The Registry That Did Not Match the Paper
A systematic review team is screening trials of a wound-care dressing. For one eligible trial, the published paper reports a mean difference of −1.8 days in healing time (95% CI −3.1 to −0.5) with 210 participants. The team finds the trial's prospective registration, which lists a primary outcome of complete healing at 12 weeks, a target sample of 300, and no mention of healing time as an outcome. The paper reports complete healing as a secondary outcome with a non-significant result. The corresponding author replies that the registry entry was "an early draft" and that healing time was always the intended primary outcome; no updated registration is available. The review's protocol says registered primary outcomes take precedence when discrepancies are unexplained.
1. How should the review team classify and handle this trial?
Read the worked answer
The team should treat this as a case of unexplained outcome switching and handle it according to its protocol, not the author's assurance.
The registry entry is prospective and therefore the strongest available record of what was planned. The paper's primary outcome differs from the registered primary outcome, and the change is not documented by any amendment. The author's retrospective claim that healing time was "always" the primary outcome is not verifiable and conflicts with the contemporaneous record.
Handling:
- Record the discrepancy in the risk-of-bias assessment, specifically under selective reporting.
- Apply the protocol rule: use the registered primary outcome (complete healing at 12 weeks) for the primary synthesis, and treat healing time as a secondary or exploratory result.
- Do not silently drop the trial. Report both results and flag the discrepancy.
- Contact the registry or author for an amendment record; if none exists, state that in the review.
- Run a sensitivity analysis excluding the trial or using the reported outcome, and report whether conclusions change.
The trial is not necessarily fabricated, but the review must not let an undocumented switch determine its conclusions.
2. The paper's sample was 210 against a registered target of 300. What does this imply, and how should it affect the synthesis?
Read the worked answer
A shortfall from 300 to 210 (70% of target) suggests the trial stopped before reaching its planned sample. That matters for two reasons.
First, precision and power: the trial was designed for 300 participants. At 210, the confidence interval around the primary effect is wider than planned, and a non-significant result may reflect low power rather than absence of effect. The review should not interpret the null secondary result as evidence of no effect.
Second, stopping: an unreported early stop can inflate apparent effects, especially if the stop was triggered by interim results. The team should check whether the paper or registry explains why enrollment ended. If no explanation exists, flag it as a possible source of bias.
In the synthesis, the team should:
- Note the sample shortfall in the risk-of-bias table.
- Avoid weighting the trial's non-significant secondary outcome as if it were adequately powered.
- Consider a sensitivity analysis excluding trials with unexplained early stopping.
- In the discussion, describe the evidence as uncertain rather than concluding the dressing is ineffective.
The review's job is to characterize uncertainty honestly, not to accept the paper's framing.