Different analyses of the same endpoint can produce different statistical conclusions. The early MMPOWER trial should be read with those analyses clearly labeled. Primary report
Study snapshot
Thirty-six adults with primary mitochondrial myopathy entered a randomized, blinded, dose-escalation study. After five days, the highest-dose walking-distance comparison had P = .053. A dose-trend analysis and a further covariate-adjusted model were significant; the full report describes the latter model as post hoc. Other efficacy measures did not show significant differences. Primary publication
Our appraisal: adjustment needs a transparent role
Adjustment is not inherently inappropriate. The important questions are whether the model was prespecified, why its variables were chosen, and how robust the result is to reasonable alternatives. A model developed after observing outcomes has a different evidentiary role from the original planned comparison.
The study should therefore not be summarized by selecting only the smallest P value. Readers need the unadjusted comparison, the trend analysis, and the post-hoc model together. Nor should a threshold narrowly missed be described as proof of no effect; the estimate and uncertainty remain relevant.
The publication has two corrections. One concerns an early-version dose detail; the other concerns a supplemental-figure reference. Neither should be silently ignored when locating the authoritative report. First correction, second correction
A short study can inform further development without demonstrating durable functional improvement. Full review should assess the model-selection process, baseline walking ability, missing observations, and the entire subsequent evidence program before making a clinical claim.
Frequently asked questions
Was the highest-dose primary comparison conventionally significant?
The reported P value was .053.
Did additional analyses favor treatment?
Yes. Their different analytical roles must remain visible.
Was the further adjusted model prespecified?
The full report describes the final model as post hoc.
Does post hoc mean automatically false?
No. It means greater caution and confirmation are needed.
Are there published corrections?
Yes, two notices linked above.
Did five days establish durable benefit?
No.
What is the main reading rule?
Do not replace the planned comparison with the most favorable later analysis.