Differential Attrition and a Switched Primary Outcome in a Supplement Trial
A manufacturer-funded trial randomized 400 adults with chronic fatigue, 1:1, to 12 weeks of a berry-derived polyphenol supplement or a color- and taste-matched placebo. The pre-specified primary outcome was distance walked in 6 minutes. After enrollment, the primary outcome was changed to a self-reported fatigue score (0–10) at week 12. Baseline fatigue was 7.2 in the supplement group and 6.5 in the placebo group (p = 0.08). At week 12, 84% of the supplement group and 68% of the placebo group had completed the trial; dropouts in both arms had higher baseline fatigue than completers. Among completers, fatigue fell 2.4 points with the supplement and 1.1 points with placebo (p = 0.02); six-minute walk distance did not differ. Mild gastrointestinal upset was reported by 12% of supplement users and 5% of placebo users. The active polyphenol has a plasma half-life of about 2 hours, and participants took their last dose the evening before the week-12 morning assessment.
1. Rank the threats to internal validity and explain the direction in which differential attrition biases the reported effect.
Read the worked answer
Differential attrition is the most serious threat. The 16-point completion gap is large, and dropouts in both arms had higher baseline fatigue. Completers in the placebo arm are therefore a selected subgroup with lower baseline fatigue and less room to improve, while the supplement arm retained more of its higher-fatigue participants. That selection would make the placebo response look smaller than it should, inflating the apparent supplement advantage. The baseline imbalance reinforces the problem: 0.7 points is more than half of the reported 1.3-point difference, and p = 0.08 is not evidence that the groups are equivalent. Unblinding through gastrointestinal upset is a third concern, since participants who experience it may infer active assignment and report benefit accordingly. The post-enrollment outcome switch compounds all of this, because the objective outcome that was originally specified showed no difference and the subjective one did.
2. What analyses would you require, and why is a completers-only comparison inadequate?
Read the worked answer
The reported comparison appears to be completers-only, which preserves the randomization only if completion is unrelated to outcome, and here it is clearly related. Intention-to-treat analysis of all 400 randomized participants is required, with missing outcomes handled by a pre-specified method: multiple imputation or a mixed model for repeated measures under a missing-at-random assumption, plus a worst-case or tipping-point sensitivity analysis that asks how different the missing outcomes would have to be to erase the effect. Baseline fatigue should be adjusted by ANCOVA rather than left to chance. Both the original and the switched outcomes should be reported in full. Adherence and plasma metabolite levels should confirm exposure. If the effect survives these analyses, it becomes credible; if it shrinks toward the 0.7-point baseline gap, the conclusion should be withdrawn.
3. Considering the pharmacokinetics and the trial conduct, what should the investigators conclude, and what would a well-designed follow-up require?
Read the worked answer
With a 2-hour half-life, a dose taken the previous evening is more than five half-lives old by the morning assessment, so less than about 3% of the parent compound remains and an acute pharmacological effect at the moment of measurement is implausible. A chronic effect is possible but would need a mechanism and a dose–exposure relationship. Combined with the industry funding, the outcome switch after the objective measure failed, and the differential attrition, the appropriate conclusion is that efficacy is unconfirmed, not that the supplement reduces fatigue. A defensible follow-up would pre-register the outcome, randomize enough participants to power the smallest clinically important difference on the fatigue scale, use an active placebo to mimic gastrointestinal effects, measure plasma metabolites to document exposure, analyze by intention to treat with pre-specified missing-data handling, and be funded and analyzed independently. The minimum clinically important difference on that scale should be stated before the trial, because a 1.3-point self-reported change may or may not be meaningful.