Case Study 1 — SELECT and the Question It Could Not Answer
An excellent trial, read for what it establishes and what it leaves open
Type: Real, public, recent · Tier 1–2 · Relevance: §8.5, §8.6
Background: the problem with weight as an endpoint
By 2020, semaglutide's weight effect was established. STEP had shown roughly 15% mean weight reduction at 68 weeks in adults with overweight or obesity without diabetes.
And that was not enough, for a reason Chapter 5 §5.6 explains.
Weight is a surrogate endpoint. It is measured because it is believed to predict outcomes — cardiovascular events, diabetes, mortality — that take years and thousands of participants to observe directly. The belief is well supported: the association between excess adiposity and cardiovascular risk is one of the most robust in epidemiology.
But obesity pharmacology has a specific and ugly history with surrogates. Drugs have reduced weight and damaged heart valves. Drugs have reduced weight and increased cardiovascular events. Drugs have reduced weight and produced psychiatric harm. In each case, the surrogate moved in the desired direction while the thing that mattered did not.
So the question SELECT was built to answer was not "does semaglutide reduce weight" — that was settled — but "does it reduce the events that weight is a surrogate for?"
The design
Population: adults with established cardiovascular disease and overweight or obesity, without diabetes. That last exclusion is deliberate and important: it isolates the question from glycemic control, which is a confounder in any trial including people with diabetes.
Size and duration: roughly 17,000 participants, followed about three years.
Intervention: semaglutide 2.4 mg weekly versus placebo, both arms on standard cardiovascular care.
Primary endpoint: MACE — a composite of cardiovascular death, non-fatal myocardial infarction, and non-fatal stroke. Pre-specified.
Funding: the manufacturer.
Note the design choices that make this a strong trial: a hard endpoint rather than a surrogate; a population at high enough baseline risk that events would accrue in reasonable time; a duration long enough for that to happen; and a size large enough to detect a moderate effect.
The result
A 20% relative reduction in MACE. Hazard ratio about 0.80.
In absolute terms: roughly 8% to roughly 6.5% over about three years — about 1.5 percentage points.
Both figures describe the same finding. Chapter 5 §5.8 exists because of exactly this pair, and the teaching point is worth restating: a reader given only the first overestimates the personal benefit by roughly a factor of ten; a reader given only the second underestimates the population importance.
Number needed to treat, roughly: about 65 to 70 people treated for approximately three years for one additional person to avoid a major cardiovascular event. For a serious outcome in a high-risk population, that is a clinically meaningful figure. It is also not a miracle, and the honest communication contains both sentences.
Why it reorganized the field
It converted a metabolic drug into a cardiovascular one. Guidelines, coverage decisions, and prescribing patterns respond differently to "reduces weight" and "reduces heart attacks." The distinction is not rhetorical: reimbursement systems that had declined to cover obesity treatment had much greater difficulty declining to cover cardiovascular risk reduction.
It answered the surrogate question in the right direction. Given obesity pharmacology's history, that was not a foregone conclusion, and a null or adverse result was entirely possible.
And it did so in a population without diabetes, which removed the most obvious alternative explanation.
The question it could not answer
Why.
SELECT establishes that semaglutide reduces cardiovascular events in this population. It does not establish how, and the candidate mechanisms are numerous and not mutually exclusive:
CANDIDATE MECHANISMS FOR THE CARDIOVASCULAR BENEFIT — none established
① WEIGHT LOSS itself
Less adiposity → less mechanical and metabolic burden. The obvious candidate.
PROBLEM: the benefit appeared to emerge relatively early, and the size of the
effect may exceed what the weight change alone would predict.
② IMPROVED GLYCEMIA
Even without diabetes, glucose handling improved.
PROBLEM: the population was non-diabetic, limiting how much room there was.
③ BLOOD PRESSURE
Modest, consistent reductions were observed.
PROBLEM: the magnitude seems small relative to the outcome effect.
④ LIPIDS
Modest improvements observed.
PROBLEM: same.
⑤ ANTI-INFLAMMATORY ACTION
Inflammatory markers fell. GLP-1 receptors are present on immune cells and
vascular tissue.
PROBLEM: inflammatory markers are among the least reliable surrogates in
medicine (Ch 5 §5.6). This is hypothesis-generating, not explanatory.
⑥ DIRECT VASCULAR AND CARDIAC EFFECTS
GLP-1 receptors are present in heart, kidney, and blood vessels (Ch 7).
PROBLEM: their functional significance in humans is not established.
A trial with a composite outcome and no mechanistic sub-study cannot
distinguish among these. SELECT was not designed to.
Why this matters more than it appears to.
It limits extrapolation. If the benefit is mediated by weight loss, any intervention producing comparable weight loss should produce comparable benefit — including diet, exercise, and surgery. If it is mediated by a direct receptor effect, it would be specific to the drug class. Those imply very different things, and SELECT cannot distinguish them.
It affects who else might benefit. Chapter 10's expanding list of indications — kidney, liver, possibly brain — is much more plausible if a shared mechanism such as inflammation is operating than if each benefit is a downstream consequence of weight loss.
And it is a reminder that clinical efficacy and mechanistic understanding are independent. Chapter 2 made this point with aspirin and insulin. Here it is in a drug approved this decade: one of the best-evidenced cardiovascular results in modern medicine, and nobody can tell you confidently why it happened.
⚠️ Hype Check — "it reduces heart attacks by 20%"
What's true: the relative reduction is real, the trial was large and well-designed, the endpoint was hard and pre-specified, and the population was well defined.
What is missing without the absolute figure: that the event rate fell from roughly 8% to roughly 6.5% over about three years, meaning roughly 65 to 70 people were treated for three years for one additional person to benefit. Neither number is the honest answer alone.
What is missing regardless: the population. This was people who already had cardiovascular disease and were overweight or obese and did not have diabetes. It says nothing about primary prevention, and the absolute benefit in a lower-risk population would be smaller — possibly much smaller — while the adverse effects would be the same.
What is also missing: the mechanism, which limits how far the result can be extended to anyone the trial did not study.
The honest sentence: "In people who already have cardiovascular disease and are overweight, semaglutide reduced major cardiovascular events by about a fifth over three years — from about eight in a hundred to about six and a half in a hundred. We don't yet know why, which limits how confident we can be about people who differ from the ones studied."
Discussion questions
-
SELECT excluded people with diabetes. Explain the design reasoning, and identify what that exclusion costs in terms of generalizability.
-
The trial used a composite endpoint (MACE). Chapter 5 §5.10 lists composite endpoints as a potential red flag. Is MACE a legitimate composite? What would make a composite illegitimate?
-
Six candidate mechanisms are listed and none is established. Design a sub-study or follow-on trial that would distinguish between mechanism ① (weight loss) and mechanism ⑥ (direct vascular effect). Is your design feasible?
-
If the benefit turns out to be mediated primarily by weight loss, what follows for someone who achieves comparable weight loss by other means? If it is mediated by a direct receptor effect, what follows instead?
-
SELECT was manufacturer-funded. Chapter 5's rule is that funding changes the weight of evidence, not its truth value. What specifically would you check, and does the trial's design address any of those concerns structurally?
-
Write the honest sentence. In one sentence a patient could understand, state SELECT's finding with its population, both effect measures, and its principal limitation. Then count the words and consider why the short version wins in practice.