How the verdict is built

linked from every drug and comparison page
01 · DIVIDE

By route, not one big list

Oral and injectable are graded as separate divisions, so a best-in-class pill is never ranked against an injection. Not because the weight-loss numbers cannot be compared across routes, they can, and we compare them directly on the oral vs injectable page. It is because a single cross-route grade would smuggle in a verdict on the needle-versus-pill trade-off, and that trade-off is yours to price, not ours.

02 · STANDARDIZE

The same question for everyone

Placebo-adjusted loss, on the conservative estimand that counts everyone prescribed the drug, at the dose you can actually be prescribed (or the pivotal trial dose while a drug is still investigational), near a 68-week window. Off-window numbers are shown and flagged, never extrapolated. Because an approved dose is often lower than the trial ceiling, a drug’s number can step down when it is approved: that is the board getting more honest, not the drug getting worse.

03 · SPLIT

Efficacy and tolerability apart

Two scores, never blended. Convenience stays a labeled profile, because the needle-versus-pill trade-off is your value judgment, not a fact we bake in.

04 · GRADE THE EVIDENCE

Confidence caps the verdict

Four levels, top to bottom. A regulator clearance is the ceiling: those read Approved. A Phase 3 result, topline or published, earns Certified and a place on the board. Below the line, Projected (only Phase 2) and NR (too early to rate) stay on the watchlist, never mixed into the ranking: a hyped number cannot be crowned until a Phase 3 readout lands.

The two ideas that do the heavy lifting

Placebo-adjusted

Credit the drug, not the diet

Every obesity trial runs a diet-and-lifestyle program in both arms, so the placebo group already loses a few percent. We report the gap above placebo, not the raw number: a drug whose arm lost 15% against a placebo arm that lost 3% really delivered 12. Absolute numbers flatter drugs tested against weaker placebo arms; the gap is the fair comparison.

The conservative estimand

Count everyone, not just the adherent

There are two ways to score a trial. One counts only the people who stayed on the drug: the bigger number, and the one companies headline. The other counts everyone randomized, including those who stopped. It runs lower, it is more honest, and it is what regulators prefer, so it is our primary. Sponsors name it differently: Novo calls it the “treatment policy” estimand, while Lilly and Boehringer call it the “treatment regimen” estimand. They are the same idea, and we check each trial’s analysis plan to use it. When a drug has heavy dropout the two numbers diverge hard, which is the entire story of the example below.

Worked example: survodutide

How a 16.6% headline becomes a Lagging verdict: the estimand and the tolerability cost, one step at a time.

16.6%
The headlineabsolute loss, efficacy estimand
13.4
Subtract placebo16.6% − 3.2%, same estimand
7.6
Switch to treatment-regimen13.0% − 5.4%: the quitters now count
Heavy
The tolerability costabout 20% stopped for GI (+17.3)
C
Verdict: Lagging, Certified. The evidence is solid Phase 3: we do not doubt the trial. But once the dropouts count and the GI cost is shown, a headline-grabbing 16.6% is really a modest, hard-to-tolerate result. This is the drug the whole method exists to catch.

← Back to the board