurs, or is prevented, in humans
A large observational study finds different cancer rates among medicine users and comparators A human association and a possible safety or benefit signal That the medicine caused the difference, without adequately addressing bias and confounding
A randomized trial records very few cancer cases What occurred during its particular follow-up; potentially stronger causal evidence if enough events accrue Reliable reassurance about a very rare cancer that may take years to develop

A short review illustrates why the verdict cannot be based on a study label alone. It describes biological plausibility for a particular thyroid-cancer concern in rodents, while noting that thyroid cancer events were rare in trials and estimates were imprecise. Observational findings were inconsistent, and increased reporting of a suspected event was not itself evidence of causation. This is an example of uncertainty to characterize, not a conclusion about personal treatment.

Glucagon-Like Peptide-1 Receptor Agonists and Thyroid Cancer: A Narrative Review

Read the review’s Summary to see several evidence types assessed for one safety question.

In “Summary,” start with the sentence beginning “The body of evidence assessing the association.” Read the design comparisons. Continue from the sentence beginning “There is biological plausibility” through the human evidence, then read the final sentence about increased reporting. Keep separate a rodent finding, a rare outcome in trials, an observational association, and a reporting signal.

The same discipline applies to a possible benefit. An animal study of reward circuitry can justify a human study; an observational association with fewer addiction diagnoses can add interest; a randomized trial measuring a specified addiction outcome is a more direct causal test. None licenses the blanket claim that all GLP-1 medicines treat every addiction.



Key takeaways

Animal experiments can test causal mechanisms in the animals studied, but translation to human outcomes needs human evidence. Observational studies can reveal associations in large, real-world populations and help investigate rare or delayed outcomes, while remaining vulnerable to differences between treatment groups. Randomized trials offer a stronger test of a treatment’s causal effect on measured outcomes, but their conclusions remain bounded by who was studied, what was compared, and how long participants were followed.

When you encounter a GLP-1 headline, first identify its exact claim and then match its wording to the study design. The next module returns to biology, beginning with how GLP-1 receptor activation affects insulin and glucagon when blood glucose changes.