Do trials that pledge to share data report their results?
Written against the design standard. On 2026-09-22 the API v2 query AREA[IPDSharing]YES returned 54887 registered trials, and the fields this design needs were present in the returned record. That is a check that the source answers, not a feasibility probe.
1. The question
A trial registration asks whether individual participant data will be shared. The answer is a pledge made before anyone knows what the trial will find. Results posting is a later, observable act of openness. Whether trials that pledged openness then behave openly, or whether the pledge is costless signaling, is a test of stated against revealed transparency on a very large sample.
2. Prior work, and the search behind the novelty claim
Log key P28.
corpus regex: (individual (participant|patient) data|IPD).{0,60}
(sharing (plan|statement)|plan to share).{0,120}(results (posting|reporting)|publication)
PubMed: ("IPD sharing"[tiab] OR "individual participant data sharing"[tiab]
OR "data sharing statement*"[tiab] OR "plan to share"[tiab])
AND ("ClinicalTrials.gov"[tiab] OR registry[tiab]) AND ("results reporting"[tiab]
OR "results posting"[tiab] OR "results posted"[tiab] OR publication[tiab] OR timely[tiab])Six PubMed records, none in the corpus. Two describe data sharing intentions in COVID-19 trials, at registration and at publication (PMIDs 33602275, 35182738); one describes sharing implementation in ophthalmology journals (PMID 37493649). None tests whether the pledge predicts results reporting.
3. Hypothesis
Trials that answer yes to the IPD sharing question post results within 12 months of primary completion more often than trials that answer no, adjusted for sponsor class, phase, and whether US law requires posting.
Null. The adjusted odds ratio is 1.
What would falsify it. An odds ratio at or below 1: the pledge carries no information about later behavior.
4. Design and unit
Retrospective cohort. One registered trial is one unit.
5. Sample
Interventional trials of phase 2, 3 or 4 with an actual primary completion date from 2017 to 2022, so that every trial has at least 36 months since completion by the data snapshot. Trials answering “undecided” are a third group, reported separately. Trials whose IPD answer was entered after primary completion cannot be identified without the registry’s change history, which is not accessible (see P30); this is carried as a threat.
6. Exposure
The IPD sharing answer as recorded at the data snapshot.
7. Outcome
Results first posted within 12 months of primary completion, from resultsFirstPostDate and primaryCompletionDate, full dates (S7). Posting by 36 months is a secondary outcome.
8. Validation gate
Registry fields are the record, not a judgment, so no human labeling gate. The gate is structural: for 100 trials the author compares the API values with the public record page. Any systematic mismatch halts the study until it is explained.
9. Analysis
In R.
glmer(posted_12m ~ ipd_share + sponsor_class + phase + applicable_law
+ log(enrollment) + (1 | lead_sponsor), family = binomial, data = d)Preregistered alongside: a Cox model on time to posting, censored at 36 months, and the same model within industry sponsors only, where the pledge and the posting obligation interact most.
10. Threats
| threat | handling | removed or disclosed | standard |
|---|---|---|---|
| The IPD answer can be edited after results are known | Not observable without change history; stated as the main limitation | Disclosed | |
| Sponsors differ in both pledging and posting | Sponsor class adjusted, lead sponsor random intercept | Disclosed | |
| US law requires posting for some trials only | Applicability entered as a covariate and stratified in a sensitivity analysis | Disclosed | |
| Completion dates are revised | The snapshot’s actual date is used | Disclosed | S7 |
11. Probes
| probe | computes | goes to | what would change the design | status |
|---|---|---|---|---|
| PA | eligible trials by IPD answer, exposure only | 5 | under 3000 yes-answers in the window narrows it | not run |
| PB | completeness of resultsFirstPostDate, sponsor class and phase, marginally |
7 | missing law-applicability for most trials drops that covariate to a sensitivity analysis | not run |
| PC | the author’s 100 record-page checks | 8 | a systematic mismatch halts the study | needs the author |
12. What this cannot show
- Whether trials that pledged sharing actually shared individual data.
- Whether a trial changed its pledge after seeing its results.