Chapter 10 — Key Takeaways

Beyond Weight Loss

The core claims

A drug that appears to work on everything should make you suspicious — and in this case some of the claims have completed randomized trials with hard endpoints, which is a genuinely different situation from supplement marketing. The discipline is to hold the suspicion while evaluating each claim separately.

Four candidate explanations for the breadth, none established, probably all operating:

  ① weight loss is upstream of many diseases
  ② GLP-1 receptors are present in heart, kidney, vasculature, immune cells
  ③ inflammation is a common thread
  ④ some of these will not replicate

FLOW (kidney) was stopped early for efficacy — genuinely good news and a statistical complication. Trials that cross an interim boundary are enriched for chance-favorable results. Direction trustworthy; magnitude held loosely.

MASH has a biopsy endpoint problem: invasive, sampling error, reader variability, and still a surrogate for cirrhosis and death, which take years to accrue.

Sleep apnea is the clearest weight-mediated benefit — less soft tissue around a collapsible airway. That is a complete explanation, not a criticism, and it makes a testable prediction.

Alzheimer's and addiction are 🔬 because their supporting evidence is mechanistic, preclinical, and observational — not because the research is illegitimate. ⚠️ requires real randomized human data that doesn't settle the question; 🔬 is for early science proceeding properly.

"Anti-inflammatory" is the vaguest mechanism claim in medicine. A mechanism that can explain benefit in any disease explains benefit in none, until someone specifies which process, in which tissue, measured how.

The ratings — one class, ten claims

A single overall rating for this class would have to be wrong about at least four of these.

The two compression errors

Upward compression — treating the ✅ indications as licensing the 🔬 ones. "It's proven for heart disease, so the Alzheimer's thing is probably right."

Downward compression — treating the 🔬 indications as discrediting the ✅ ones. "They claim it treats everything, so I don't believe the heart result."

Both are the same error: attaching a rating to a molecule rather than to a claim. Rule 6 prevents both.

Two selection effects, one logic

Chapter 6 (testimonials) · Chapter 9 (Phase 2) · Chapter 10 (interim stopping):

When something is reported because it crossed a threshold, the reported value is enriched for chance.

And from Case Study 2, the observational asymmetry:

Observational data overestimates drug benefits — because clinicians choose who gets treated — and is genuinely strong for detecting harms, because nobody selects patients for a bad outcome.

The reflex: on any claim that a drug prevents a disease, ask "randomized or observational?" first.

Key terms

pleiotropic effect · indication expansion · FLOW · MASLD / MASH · biopsy endpoint · albuminuria · eGFR · HFpEF · obstructive sleep apnea · apnea-hypopnea index · stopped early for efficacy · confounding by indication · healthy adherer effect · immortal time bias

What you can now evaluate

• ✅ each indication in this class separately, with its own population and endpoint
• ✅ what "stopped early for efficacy" costs you
• ✅ why a biopsy endpoint is both invasive and a surrogate
• ✅ when "it's just the weight loss" is a complete explanation and when it is a confound
• ✅ why an observational dementia signal is a reason to run a trial and not a result
• ✅ your own upward compression, in your own dossier
• ❌ not yet: the drug against which all of this should be measured. Chapter 11.

The one-sentence version

One molecule, ten claims, five ratings — and the two ways to get this wrong are opposite to each other and identical underneath.

Next: Chapter 11 — insulin. A century of evidence, and the standard everything in Part II is measured against.