When the Primary Endpoint Fails and a Secondary Doesn't

A randomized trial gave oral semaglutide or placebo to 72 people with major depressive disorder, to see whether it improved the cognitive difficulties that often come with depression. It did not improve the outcome it was built to test. A secondary measure did come back significant, and that is the one you will hear about.^[1]

What the trial found

The primary outcome was a composite of the digit symbol substitution test, the Stroop test and an n-back task — standard measures of executive function. Its confidence interval runs from clearly negative to clearly positive, which is what a null result in a small trial looks like.^[1]

Why the secondary result should not be the headline

A pre-specified primary endpoint exists precisely so that a trial cannot go looking for a win after the fact. When the primary fails and a secondary succeeds, three things are true at once, and only reporting the third is how a null trial becomes a positive headline.

1. The trial answered its question, and the answer was no. That is the finding.
2. Seventy-two people is small. With several secondary outcomes measured, one clearing p=0.05 is unremarkable arithmetic.
3. A p of 0.03 with a lower bound of 0.19 is a result that barely clears the line, in a measure that was not the target.

When the primary fails and a secondary succeeds, the honest headline is the failure. The secondary is a suggestion for the next trial.

That is not a criticism of the researchers, who reported all of it plainly and describe the secondary results as suggesting effects on specific domains — appropriately cautious language. It is a warning about what happens to a result like this on its way to a reader.

The two findings that are actually useful

It did not improve depression. Depressive symptom severity was unchanged. Anyone hoping these drugs treat mood directly, rather than the metabolic burden that often accompanies it, has an answer here.

It did not increase suicidal ideation. The frequency of suicidal ideation was unaffected, and there were no serious adverse events.^[1] That is a small data point in a much larger question we cover in what happened with the suicidality signal — and it comes from a population with major depressive disorder, which is exactly where the concern is sharpest.