Five days to the abstract deadline and the numbers moved
An investigator-initiated, 14-site trial of a perioperative antibiotic regimen has an abstract deadline at a major surgical congress in five days. The first author, a second-year fellow, drafted the abstract from an interim snapshot (n=231 of a planned 300) and reported a primary-endpoint odds ratio of 1.85, p=0.02. The sponsor's biostatistician has since run the locked analysis on n=298: the OR is 1.42 and p=0.11. One site contributing 26 patients is under an open GCP query and may yet be excluded. The draft has already been circulated to 11 co-authors, and the department chair has posted on social media that the trial 'showed a strong positive effect'. The congress embargo policy prohibits prior public disclosure of results. A co-author emails you, the publication lead, asking whether the team can 'just submit the interim numbers and update them at the meeting'.
1. What do you do about the abstract in the next five days? Give your reasoning and the sequence of actions.
Read the worked answer
The proposal is not acceptable. Submitting interim estimates while knowing a locked analysis exists, and then 'updating at the meeting', is a misrepresentation of the evidence at the point of publication. The abstract is a citable, indexed record; readers and guideline panels will use the number that was submitted, not the one corrected on a slide.
Sequence.
- Same day: confirm with the trial statistician which dataset is authoritative and whether the open GCP query affects the primary analysis population. Get this in writing.
- Rebuild the abstract from the locked n=298 analysis, with the pre-specified analysis population clearly described and the site under query handled per the SAP, not per convenience.
- If the locked result is null, the abstract reports a null result. A congress abstract reporting a well-conducted neutral trial is publishable; a false positive is not recoverable.
- If the locked analysis cannot be finalised before the deadline, do not submit. Missing one congress is a scheduling cost; a retraction is a career and institutional cost.
- Circulate the corrected draft to all 11 co-authors with a covering note explaining why the numbers changed, and require written approval before submission.
Judgement point. Do not average the two analyses or present a range to preserve the positive framing. The honest option is a clean null result, or no abstract.
2. The department chair has already posted on social media about a 'strong positive effect', and the congress embargo applies. How do you handle this?
Read the worked answer
Two separate breaches to address. First, the chair's post is a public disclosure of trial results before presentation, which conflicts with the congress embargo and could jeopardise the abstract. Second, the post states a result the locked analysis no longer supports.
Immediate actions. Contact the chair directly and privately, explain both problems, and ask for the post to be deleted or corrected promptly. In parallel, notify the congress publication team proactively rather than waiting to be challenged - self-reporting is treated far more leniently than discovery. Ask whether the abstract still qualifies; embargo policies vary on whether a pre-presentation social post is disqualifying.
Correcting the record. A deletion alone is insufficient if the post was widely shared. Agree a short factual statement with the chair and the institution's communications office: the analysis is ongoing and no result is final. Do not repeat the OR in any correction.
Governance follow-up. Social media is the most common route to accidental unblinding. Add a one-page rule to the author pack: no posting of trial results before presentation; direct queries to the publication lead. Brief the chair's office, the institutional press office and the co-authors in the same week.
Why it matters beyond this abstract. Embargo breaches get institutions excluded from future late-breaker slots, and a chair's post carries more weight with journalists than a press release. Treat it as a media incident, not an HR matter.
3. What process changes would you propose to the trial team to prevent a repeat on the next analysis?
Read the worked answer
Fix the handoff, not the person. The failure was structural: the fellow had no authoritative dataset and no gate before circulation.
Controls to propose.
- Single source of truth. Only the statistician's locked output may be used in any external document. Interim snapshots are watermarked 'not for external use' and never circulated to authors.
- Data verification sign-off. A one-page checklist, signed by the statistician and the PI, confirming the analysis population, the N, and the primary estimate before any abstract or manuscript leaves the team.
- Freeze date. Set the analysis freeze at least 10 working days before any abstract deadline so there is time to absorb a change of result.
- Author pack. Distribute the congress embargo policy and a plain statement that results may not be posted or discussed publicly before presentation, with named contacts for queries.
- Open queries register. The publication lead should see a list of sites with unresolved GCP queries before, not after, an abstract is drafted.
Culture point. The fellow did nothing dishonest; the fellow was given an interim number and a deadline. Say so in the review, or the next junior author will hide a data change instead of raising it.
Measure. Track how often an abstract's numbers differ from the locked analysis between draft and submission. The target is zero, and the metric makes the risk visible to the PI.