A clinical trial is marked completed, but a reader cannot find a journal article. That is a legitimate reason to investigate. It is not yet a complete finding about whether results are public, whether a reporting deadline applies, whether a submission is under review, or whether someone concealed an unfavorable outcome.
The public-interest question is important because the evidence available to readers can differ from the evidence generated by research. But an audit of that gap needs a defined cohort, a distinction between trials and publications, and a record of where results were sought. Otherwise a seemingly precise percentage can combine unlike objects and turn missing information into unsupported accusations.
The World Health Organization's 2017 joint statement distinguished registry disclosure from journal publication. It described a twelve-month global norm for summary results after primary completion and an indicative twenty-four-month journal-publication timeframe. Those were the statement's policy expectations, not a declaration that one worldwide legal deadline governs every trial. WHO's 2017 joint statement on public disclosure
This article develops an original twenty-trial reconciliation. All trial identifiers, sponsors, reporting histories, participant counts and outcomes in the worked examples are fictional. I actually ran a local script to check the invented rows and their totals. No live registry export was audited, no real sponsor was evaluated for compliance, and no treatment is being recommended or assessed for clinical effectiveness.
The exercise separates three questions that are often collapsed. What can a member of the public read? What reporting activity can be observed? And what obligation, if any, applies to the particular trial? Each deserves its own evidence. A good audit can expose an information gap forcefully while remaining careful about the explanation for that gap.
1. A trial can have registry results, a paper, both or neither
Start with twenty fictional trials, labeled Q01 through Q20. Eight have publicly accessible registry summaries. Nine have at least one journal results article. Five appear in both groups. The number with either form of disclosure is therefore twelve, not seventeen: eight plus nine, minus the five counted twice.
The four mutually exclusive categories make the picture easier to read. Five trials have both forms. Three have only a registry summary. Four have only a journal results article. Eight have neither form found in our defined exercise. These categories add to twenty and preserve the relationship between the two publication routes.
The choice of headline measure changes the reported percentage. Registry-summary availability is eight of twenty, or forty percent. Journal-article availability is nine of twenty, or forty-five percent. Availability through either route is twelve of twenty, or sixty percent. None is the same claim as the other two.
A report saying that fifty-five percent lack journal articles would be arithmetically correct in this fixture. Calling those eleven trials wholly unreported would be wrong, because three of them have registry summaries. Conversely, treating every journal article as fulfillment of a registry-reporting requirement would assume an equivalence the audit has not established.
Our labels concern access to specified outputs. They do not grade the scientific quality or completeness of those outputs. A registry summary could omit information needed for a particular analysis; a journal article could report only one part of a larger trial. The presence of a document opens another question about what it contains.
For this exercise, the relevant journal output must actually report results from the trial. A protocol paper, an editorial mentioning the trial, or a conference announcement is retained as a related document but does not automatically satisfy that definition. The classification rule should be chosen before the counts are calculated, not adjusted to make the final number look better.
A clear first finding is therefore possible without accusing anyone: in the constructed cohort, twelve trials have at least one qualifying public output and eight do not. The next task is to understand the timing, coverage and obligations behind those states. The arithmetic is the beginning of the investigation, not a substitute for it.
2. Define the cohort before counting missing results
Imagine that the twenty trials came from a fictional search returning thirty records. Four were observational studies outside the stated question. Three were duplicate records for already counted trials. Two described studies withdrawn before anyone enrolled. One had not reached the completion condition used for this audit. Removing those ten records leaves the twenty-trial cohort.
Those exclusions are rules for this particular educational exercise, not a general instruction to omit every stopped study from a reporting audit. A terminated trial that enrolled participants can still have informative results and reporting obligations. A withdrawn, never-enrolled study presents a different situation. The distinction should be visible in the inclusion record.
The research question might instead include all registered interventional studies, whether completed or not. That would require a different denominator and separate treatment of ongoing studies. Neither question is inherently illegitimate, but their percentages should not be compared as though they measure the same population.
The duplicate decision also needs an explanation. In the fictional search, three pairs share a documented trial identity across records. They are not removed merely because their titles look similar. Two separately conducted trials can have nearly identical names, and one trial can appear under several descriptive titles. A title-only deduplication rule could erase real studies or preserve duplicates.
A useful inclusion log records each source record, the trial to which it is assigned, and the reason for inclusion or exclusion. For a disputed row, the analyst should be able to reverse the decision and recalculate the total. A spreadsheet with only the final twenty identifiers hides the choices that produced the cohort.
Apply a sensitivity check to the example. If the not-yet-complete study were included and had no qualifying public output, the either-route rate would become twelve of twenty-one, about 57.1 percent instead of sixty percent. That change reflects the denominator, not a newly discovered publication failure among the original twenty.
The point is not to find an exclusion rule that produces the highest disclosure rate. It is to match the cohort to the stated question and disclose consequential judgment calls. An audit that preserves those decisions is more credible than one that reports a sharp percentage while leaving the study population undefined.
3. Completion is a set of dates, not one universal clock
ClinicalTrials.gov distinguishes primary completion from completion of all study data collection. Its description of primary completion concerns the final collection of data for the primary outcome, and distinguishes an estimated date from the date that actually occurred. These labels should not be replaced by a generic field called finished. ClinicalTrials.gov study definitions
For an original timeline, suppose a fictional trial reaches primary completion on January 1, 2025, completes later follow-up on July 1, 2025, and is examined by an auditor on February 1, 2026. The elapsed times are 396 days from primary completion and 215 days from final completion. Choosing the wrong starting point changes the apparent age of the reporting gap substantially.
Those dates do not, by themselves, establish a legal deadline. They show why the event associated with a clock must be identified before a rule is applied. A policy keyed to primary completion cannot be assessed by quietly substituting the later end of follow-up. A rule with a different trigger must be evaluated according to that trigger instead.
Estimated dates require another distinction. Suppose an earlier version of the fictional record predicted primary completion in January, but a later version records an actual date in April. An analysis using the earlier forecast as though it were an observed event would overstate elapsed time. The correction is to preserve the estimate and the actual date as different pieces of information.
Now give the twenty-trial cohort sixteen actual primary-completion dates, three estimates and one missing value. A date-dependent classification cannot treat all twenty as equally established. The three estimates need verification or a separately labeled sensitivity analysis. The missing value cannot become January first merely because the spreadsheet requires a date.
A further problem occurs when the study stops early. The word terminated can provoke an assumption that there is nothing to report. Yet the relevant question is what participants and data existed before termination and which reporting framework applies. Our audit should preserve the stopping reason and the dates rather than erase the trial from view solely because the original plan was not completed.
The useful output of this stage is a timing record: which event, which date, whether actual or estimated, which version supplied it, and what remains unresolved. That record supports a later deadline assessment. It also prevents the publication-gap chart from quietly mixing mature trials, recent completions and studies whose completion status has not been established.
4. A missing result is not automatically a violation
NIH's current guidance distinguishes the federal regulation for applicable clinical trials from its complementary funding policy. It also identifies a 2026 change for basic experimental studies with humans: applications for due dates on or after May 25, 2026 are treated differently under NIH's clinical-trial definition. Applicability depends on the framework and relevant dates, not simply on whether a record appears in a registry. NIH's policy-and-regulation comparison
NIH's dissemination policy describes a generally one-year results-submission timeline after primary completion for the covered context, with specified possibilities for delay. That summary is not a compliance determination for every registered study; coverage, effective dates and qualifying circumstances must be checked against the applicable requirements. NIH dissemination policy
In our fictional twenty-row exercise, twelve trials are stipulated to fall within one selected reporting regime. Three are outside that particular regime, and five remain unresolved. These are separate categories. Outside one regime does not mean free from every ethical, contractual, funding or legal obligation, while unresolved does not mean exempt.
Within the twelve stipulated in-scope trials, eight have submitted results and four have no submission observed in the fixture. The four are Q13 through Q16. Q13 has not reached the relevant standard date. Q14 has a documented delay. Q15 lacks enough timing information for a conclusion. Q16 is past the assumed standard date without an explanation found in the supplied packet.
Q16 is the priority follow-up case. It is not automatically an adjudicated violation. A real finding would require confirmation of the applicable obligation, the controlling dates, the submission record and any relevant exception or extension. The exercise supplies a flag for investigation, not a substitute for that verification.
The full cohort has nine trials without a submission recorded, but calling all nine delinquent would ignore the distinctions just made. Some are outside the selected regime, some have unresolved coverage, and the in-scope cases themselves have different timing states. A single red cell labeled late would conceal rather than summarize this evidence.
For a public-interest audit, report the observable gap and the applicability assessment separately. The public may reasonably care about inaccessible results even when a particular legal deadline does not apply. Conversely, identifying a legal requirement does not establish that every public information need has been satisfied. Keeping those questions separate makes the criticism more precise, not less serious.
5. Submitted, publicly accessible and quality-checked are different states
ClinicalTrials.gov's definitions distinguish results submission, public posting with quality-control comments, and posting after the quality-control process has concluded. They describe NLM's review as a limited check for apparent errors, deficiencies or inconsistencies. A researcher should preserve those distinctions rather than convert every intermediate state into either no results or fully verified science. ClinicalTrials.gov definitions of results and QC states
Our fictional cohort illustrates the difference. Eleven trials have a recorded submission. Eight have a publicly accessible summary. Of those eight, six have completed the stipulated QC process and two are publicly visible with comments still attached. Three submitted summaries are not yet publicly accessible in the exercise.
The resulting rates are fifty-five percent submitted, forty percent publicly accessible, and thirty percent with the stipulated QC-complete status. These are not competing estimates of one quantity. They measure three different stages. The audit should identify which one its headline uses and retain the other stages where they affect interpretation.
A reader concerned with access cannot treat an unseen submission as equivalent to a readable summary. A reader investigating whether a sponsor acted cannot ignore the submission simply because posting has not occurred. A reader assessing scientific reliability should not mistake completion of an administrative quality check for an independent validation of the underlying research.
NLM's archived January 2020 bulletin described the introduction of public posting with QC comments for specified applicable trials. That historical notice is useful evidence that public visibility and completed review were deliberately distinguished; it is not being used here as a guarantee about every current posting timeline. NLM's 2020 account of QC and posting procedures
Give Q07 a hypothetical comment identifying an inconsistency between two participant totals. An auditor can report that the comment exists and identify the unresolved inconsistency. The comment does not, without further evidence, establish fraud, invalidate every outcome, or reveal which total is correct. The status is an evidence issue to resolve.
A later version might address the inconsistency. The audit should then record what changed and when rather than leave the old flag permanently attached as though nothing happened. At the same time, it should not rewrite the earlier observation: the summary had that status at the first audit date. A transparent time series preserves both facts.
6. Fourteen papers can describe only nine trials
PubMed's help documentation explains that its Secondary Source ID field can contain trial-registry identifiers, including ClinicalTrials.gov identifiers. These links provide a useful route between a bibliographic record and a registered study, but the article's role still needs to be inspected. PubMed help on secondary-source identifiers
In the original reconciliation, Q01 has three results articles, Q02 and Q03 have two each, Q04 and Q05 have one each, Q09 through Q11 have one each, and Q12 has two. That totals fourteen articles covering nine trials. Counting fourteen as the number of publicly reported trials would inflate the trial-level rate.
The extra papers are not necessarily wasteful duplicates. A trial may have a main report, a later follow-up and a distinct secondary analysis. Each can contain useful information. The counting problem arises only when publications are treated as independent trials or when one secondary report is assumed to disclose every outcome from the parent study.
For the audit, maintain a trial-to-report relationship rather than forcing one publication into one spreadsheet cell. Each linked paper can carry a role: protocol, primary results, follow-up, secondary analysis or another clearly described category. The categories are local analytical labels and should be supported by the paper, not assigned solely from a matching acronym.
Consider an ambiguous paper with the same trial nickname but no registry identifier and a different enrollment count. It could describe a subgroup, a companion study, or another trial with a similar name. A responsible linkage process records the ambiguity instead of guessing the match to eliminate a missing-publication cell.
A paper that lists two registry identifiers introduces the reverse relationship: one report may combine information from more than one trial. The analyst should examine which results belong to each study rather than automatically credit both with complete reporting. A link establishes a connection worth reading, not the completeness of disclosure.
For our fictional cohort, the fourteen-to-nine crosswalk is explicit and fully known. In a real audit, some links would remain uncertain after reasonable searches. The final output should distinguish confirmed links, candidate links and no qualifying report found. That is more informative than converting every failed exact-identifier search into a claim that no publication exists anywhere.
7. Journal publication and registry disclosure should not be played against each other
ICMJE's recommendations say that posting brief structured trial results in an acceptable registry is not treated as prior publication under the conditions it describes. The recommendations also ask authors to explain discrepancies between registry and journal results. This is an editorial policy statement, not a substitute for a trial's funding or regulatory requirements. ICMJE recommendations on clinical trials
That distinction matters when a fictional investigator explains a missing registry summary by saying a journal article is planned. A plan is not a posted result. The audit can record the plan, but it should not count it in the same category as a summary the public can actually read. Nor should the investigator's statement be converted into a conclusion about the journal's policy without checking that policy.
Give Q09 a journal article and a submitted but not yet public registry summary. A reader looking only at the registry's public-result state would miss the available paper. A reader looking only at the paper would miss the unresolved registry stage. The two observations should be reconciled rather than used to cancel each other out.
Now suppose the journal article reports a six-month follow-up, while the registry submission includes the prespecified earlier outcome as well. The outputs may be complementary. The audit should identify which questions each answers instead of declaring one the authoritative complete version merely because it appeared first or in a prestigious venue.
For Q12, the fictional packet contains two journal articles but no registry submission. That can support a finding of journal disclosure and a separate registry-submission gap. Whether the latter violates an obligation remains unresolved in our scope assessment. The existence of papers does not settle that legal question, but the absence of a registry submission should not erase the papers from an access assessment.
A useful reconciliation note records the publication date, the type of report, the trial identifier, the outcomes covered and any apparent differences requiring explanation. The point is not to perform a full clinical evaluation inside a disclosure audit. It is to avoid calling a trial wholly unpublished when results exist, or fully reported when the located output addresses only a different question.
The same approach handles a protocol-only paper. It may provide valuable evidence of the planned design and analysis, but its existence does not establish that completed outcomes were publicly reported. That document belongs in the record map with its actual role. Treating every linked paper as results would create a false sense of completeness before anyone examined the evidence.
8. Check whether the located output answers the registered question
The CONSORT 2025 statement includes reporting items for trial registration, access to the protocol and statistical analysis plan, and specified outcome details such as the measurement, analysis and time point. Those features support a more exact comparison than matching a trial title alone. They are reporting guidance, not proof that a particular trial was well conducted. CONSORT 2025 statement
Create a fictional Trial R with three planned outputs: a primary score at twelve weeks, a secondary function measure at six weeks, and a count of adverse events over twenty-four weeks. A paper reports the score at six weeks and the function measure at six weeks. A registry summary reports the twelve-week score. Neither output alone, as described, supplies the twenty-four-week adverse-event count.
A simple indicator saying results found would hide that pattern. A more informative evidence map has one row per intended outcome and identifies which located output addresses it. In this example, the primary score is available in the registry, the secondary measure in the paper, and the longer adverse-event account remains unlocated within the supplied material.
The six-week score is not automatically a substitute for the twelve-week score merely because the instrument has the same name. Timing is part of the measured question. The difference might be explained by a planned intermediate analysis or a documented amendment, but the audit should locate that explanation rather than assume either harmless equivalence or misconduct.
Analysis populations can produce another apparent discrepancy. Suppose 120 participants were randomized, ninety supplied a particular measurement, and twenty-seven met an invented response criterion among those measured. Twenty-seven divided by ninety is thirty percent. Twenty-seven divided by 120 is 22.5 percent. The two percentages use different denominators and cannot be compared responsibly without stating that fact.
This arithmetic does not estimate a treatment benefit. We have supplied no control-group comparison, uncertainty analysis or clinical context. It illustrates why a disclosure audit should preserve the population attached to a result. Calling one of the percentages false solely because it differs from the other would be an unwarranted conclusion.
The appropriate finding for Trial R is specific: the located reports cover some planned outcome-timepoint combinations, while another has not been found in the defined search. The next evidence request concerns the missing combination and any documented changes. It does not require the auditor to speculate about whether the absent result would favor or disfavor the intervention.
That distinction also protects the audit from becoming a superficial checklist. The goal is not to reward a filled cell regardless of content. It is to identify which public records answer which research questions and to describe the important gaps without pretending that publication status alone establishes scientific validity.
9. An unseen result cannot be labeled unfavorable by inference alone
Missing results create uncertainty about the available evidence. They do not reveal their own contents. An audit can investigate whether reporting patterns are associated with outcomes when it has appropriate data, but the fact that one particular result was not found does not establish that the result was negative or deliberately suppressed.
Use a separate fictional ten-trial example. Five trials have visible results: three fall into an invented favorable category and two do not. The other five results are unavailable to the reader. Among the visible trials, the favorable share is three of five, or sixty percent. That number describes the observed subset only.
Now construct two possible unseen worlds consistent with the same visible record. In World A, all five unavailable results would fall into the favorable category, producing eight favorable trials out of ten overall. In World B, none would, producing three out of ten. The reader's observed five-trial subset is identical in both worlds.
The resulting thirty-to-eighty-percent range is a simple count-based illustration, not a clinical meta-analysis or a statistical confidence interval. The categories and equal trial weights were invented for the exercise. The point is that the visible record alone does not choose between these different underlying patterns.
A writer should therefore resist two opposite shortcuts. It would be wrong to assume that every unseen result confirms the visible pattern. It would also be wrong to assume that every unseen result contradicts it. Both assertions fill missing evidence with a preferred story rather than an observation.
The public-interest finding can remain strong without either shortcut. Five of the ten results are not available in the defined record set, and that absence prevents a complete account under our stated classification. The audit can identify who is responsible for the relevant record, which output is missing and what explanation has been documented, without inventing the missing outcome.
A later discovery would update the evidence rather than vindicate an unsupported guess. If one unavailable result is found and classified, the remaining uncertainty narrows by that observation. The audit should record when it became available and how it was classified, not retroactively describe the earlier incomplete picture as though the result had always been known.
This is also why disclosure findings should not become patient-specific advice. Even a complete set of publication-status labels does not establish whether a treatment is suitable for an individual. The contribution of this audit is a clearer map of accessible evidence and its gaps, not a shortcut around clinical interpretation.
10. A stable percentage can hide improvement and a changing population
Return to the original twenty-trial cohort, where twelve have a qualifying public output through either route. Imagine a later fictional snapshot in which two previously unreported trials acquire such an output. Within the fixed cohort, the count rises to fourteen of twenty, or seventy percent. That is a genuine change in the observed disclosure state of the original group.
Now add five newly eligible trials to the audit, only one of which has a qualifying public output. The expanded cohort contains fifteen reported trials out of twenty-five, or sixty percent. The headline percentage is back where it began, even though two older trials gained reports and the population changed.
The correct account separates those movements. The original cohort improved from twelve to fourteen. The five additions contributed one reported and four unreported trials. A single comparison of sixty percent with sixty percent would conceal both the improvement and the new gap. It would be wrong to conclude that nothing changed.
Trial size introduces a different denominator. In the original fixture, each of the eight trials with a public registry summary has fifty participants. Each of the other twelve has two hundred. The cohort therefore represents 2,800 participants, with four hundred enrolled in trials that have a public registry summary.
At the trial level, registry-summary availability is forty percent. The share of participants enrolled in those trials is about 14.3 percent. Neither should be relabeled as the other. The larger trials make the participant-weighted view different, even though no trial's publication label has changed.
The either-route group contains the eight smaller trials and four of the larger trials. Its enrollment total is 1,200, about 42.9 percent of the cohort's 2,800 participants, while its trial-count share is sixty percent. These are descriptive coverage calculations on invented enrollment counts, not estimates of how much clinically useful information has been disclosed.
A published report may omit some outcomes or participants, and an enrollment total does not establish a complete individual-level dataset. The participant-weighted figure therefore means enrolled in a trial with a qualifying output, not every participant's data are publicly available. That wording matters for both interpretation and privacy.
A useful dashboard could show fixed-cohort change, expanded-cohort status and enrollment-weighted coverage separately. The analyst should explain why each measure is included rather than merge them into a single score. The resulting picture is more demanding to read, but it answers questions the simple missing-papers percentage cannot.
11. Test the counting pipeline before trusting the dashboard
The twenty-row reconciliation was actually checked with a local Python script. Its inputs were original fictional JSON, not a ClinicalTrials.gov export or an implementation of the registry's schema. The script grouped the supplied trial labels and calculated the sets, publication counts, reporting stages and enrollment totals described above.
The base output confirmed eight public registry summaries, nine trials with journal results, five in both groups, twelve in either group and eight in neither. It separately confirmed eleven submissions, six QC-complete records, two public summaries with comments and three submitted summaries not publicly accessible in the example. Those outputs are retained with the input file.
I then ran a duplicate-row control. It added copies of Q01, Q02 and Q03, producing twenty-three rows but still twenty unique trial identifiers. Counting rows gave eleven registry-positive entries and fifteen entries with either-route disclosure. Counting unique identifiers restored the intended totals of eight and twelve.
The duplicate control shows a concrete failure that a plausible-looking dashboard could hide. Fifteen divided by twenty-three is about 65.2 percent, while twelve divided by twenty is sixty percent. No new trial or result appeared. The change came entirely from repeating three already reported trials in the input.
A second executed control removed the public-registry-summary field from Q20. The modified file then had eight known public summaries, eleven known nonpublic states and one unknown state. A parser that treats a missing field as false would silently turn the unknown into another established absence. That is a classification decision, not a neutral convenience.
The missing-field control permits an explicit sensitivity calculation. Counting all twenty trials, the known public share is eight of twenty, or forty percent. If the unknown trial later proves to have a public summary, the share becomes nine of twenty, or forty-five percent. Restricting the denominator to nineteen known states instead produces about 42.1 percent. That last figure answers a different question and should not be presented as though the twentieth trial had been resolved. The interval here comes from one unclassified row, not from statistical sampling uncertainty.
The control does not establish that Q20 has public results. It establishes that the supplied modified row does not say. If a chart needs a denominator, it can display the unknown separately and explain how it affects the calculation. The analyst should not invent the absent value merely to produce a complete-looking table.
These tests validate only the local counting behavior against the constructed inputs. They do not validate real registry data, legal classifications, article-to-trial links or clinical findings. A perfect script would still produce a misleading public claim if the underlying labels were unsupported. Data checking and substantive review solve different problems.
For a reproducible audit, preserve the cohort definition, the retrieval date, the crosswalk, the classification rules and the output used in the published calculation. When a row changes, the analyst should be able to identify whether the cause was a newly available result, a corrected link, a changed scope decision or a data-processing error. The dashboard should be the visible end of that account, not its only surviving artifact.
12. Public disclosure and communication to participants are separate responsibilities
An NIH notice released August 27, 2026 sought input on a draft policy for sharing summary-level study results with clinical research participants. The notice listed an October 26 response date and explicitly presented a draft, not an enacted final requirement. It highlights a question different from whether a technical registry table or journal article exists. NIH request for information, NOT-OD-26-113
A fictional trial can have a public results article while its participants do not know where to find it or what it means. Conversely, a team might send participants a plain-language summary without satisfying a separately applicable registry requirement. An audit should not let either activity stand in for the other without examining the relevant obligation and the actual communication.
For the twenty-trial exercise, imagine an additional participant-communication field. It would need its own evidence and definition. A planned email, a completed distribution, a public webpage and a message returned as undeliverable are different states. Our current fixture contains none of those observations, so it cannot produce a truthful participant-communication rate.
Nor should a request for transparency become a demand to expose personal participant information. Summary reporting, access to de-identified research data and public release of identifiable records are different matters. This article's arithmetic uses only invented aggregate counts and does not propose collecting or publishing private medical details.
The final audit finding should therefore be written at the level the evidence supports. In the constructed cohort, twelve of twenty trials have a qualifying registry or journal output, eight do not, and the reporting-stage and applicability records explain why those categories cannot be translated directly into a count of legal violations. Fourteen journal papers cover nine trials rather than fourteen independent studies.
The audit also identifies the next questions precisely. Some results are submitted but not publicly accessible. Some visible outputs have unresolved comments. Some located papers cover only part of the planned outcome set. Some timing or applicability records remain incomplete. These are different work items, not interchangeable explanations for one empty results cell.
A serious public-interest investigation can use that specificity to ask sharper questions and publish more defensible findings. It need not soften a genuine lack of disclosure, but it should not inflate one by confusing papers with trials, forecasts with actual completion dates, missing links with absent publications, or unresolved coverage with proven noncompliance.
The trial finished is the start of the question. A reliable answer follows the evidence into the registry, the publication record, the reporting history and the applicable framework, while keeping their meanings separate. That is how an audit can make hidden gaps visible without filling those gaps with invented results or allegations.