“Cognition improved” can hide important differences between tests. This randomized study is best read through its separate cognitive domains and analysis populations. Primary report
Study snapshot
Older adults with normal cognition or mild cognitive impairment received tesamorelin or placebo in a double-blind, 20-week trial. The overall intention-to-treat cognitive analysis favored treatment. Executive-function results were significant, verbal-memory results showed a trend, and visual memory did not show a significant effect. The report also examined outcomes after a washout period. Its registration is NCT00257712. Primary report
Our appraisal: do not turn an omnibus result into three victories
An overall multivariate test asks about a collection of measurements. It does not guarantee that every component improved. A useful evidence summary names the stronger and weaker findings instead of replacing them with an undifferentiated claim about memory.
Analysis populations matter too. Results among participants who completed follow-up can differ from results using the randomized population. The reason for that difference should be investigated, not resolved by selecting whichever analysis produces the smallest P value.
The diagnostic subgroups introduce a second interpretive boundary. Failure to detect a treatment-by-diagnosis interaction does not establish equal effects in people with and without impairment. Subgroup estimates and uncertainty are needed to assess that question.
Finally, performance on cognitive tests is not identical to prevention of dementia, preservation of independence, or a durable effect after treatment ends. Those claims require outcomes and follow-up designed for them. Full review should connect the statistical hierarchy to the protocol and include adverse events before drawing a benefit-risk conclusion.
Frequently asked questions
Were participants and investigators blinded?
The report describes a double-blind, placebo-controlled design.
Did all cognitive domains improve significantly?
No. The domain results differed.
Was the executive-function finding the same as visual memory?
No. Their statistical results should remain separate.
Can a trend be described as established benefit?
No. Preserve the uncertainty.
Does a nonsignificant interaction prove subgroup equivalence?
No. It does not establish that effects are identical.
Did this establish dementia prevention?
No. That is a different clinical claim.
What should readers check next?
The endpoint hierarchy, missing-data approach, washout findings, and complete safety reporting.