What 389 trials reveal about outcome switching

Prospective trial registration exists specifically to prevent a particular kind of quiet distortion: a trial that measures several things, and then, once the results are in, presents whichever measure looked most favourable as if it had been the plan all along. A study examining 389 trials with clearly prospectively registered primary outcomes, drawn from a random sample of 1,488 articles out of 29,749 publications searched, measured how often this actually happens, and what it does to the reported results when it does.

A third of trials didn't publish what they registered

The study found that 33.4% of trials, 130 of the 389 examined, experienced at least one primary outcome change between registration and publication. That's a substantial share, roughly one in three, and it's worth being precise about what kind of change was actually most common: 66 of the 130 changes, just over half, were cases of not reporting or simply omitting the registered primary outcome from the publication entirely, rather than an explicit, acknowledged substitution of one outcome for another.

That distinction matters. A trial that clearly states "we changed our primary outcome from X to Y, and here's why" is at least being transparent about the change, even if the change itself raises questions. A trial that registers primary outcome X and then publishes results without ever mentioning X again is doing something quieter and harder for a reader to catch, since nothing in the publication itself flags that anything was altered. A reader would need to actively check the registry entry against the publication to notice the gap at all.

The trials that changed outcomes reported bigger effects

The more consequential finding is what happened to the actual reported results. Trials with a primary outcome change reported intervention effect sizes 16% larger than trials without such a change, a pooled ratio of odds ratios of 0.84 (95% CI 0.73-0.96). This inflation held up even after adjusting for potential confounding factors, with the adjusted analysis showing a similar pattern (0.81, 95% CI 0.71-0.93).

That's the mechanism undue outcome switching is specifically feared to enable, demonstrated with actual comparative data rather than left as a theoretical risk. Trials that altered what they measured as their primary outcome, whether through outright substitution or simply not reporting the original measure, systematically reported more favourable results than trials that stuck with their registered plan. The size of the effect, 16% larger, is not a trivial rounding difference; it's large enough to meaningfully change how a reader, or a subsequent meta-analysis, would interpret the treatment's apparent benefit.

Why this isn't necessarily about deliberate manipulation

It's worth being fair to the range of reasons an outcome might genuinely change between registration and publication. A measurement instrument might turn out to be impractical to administer as planned. A data collection problem might make the originally registered outcome unusable. A protocol amendment, properly reviewed and approved, might legitimately update the primary outcome mid-study for a defensible scientific reason. None of these scenarios is inherently improper, and prospective registration doesn't forbid legitimate, well-documented change.

What the data does suggest is a systematic pattern across the trials examined, not that every individual case of outcome switching was improper, but that the population of trials which did switch outcomes reported meaningfully better results than the population that didn't. Whether that pattern reflects a mix of legitimate methodological necessity and less legitimate outcome shopping isn't something this kind of aggregate comparison can fully disentangle on a case-by-case basis. What it does establish is that the switching itself correlates with more favourable reporting closely enough, and consistently enough, to be a genuine red flag worth checking whenever it appears.

What this means for how outcome changes get handled and reviewed

A few practical implications follow directly:

• Any change to a registered primary outcome should be documented explicitly in the eventual publication, with a clear rationale, rather than the outcome simply disappearing from the reported results. The finding that half of all changes in this study were omissions rather than acknowledged substitutions is precisely the pattern that undermines a reader's ability to evaluate the trial's results fairly.
• Reviewers and readers evaluating a trial's results have good reason to check the registry entry against the publication as a standard step, not an optional extra. Given that outcome-switching trials in this study reported effects 16% larger on average, a discrepancy between registry and publication is a specific, checkable signal worth investigating before accepting a trial's reported effect size at face value.
• A protocol amendment changing the primary outcome should be captured with the same rigour as any other substantial amendment, timestamped, justified, and reflected consistently across the registry, the protocol, and eventual publications. Treating an outcome change as a footnote rather than a tracked, documented decision is exactly what allows the omission pattern this study found to happen unnoticed.
• Systematic reviewers and meta-analysts should treat outcome-switching trials as a specific category worth flagging or adjusting for, given the demonstrated correlation with inflated effect sizes, rather than folding them into a pooled analysis without accounting for this known bias.

The broader point is that prospective registration only delivers its intended protection against outcome shopping if changes to what's registered are tracked, disclosed, and scrutinised as carefully as the results themselves. A third of trials in this study didn't publish what they registered, and the ones that didn't reported meaningfully rosier results. That's not a coincidence worth ignoring.