What 92 trials reveal about registries versus journal articles

Most people encountering the results of a clinical trial read about it in a journal article, not a registry entry. An observational study of 92 completed trials of targeted therapies for endocrine and metabolic disorders, all registered on ClinicalTrials.gov after September 2009, compared what the registry recorded against what actually made it into the published journal article, and the gap between the two is larger than a casual reader of either source would likely assume.

The registry was essentially complete. The journal articles weren't

Every trial in the study reported serious adverse events and other adverse events in the registry, and every eligible trial reported all-cause mortality data there as well. That's a fully complete safety-reporting record, at least at the registry level.

The corresponding journal articles told a different story. Serious adverse events appeared in 86% of the published articles. Other adverse events appeared in 91%. All-cause mortality data, among the trials where it was applicable, appeared in only 64%. That means more than a third of the relevant journal publications in this sample omitted mortality data entirely, information that was sitting complete and available in the registry the entire time.

Where both sources reported the same event, they often disagreed

Completeness is one problem. Accuracy, where both sources did report something, turned out to be a separate and in some ways more troubling one. Among trials where reporting was complete in both the registry and the journal article, allowing a direct comparison, 38% showed numerical discrepancies in serious adverse event counts. For other adverse events, the discrepancy rate was 86%, meaning the great majority of trials with complete data in both places still reported different numbers depending on which source you consulted. Mortality data discrepancies appeared in 36% of comparable cases.

An 86% discrepancy rate for other adverse events is a genuinely striking figure. It suggests that even when a study bothered to report the same category of safety data in both its registry entry and its journal publication, the actual numbers frequently didn't match, not as an occasional slip but as something close to the norm within this sample.

Why the numbers didn't match: terminology and thresholds

The study identified two specific, mechanical reasons behind these discrepancies, both of which are fixable rather than mysterious. Eighteen per cent of journal articles used different terminology than the registry to describe the same adverse events, meaning a reader comparing the two sources might not even recognise they were looking at the same underlying event, described in different clinical language. More significantly, 60% of articles applied different frequency thresholds for reporting adverse events than the registry did, meaning an event that crossed the registry's reporting threshold might fall below the journal article's higher bar, and simply not appear, or vice versa.

That 60% figure matters enormously for interpreting the discrepancy rates above. A numerical mismatch isn't necessarily evidence of an error or a deliberate omission. It's frequently a consequence of two different reporting standards being applied to the same underlying data, registries and journals operating with different conventions about what counts as reportable, applied inconsistently across this body of research.

Why this matters beyond an academic curiosity about reporting formats

A clinician, a systematic reviewer, or another researcher building on this evidence base who only reads the journal article is working from a version of the safety data that's both less complete and, where comparable, frequently different from the registry's numbers. Given that journal articles are generally the more visible, more cited, and more widely read version of a trial's results, this isn't a minor technical footnote. It's a real gap in what the most commonly consulted source of trial safety information actually contains, relative to the fuller record sitting in the registry.

What this means for how safety data gets reported and used

A few practical implications follow directly:

• Registry data deserves to be treated as at least equally authoritative as the journal publication, not a secondary or supplementary source. This study's own finding, that the registry was complete in every case examined while journal articles frequently weren't, argues for checking the registry directly rather than relying solely on the published article for safety information.
• Terminology and reporting thresholds need to be harmonised between a trial's registry entry and its eventual publication, not treated as independent editorial choices. The 18% terminology mismatch and 60% threshold mismatch identified here are both specific, addressable process gaps, not an unavoidable feature of how trial reporting works.
• A trial team preparing a publication should reconcile it against the registry entry as a deliberate step, not assume consistency will happen automatically. Given how frequently discrepancies appeared even in this study's "complete reporting in both places" subset, consistency clearly isn't the default outcome without active reconciliation.
• Anyone conducting a systematic review or meta-analysis using journal-reported safety data alone should be aware they may be working from an incomplete or inconsistent version of the underlying evidence, and cross-checking against the registry, where feasible, is a genuine methodological safeguard rather than an excessive precaution.

The broader message is that a trial's registry entry and its eventual publication are sometimes treated, implicitly, as two views onto the same underlying truth. This study's findings suggest they're often more like two separately maintained records that happen to describe the same trial, with all the potential for drift and inconsistency that separate maintenance implies.