A statistically detectable average change does not directly tell readers how many people experienced a meaningful improvement. Responder analyses address that question, but their thresholds require scrutiny. Primary study
What this paper added
The report reanalyzed a bremelanotide phase 2b trial using seven patient-reported endpoints. It combined planned expert-based thresholds with post hoc approaches involving global benefit ratings, receiver-operating-characteristic curves, and cumulative distributions. Reported responder comparisons favored the selected active-treatment group. These were analyses of the existing trial, not newly enrolled participants. Primary report
Our appraisal: the threshold is part of the result
Classifying a person as a responder turns a continuous change into a yes/no outcome. Small movements around the cutoff can change the classification even when the underlying experiences are similar. Showing the full distribution alongside the threshold helps readers see that sensitivity.
An expert estimate and a patient-reported anchor provide different kinds of justification. Their agreement may be reassuring, but neither should be presented as an immutable universal boundary. Relevance can depend on population, baseline severity, follow-up, and the wording of the questionnaire.
Deriving and evaluating a cutoff in the same dataset risks optimistic performance. Independent validation is valuable, particularly when several endpoints, subgroups, or candidate thresholds were examined. Planned and exploratory analyses should remain labeled separately.
This paper should be linked to its parent phase 2b report rather than counted as another independent efficacy trial. It contributes a distinct interpretation of clinical meaning, not a second randomization. Later use of a responder definition can test its utility, but that does not retroactively make every original exploratory choice prespecified.
The study is informative about measurement and endpoint interpretation in a selected population. It does not establish that everyone meeting a cutoff has the same benefit or that a questionnaire response alone captures the complete benefit–risk profile.
Frequently asked questions
Was this a new independent trial?
No; it reanalyzed an existing trial.
Were all threshold analyses prespecified?
No.
Does dichotomizing preserve all information about change?
No.
Can a cutoff depend on population and context?
Yes.
Why show cumulative distributions?
They reveal response patterns beyond a single cutoff.
Is same-dataset threshold validation fully independent?
No.
What should evidence summaries avoid?
Counting the parent trial and responder paper as separate participant cohorts.