ClinicalTrials.gov: Reconcile Outcomes and Adverse-Event Denominators
TL;DR
A results record needs a traceable relationship to the protocol, statistical analysis plan and participant data. Matching the number of outcome rows is only the beginning: the outcome definition, observation period, analysis population and unit must also agree, or carry an explained and approved difference. ClinicalTrials.gov registration and results definitions describe different parts of this record; 42 CFR Part 11 sets the requirements for trials within its scope. [1] [2] [3]
Our August 1, 2026 AACT snapshot contains 660,411 posted outcome rows across 79,442 studies. These are outcome records, with several possible records per study. Their distribution illustrates the scale of reconciliation work; it measures neither reporting quality nor legal compliance. [4]
For a sponsor preparing results, the useful deliverable is an outcome-by-outcome reconciliation record with versioned sources, population definitions and resolved differences. The workflow below is an editorial implementation proposal. The cited regulations, registry definitions and study-specific analysis plan determine what belongs in an actual submission.
Start with the version that defines the result
The registration module describes the planned study, including primary and secondary outcomes and their time frames. The results module describes what was measured and analyzed. Under 42 CFR §11.28(a)(2)(i)(W)–(X), primary and secondary outcome information forms part of registration information for applicable trials. Section 11.48 describes results information, including participant flow, baseline characteristics, outcomes and adverse events. [1] [2] [3]
A current registry page is a changing record. A comparison between its current protocol and results sections can establish their current agreement. Assessing a change from the original plan requires dated protocol, registry and analysis-plan versions. An apparently perfect current match can coexist with an earlier amendment; an apparent difference can reflect a documented amendment or a different analysis unit. Review the history and reason before assigning a discrepancy category.
Create a compact source index before drafting results: protocol and amendments, the relevant statistical analysis plan, registry history, database cut and approved analysis outputs. Record the version and date of each. For the broader protocol-to-study-build relationship, see our analysis of ICH M11 and the study build map.
What the snapshot contributes
The retained AACT extract groups the 660,411 result outcome rows as 183,628 primary, 445,982 secondary, 28,557 other prespecified and 2,244 post hoc outcomes. The four counts sum to the reported total. AACT reorganizes ClinicalTrials.gov information for analysis; these categories remain registry classifications rather than independent verification of prespecification. [4]
Posted outcome records by registry category
August 1, 2026 AACT snapshot: 660,411 outcome rows across 79,442 studies. Descriptive classifications; no reporting-quality inference.
Scroll sideways for the full figure.
View chart data
| Category | Outcome rows |
|---|---|
| Primary | 183,628 outcome rows |
| Secondary | 445,982 outcome rows |
| Other prespecified | 28,557 outcome rows |
| Post hoc | 2,244 outcome rows |
The chart gives a workload context. Assessing outcome changes, conditionally required fields or individual review comments requires version-level or case-level evidence beyond this aggregate.
Reconcile meaning, population and unit
Use one row per planned outcome and link it to the corresponding results entry or an explained disposition. Titles alone are insufficient identifiers: the same phrase can describe different visits, scoring rules or aggregations. Conversely, a clearer results title can still represent the same planned measurement.
| Reconciliation field | Comparison to make | Evidence to retain |
|---|---|---|
| Outcome and hierarchy | Planned primary/secondary status versus the results entry and any documented change | Dated protocol, registry history and change rationale |
| Measurement | Instrument, score direction, endpoint calculation and summary statistic | Analysis-plan definition and approved output |
| Observation period | Planned time frame versus the period represented in the result | Visit windows, assessment rules and any amendment |
| Population | Assigned participants versus participants or units actually analyzed | Inclusion/exclusion derivation and reason for differences |
| Unit | Participants versus another unit such as lesions or observations | Unit definition, counting rule and denominator |
| Analysis | Reported estimate, uncertainty and comparison groups versus approved analysis | Output version, method and review decision |
These are proposed review fields, not an additional registry schema. ClinicalTrials.gov results definitions and §11.48 distinguish participants from other units analyzed. The analysis-population description is required in specified circumstances, including when the analyzed number differs from the number assigned to the arm or comparison group. A blank description, considered alone across an entire registry, therefore fails to identify a compliance defect. [2] [3]
The ICH E9(R1) estimand framework also helps explain why analysis populations and handling of events after treatment assignment matter. Its purpose is to make the treatment effect being estimated explicit. A withdrawal, treatment switch or rescue medication can affect the chosen analysis strategy; its presence does not by itself dictate excluding the participant. Follow the trial’s defined estimand and analysis rules. [5]
Treat adverse-event denominators as separate, explainable quantities
ClinicalTrials.gov adverse-event reporting distinguishes participants affected by an event, participants at risk and numbers of events where applicable. The results definitions also describe the reporting period, collection approach and frequency threshold for other adverse events. These are different quantities from the efficacy population and from simple enrollment. [2] [3]
A useful review asks which participants contributed to each safety table, during which period, under which grouping and ascertainment rule. A denominator difference should lead to a documented explanation or correction. Automatically forcing every table to use the randomized count can introduce an error when the approved definitions differ.
Consider a hypothetical parallel-group trial: 100 participants are randomized to an arm, 96 receive treatment, and 90 have the observation needed for a particular outcome. Those numbers could represent three legitimate populations. Their validity depends on the study’s definitions and handling of missing observations; the numbers alone settle nothing. If five participants have a particular adverse event in a defined at-risk group of 96, the participant count is five even if some have repeated episodes. An event count may be larger. This example illustrates counting logic, rather than recommending a population or statistical method.
| Safety review question | Record to inspect | Decision to document |
|---|---|---|
| Who was at risk? | Population derivation and exposure/observation definitions | Included participants and justified exclusions |
| What period is covered? | Safety reporting period and follow-up rules | Start/end boundaries and handling of later events |
| Are counts of people and events distinct? | Participant-level and event-level summaries | Counting convention and reconciliation |
| Why do table denominators differ? | Group definitions, event collection and population notes | Explained difference or corrected value |
| Does the narrative match the table? | Approved safety output and registry draft | Consistent population, time frame and units |
Separate submission duties from public posting
For applicable clinical trials subject to §11.42, §11.44 generally requires results submission within one year of the primary completion date, subject to the rule’s provisions for delayed submission and extensions. Determine applicability and the relevant exceptions for the actual study. Submission, quality-control review and public posting are separate steps. [6]
A registry-wide interval from primary completion to public posting combines studies with different legal scope, histories and review pathways. A late-submission assessment instead requires the applicable deadline and actual submission date for each eligible study. Pending-queue snapshots have a further selection limit: records that have already left the queue are absent. Reset-event shares within that selected queue describe pending records, with no valid all-submission failure-rate denominator or evidence of a software-related cause.
For operational tracking, retain the initial submission date, subsequent releases or corrections, review correspondence and posting date separately. The dates answer different questions. Follow the current PRS submission instructions for entering and releasing the record. [7]
Deliver a reviewed reconciliation package
Before release, give the responsible reviewer a versioned registry draft, the reconciliation record, the supporting outputs and a short list of resolved changes. Make the evidence easy to navigate: each results field should lead to the definition and output that supports it. Where the record changed, retain who approved the change and why. This proposed package helps reviewers examine the result without reconstructing a series of unlabelled spreadsheets.
EClinCloud’s EDC solution is relevant to study data collection and management. A buyer should assess the configured workflow, exports and review responsibilities against the actual trial. Submission and analysis compliance require study-specific evidence beyond a product description. [8] The same source-to-output discipline is discussed in our report on decentralized elements.
Methods and limits
The descriptive outcome chart uses the inherited August 1, 2026 AACT snapshot and its retained outcome-category aggregate. It counts result outcome rows, not unique trials, and uses no missing-field score, queue-based failure estimate or posting-based lateness classification. The article makes no inference about sponsor software, submission defects or outcome switching from those aggregates. Regulatory statements were checked against current Part 11 and the cited primary definitions; proposed review tables express editorial implementation choices.
Sources
1. National Library of Medicine. ClinicalTrials.gov protocol registration data element definitions.
2. National Library of Medicine. ClinicalTrials.gov results data element definitions.
3. eCFR. 42 CFR Part 11: Clinical Trials Registration and Results Information Submission, particularly §§11.28 and 11.48.
4. Clinical Trials Transformation Initiative. AACT database. Author analysis of the August 1, 2026 snapshot, retained outcome-category extract: 660,411 outcome rows across 79,442 studies.
5. ICH. E9(R1): Addendum on estimands and sensitivity analysis in clinical trials, final November 20, 2019.
6. eCFR. 42 CFR §11.44: Results information submission deadlines.
7. National Library of Medicine. How to submit results to ClinicalTrials.gov.
8. EClinCloud. EDC solution. Used only for the stated product scope.
