At a Glance

A drug can move a number while the patient gets no better.

1. Part 1Each patient's plaque change explained nothing of the drug's effect (p = 0.884).
2. Part 2Predicting who gets sick and moving when a treatment changes are two different jobs.
3. Part 3A negative-control trial checks whether the marker can move without the patient.

One trial, or one person, cannot show a number stands in for health.

1. Part 1Two trials built the same way: one yes (p=0.01), one no (p=0.83).
2. Part 2Blood pressure earned its trust as strokes fell in trial after trial.
3. Part 3Make each dot one randomized trial, not one patient.

A stand-in for aging must hold for every kind of drug.

1. Part 2An aging stand-in would need the word on blood pressure's row: mechanism agnostic.
2. Part 3Hold back every trial of one mechanism and test whether the line still predicts them.

Approving on a stand-in only "reasonably likely" to work raises doubt about the next one.

1. Part 1The plaque fell in both trials, and the rule let the plaque stand in.
2. Part 2The plaque's row says Accelerated: the lower yes.
3. Part 3Doubt raises the bar for any stand-in, and a higher bar means fewer approvals.

The test that counts is the one named before the numbers came in.

1. Part 1A new reading of the rules put the chance of a false win as high as .0975.
2. Part 3Part one's tests ran on the plaque after the fact; here they are set before any data.