Improvement measured inside one study arm is not the same as improvement caused by treatment. This trial provides a concrete example. Primary report
Study snapshot
The open-label trial randomized 73 people with suppressed HIV, abdominal obesity, and neurocognitive impairment to tesamorelin or standard care for six months. The cognitive change comparison between groups was not statistically significant, with P = .673. Waist circumference favored tesamorelin, but the investigators did not establish a corresponding cognitive benefit. They identified limited power and the absence of placebo as limitations. Primary report
Our appraisal: two separate outcome stories
It would be misleading to compare a smaller within-group P value in the intervention arm with a larger one in the comparison arm and call that a treatment difference. The relevant test compares the changes directly. A before-and-after pattern can also contain practice effects, background care, and random variation.
The waist measurement addresses another question. A favorable physical measurement cannot rescue an unsuccessful cognitive comparison simply because both outcomes were collected in the same participants. A mechanism linking them would require its own analysis and supporting evidence.
Limited power also needs balanced language. It leaves uncertainty about smaller effects; it does not convert the observed null comparison into evidence that treatment works. Conversely, the result should not be generalized into proof of no cognitive effect in every population or setting.
For review, prioritize the cognitive endpoint definition, baseline balance, missing data, assessor blinding, and prespecified analysis. The practical evidence conclusion should follow the actual comparison, not the most favorable-looking change in the results table.
Frequently asked questions
Did the trial establish cognitive superiority?
No. Its reported between-group cognitive comparison was nonsignificant.
Was there a placebo arm?
No. The comparator was standard care.
Did every outcome tell the same story?
No. The waist and cognitive findings differed.
Can separate within-arm P values prove a treatment difference?
No. The direct between-group test is needed.
Does low power prove a missed benefit?
No. It identifies uncertainty, not a favorable result.
Can the waist result support a memory claim?
Not by itself.
What remains unresolved?
Effect precision, outcome completeness, and applicability beyond this particular trial population.