What CONSORT Now Requires When You Report Outcomes
Consider a common situation for a team preparing a phase 3 manuscript. The protocol and locked SAP prespecify six outcomes. The registry entry lists five, in shorter free-text wording. The first manuscript draft reports four, including a primary outcome that has been reworded. Without a structured reconciliation, the team risks submitting a paper whose outcome differences are undocumented, and therefore hard to defend when an editor or peer reviewer compares it with the registry or the published protocol.
The defensible answer is to reconcile the four documents in a single, dated outcome disposition matrix before the results section is drafted. The matrix lists every outcome in the locked protocol and SAP, records its registry wording, and tracks how it appears in the manuscript. Each outcome then gets one disposition. It may be reported as prespecified. Its definition may have changed, disclosed with date and reason under CONSORT 2025 item 10. It may be new, labeled as not prespecified under CONSORT-Outcomes item 6a.7 and CONSORT 2025 item 10. Or it may be prespecified but not reported in the manuscript, with a sentence saying where the results can be found (CONSORT-Outcomes item 17a.1).
Cite the right version. The CONSORT 2025 statement is a 30-item checklist, published simultaneously in April 2025 in The BMJ, JAMA, The Lancet, Nature Medicine and PLOS Medicine with identical content. It states that it supersedes CONSORT 2010, “which should no longer be used.” Its summary of changes lists the integration of items from the Outcomes extension on how outcomes were measured and analyzed, as items 14 and 26. The protocol-side counterpart, SPIRIT 2025, was published alongside it.
The CONSORT-Outcomes 2022 extension remains the outcome-specific elaboration used alongside the main checklist. It was developed through a scoping review, a three-round international Delphi survey completed by 124 panelists from 22 countries (November 2018 to February 2019), and a consensus meeting attended by 25 panelists. Its 17 items attach to CONSORT 2010 checklist items, most of them to item 6a (outcome definitions) and item 12a (statistical methods). Because those numbers follow the 2010 checklist, cite them as extension items (for example, “CONSORT-Outcomes 6a.7”) and cite CONSORT 2025 items by their new numbers. The EQUATOR Network record lists CONSORT 2025 as the extension's generic parent guideline. It also points to five optional outcome-reporting items proposed in eTable 6 of the JAMA supplement. The developer team's project page confirms that the checklists are finalized and links to the CONSORT extensions library.
The extension defines a fully specified outcome by five elements. CONSORT 2025 item 14 names the last four explicitly. The extension adds the domain and asks for a rationale for the primary outcome's domain (item 6a.1).
Domain: the title or concept the outcome describes, such as knee pain. If the domain came from a core outcome set, record which one; our core outcome sets guide covers that upstream choice.
Measurement variable: the specific measurement made on participants, including the instrument, such as the WOMAC pain subscale score.
Analysis metric: the participant-level form used in the analysis, such as change from baseline, final value or time to event.
Method of aggregation: how participant-level values are summarized in each group, such as a mean, a median or the proportion meeting a threshold. If continuous data are analyzed as categories, extension item 6a.4 asks for the cutoff.
Time point: when the outcome is assessed for analysis, such as week 24 after randomization. If assessments happen at several time points, extension item 6a.5 asks which ones the analysis uses.
When any of these elements is missing or loosely stated, the outcome's identity becomes ambiguous. A later change can then pass unnoticed between protocol lock and final report. The table below uses an illustrative rheumatoid arthritis outcome.
| Element | Vague wording | Fully specified (illustrative) | Where CONSORT asks for it |
|---|---|---|---|
| Domain | Disease activity | Rheumatoid arthritis disease activity | Extension 6a.1 (rationale for the primary outcome's domain) |
| Measurement variable | Joint assessment | DAS28-ESR composite score | CONSORT 2025 item 14; extension 6a.2; instrument properties, extension 6a.8 |
| Analysis metric | Improvement | Change from baseline | Item 14; extension 6a.2; minimal important change for within-participant change, extension 6a.3 |
| Method of aggregation | Response | Mean change in each group (or the proportion below a stated cutoff) | Item 14; extension 6a.2; cutoffs for categorized continuous data, extension 6a.4 |
| Time point | End of treatment | Week 24 after randomization | Item 14; extension 6a.2; time points used when assessed repeatedly, extension 6a.5 |
The between-group comparison is not part of the outcome definition. An example is the difference in adjusted means from a repeated-measures model. That comparison belongs in the statistical methods (CONSORT 2025 item 21a) and the results (item 26). Keeping the two apart makes it easier to see whether the outcome changed or only the analysis did.
Where Outcome Identities Drift Between Protocol, SAP, Registry, and Publication
Differences between prespecified and published outcomes are common, and meta-research has measured them for more than a decade. Two studies anchor the numbers here. Read them with their limits.
Jones and colleagues (BMC Medicine, 2015) systematically reviewed 27 studies that compared registry entries with publications. The median share of trials with a discrepancy between the registered and published primary outcome was 31% (interquartile range 17% to 45%). Six studies compared publications with outcomes registered before enrollment, and in those the median was 41% (IQR 33% to 48%). That was higher than the 26% median in the 18 studies that used the registry entry as it stood at the time of the search. The authors read the difference as a sign that registry entries are often changed retrospectively to match publications. Ten studies examined secondary outcomes, with a median discrepancy rate of 54%. Eighteen studies counted registry primary outcomes that were undefined or unclear: the median was 15%, and that ambiguity often limited the comparison. The review presented medians rather than a pooled estimate because the studies were heterogeneous, and their risk of bias was moderate to high.
More recent evidence comes from a different design. Nguyen and colleagues (BMC Public Health, May 2026) examined interrupted time series (ITS) studies of health interventions, not randomized trials. They used 44 published protocols (2010 to 2022) and their results reports, covering 553 outcomes. Of the protocol-defined outcomes, 24% were never reported, and 60% of protocols (25 of 42) had at least one unreported outcome. The study counted three kinds of discrepancy: an outcome missing from the protocol, one missing from the report, or a change in primacy. By that definition, 52% of the 388 assessable outcomes had a discrepancy, affecting 74% of studies. Authors justified only 28% of them. The authors note that ITS studies with published protocols probably represent best practice, so RCT teams should treat these figures as indirect evidence rather than a benchmark for trials.
The empirical literature sorts discrepancies into a small number of types. Each type maps to a different disclosure route in the report. Jones 2015 tabulated these types for 1,288 trials in 16 studies. Not every study reported every type, so read the crude totals below as rough indications, not prevalence estimates.
| Discrepancy type | What differs between documents | Evidence on frequency | Disclosure route in the report |
|---|---|---|---|
| Omission of a prespecified outcome | An outcome in the protocol, SAP or registry is absent from the report | Jones 2015 crude total: registered primary outcome not reported in 9% of trials. Nguyen 2026 (ITS studies): 24% of protocol outcomes not reported; most common reasons given were event rates too low to analyze and the outcome being judged unsuitable | Report it, or state where the results can be found (extension 17a.1); if it was dropped after the trial began, disclose the change and reason (CONSORT 2025 item 10) |
| New or non-prespecified outcome | An outcome appears in the report but not in the protocol, SAP or registry | Jones 2015: median 13% of trials introduced a new, unregistered outcome. Nguyen 2026: 14% of assessable outcomes were missing from the protocol | Label it as not prespecified (extension 6a.7; CONSORT 2025 item 10) and explain why the analysis was done (extension 18.1) |
| Primacy change | A registered primary is reported as secondary, or a secondary is promoted to primary | Jones 2015 crude totals: registered primary reported as secondary in 7% of trials; published primary listed as secondary in the registry in 5%. Nguyen 2026: 4% of assessable outcomes, none with a justification | Disclose the change, its date and reason (item 10); keep the prespecified primary outcome as primary in the analysis and interpretation (our recommendation) |
| Timing or definition change | Same domain, but a different time point, measurement variable, analysis metric or aggregation | Jones 2015 crude total: timing of primary-outcome assessment differed in 8% of trials. Nguyen 2026 did not assess this type | Disclose the change with version, date and reason (item 10), and report the complete definition actually used (item 14) |
Build the Outcome Disposition Matrix Before You Write Results
To stop undocumented differences from reaching the manuscript, run the reconciliation as a cross-document matrix with one row per outcome. It links four records:
Protocol of record: the current approved protocol, with every amendment's version number and date. Protocol endpoint wording is covered in our ICH M11 protocol endpoints guide.
Locked SAP: the analysis specification locked before unblinding, including estimands, intercurrent-event handling and models. See our guide to statistical analysis plan endpoint definitions.
Registry record: the ClinicalTrials.gov, EU Clinical Trials Information System (CTIS) or other registry entry, including its change history. Registry wording and update obligations are covered in our ClinicalTrials.gov outcome measures guide.
Manuscript package: the main text, tables, figures and supplementary appendix as planned for submission.
Decide first which document holds the definition of record. Registry free-text fields often cannot carry all five elements. In Jones 2015, a median 15% of registry primary outcomes were too ambiguous to compare at all. We recommend treating the latest approved protocol together with the locked SAP as the definition of record. Then note whether the registry wording is consistent, merely abbreviated, or genuinely different. Abbreviation is a registry-hygiene task for the registry owner. A different time point or measure is a discrepancy that needs a disposition.
The template below is the matrix itself: one row per field, filled in for every outcome.
| Matrix field | What to record | Where it comes from | CONSORT hook |
|---|---|---|---|
| Outcome ID and tier | A stable ID (O1, O2 …) and its tier at protocol lock: primary, key secondary, secondary or other | Protocol and SAP | CONSORT 2025 item 14 |
| Definition of record | Domain, measurement variable, analysis metric, method of aggregation and time point | Latest approved protocol plus locked SAP | Item 14; extension 6a.1 and 6a.2 |
| Registry wording | Verbatim registry text and the date of the record version checked | Registry entry and its change history | Item 2 (trial registration) |
| Manuscript wording and location | Verbatim text plus the table, figure or supplement location | Draft manuscript | Items 14 and 26 |
| Element comparison | Which of the five elements differ, and between which documents | Reviewer's comparison | Feeds the disposition |
| Disposition | One of the four states below | Writing team and trial statistician | Items 10, 14 and 26; extension 6a.7, 17a.1, 18.1 |
| Change evidence | Amendment or SAP version, approval date, whether before unblinding, and the reason as recorded at the time | Amendment log and SAP change history | Item 10 |
| Pointer | Exact location of results not in the main report | Supplement, registry results record or companion paper | Extension 17a.1 |
| Owner and sign-off | Named row owner, reviewer and sign-off date | Matrix record | Internal control (not a CONSORT item) |
Every outcome row ends in exactly one disposition state:
Reported as prespecified: all five elements match the definition of record, even if the wording is shortened. Report the definition under CONSORT 2025 item 14 and the results under item 26.
Definition changed: one or more elements changed after the trial began. Record the amendment or SAP version, its date and the reason, and disclose the change under item 10.
Added, not prespecified: an outcome or analysis introduced after the trial began or after unblinding. Label it as not prespecified in text and tables (extension 6a.7; item 10) and explain why it was done (extension 18.1).
Prespecified, not in the report: an outcome or analysis in the protocol or SAP that the manuscript does not present. Add a pointer sentence saying where the complete results are (extension 17a.1).
Synthetic Worked Example: Phase 3 Outcome Reconciliation
The table walks a hypothetical trial that matches the opening scenario. It has six prespecified outcomes, five registry entries, and a draft that reports four prespecified outcomes plus one post hoc analysis. It shows a reworded primary outcome, an outcome missing from the registry, results moved to a supplement, and a post hoc subgroup analysis.
| Outcome | Protocol and locked SAP | Registry record | Draft manuscript | Disposition | Action in the report |
|---|---|---|---|---|---|
| O1 Primary | Mean change from baseline in DAS28-ESR at week 24; amendment 2 (before unblinding) moved ESR testing to a central laboratory | “Change in DAS28 at 6 months” | “Improvement in disease activity at week 24” | Definition changed | State the full five-element definition in Methods (item 14) instead of the reworded label; disclose amendment 2, its date and reason (item 10); log the registry's time-point and instrument wording for the registry owner |
| O2 Key secondary | Mean change from baseline in HAQ-DI at week 24 | “HAQ-DI at week 24” | HAQ-DI mean change at week 24, between-group difference with 95% CI | Reported as prespecified | Standard reporting under items 14 and 26 |
| O3 Secondary | Proportion of participants with ACR50 response at week 24 | “ACR50 response at week 24” | ACR50 proportions with risk difference and risk ratio, 95% CIs | Reported as prespecified | Item 26: absolute and relative effects for a binary outcome |
| O4 Secondary (patient-reported) | Mean change from baseline in SF-36 physical component summary at week 12 | “SF-36 PCS at weeks 12 and 24” | Not in main text; week 12 results in Supplementary Table S3 | Prespecified, not in the report | Pointer sentence to Table S3 (extension 17a.1); log the registry's extra week 24 time point, which is not in the protocol or SAP, and do not present week 24 as prespecified |
| O5 Secondary (safety) | Incidence of adjudicated serious infections through week 52 | Not listed | Not in main text (two events) | Prespecified, not in the report | Pointer to the supplementary safety table and registry results (extension 17a.1); log the registry omission |
| O6 Safety | Proportion discontinuing study treatment because of adverse events through week 52 | “Discontinuation due to adverse events through week 52” | Counts and percentages by group with risk difference and risk ratio | Reported as prespecified | Item 26 reporting by group |
| Post hoc analysis | Not in protocol or SAP; conceived after unblinding | Not registered | Subgroup analysis by a baseline serum biomarker at week 24 | Added, not prespecified | Label as post hoc and hypothesis-generating (items 10 and 28); explain why it was done (extension 18.1) |
The reworded primary is the row to learn from. “Improvement in disease activity” sounds harmless, but it drops the measurement variable, the metric and the aggregation method. Restoring the five-element definition in Methods is what lets a reader see the amendment. The registry's “6 months” and unqualified “DAS28” are a registry-side discrepancy. Log it and route it to the registry owner rather than letting the manuscript inherit the registry wording. Rows O4 and O5 show the opposite problem: outcomes that really were prespecified but are easy to lose silently.
Disclose Changes Without Rescuing Them Post Hoc
Transparency and inferential status are separate questions. When a team changes an outcome definition or runs an unplanned analysis, disclosure under CONSORT 2025 item 10 meets the reporting obligation. It does not turn an unplanned result into a confirmatory one.
Item 10 asks for “important changes to the trial after it commenced including any outcomes or analyses that were not prespecified, with reason.” Items 21d and 28 ask authors to distinguish prespecified from post hoc subgroup and sensitivity analyses. Extension item 18.1 asks authors to explain why any analysis that was not prespecified was performed. Disclosure tells readers where an outcome came from. How much weight it can carry is a methodological question, and CONSORT does not settle it.
CONSORT does not set a threshold for what counts as an “important” change. The decision rule below is our recommendation. When in doubt, disclose.
Operational clarification: a refinement that leaves all five elements and the analysis unchanged. Examples are stating a visit window around a defined time point or describing sample handling. Record it in the matrix; it usually needs no item 10 disclosure.
Substantive change: anything that changes what is estimated. Examples include a new time point (week 12 to week 24), a different instrument, a switch from mean change to a responder cutoff, different handling of intercurrent events such as treatment discontinuation, or a change in which outcome is primary. Disclose it under item 10.
For each disclosed change, record four things: the protocol or SAP version that made it, the approval date, whether it came before or after unblinding, and the reason as documented at the time. A reason reconstructed during peer review, with no document behind it, is much weaker than an amendment that already says why.
Model Language for Manuscript Disclosures
The two statements below continue the synthetic example. Bracketed text marks details a real team would fill in from its own records.
Changed outcome definition: “Protocol amendment 2 (approved [date], before database lock and unblinding) specified that ESR for the primary outcome, change from baseline in DAS28-ESR at week 24, be measured by a central laboratory rather than site laboratories, to reduce between-site assay variability. The outcome was otherwise unchanged (CONSORT 2025 item 10).”
Post hoc analysis: “The subgroup analysis by baseline serum [biomarker] concentration was not prespecified in the protocol or statistical analysis plan. It was conceived after unblinding because [reason]. It is reported as post hoc and hypothesis-generating (CONSORT 2025 items 10 and 28; CONSORT-Outcomes 18.1).”
Missing Analyses: Report Them or Point to Them
Protocols and SAPs often specify more secondary, exploratory, pharmacokinetic and health-economic analyses than a journal article can hold within its word, table and figure limits. The question is not whether every analysis appears in the main text. It is whether a reader can find each one.
Silent omission is what the meta-research keeps finding: in Nguyen 2026, a quarter of protocol-defined outcomes never appeared in the results report. CONSORT-Outcomes item 17a.1 addresses this directly. It asks authors to include the results for all prespecified outcome analyses, or to state where the results can be found if they are not in the report.
For outcomes and analyses the main text does not present, there are three citable places for the results:
Journal supplement: online supplementary tables published with the article, for full secondary-outcome results, subgroup tables and additional time points.
Registry results record: summary results posted to ClinicalTrials.gov (which has hosted a results database since 2008) or to the EU registry, for trials subject to results-posting requirements. Registry obligations themselves are covered in our ClinicalTrials.gov guide.
Companion publication: a separate paper on secondary, quality-of-life or exploratory outcomes, cited by DOI or, until publication, by preprint identifier.
Drafting Compliant Pointer Statements
A pointer has to be specific. “Additional data are available on request” does not tell a reader where a prespecified result is. Name the omitted outcome or analysis and give an exact, accessible location. Two synthetic examples:
Supplement pointer: “The prespecified secondary outcome, change from baseline in SF-36 physical component summary at week 12, was analyzed as specified in the statistical analysis plan; results by group and the between-group estimate are in Supplementary Table S3 (CONSORT-Outcomes 17a.1).”
Registry pointer: “Two adjudicated serious infections occurred through week 52, too few for the prespecified time-to-event analysis; event counts by group are in Supplementary Table S5 and in the trial's ClinicalTrials.gov results record (NCT[number]).”
The Pre-Submission Reconciliation Pass
Before submission, the writing team, the trial statistician and the clinical operations lead should run one last structured pass over the matrix and the manuscript. The steps below are a practical sequence, not a CONSORT requirement.
Confirm the checklist version. Check whether the target journal asks for CONSORT 2025 or still cites CONSORT 2010. For CONSORT 2025, check items 10, 14 and 26 in particular. If the journal still asks for the 2010 checklist, complete it as instructed and use the extension's items, which are numbered against CONSORT 2010, for outcome detail.
Run the five-element audit (item 14). Every reported primary and secondary outcome in Methods should state the measurement variable, analysis metric, method of aggregation and time point, with the domain clear from the outcome name.
Check results completeness (item 26). For each primary and secondary outcome, by group, confirm four things: the number analyzed; the number with available data at the outcome time point; each group's result with the effect estimate and its precision; and, for binary outcomes, both absolute and relative effects.
Close every matrix row. No row may be left without a disposition. Every “definition changed” row needs a version, date and reason. Every “added” row needs a not-prespecified label. Every “not in the report” row needs a pointer that resolves.
Hand registry discrepancies to the registry owner. Where the matrix shows registry wording that differs from the definition of record, route it to whoever owns the registry record. Our ClinicalTrials.gov outcome measures guide covers update obligations; this pass only documents the difference.
Check SAP alignment and access. Reported metrics, intercurrent-event handling and missing-data methods should match the locked SAP or be disclosed as changes. CONSORT 2025 item 3 also asks where the protocol and SAP can be accessed. For the lock itself, see our guides to SAP endpoint definitions and ICH M11 protocol endpoints.
File the signed matrix. Store the final matrix, with owners and sign-off dates, in the trial master file location your TMF plan designates for statistical and publication records. Our TMF Reference Model guide covers zone-level filing.
The matrix does not make a trial's outcomes better. It makes every difference between what was planned and what is reported visible, dated and explained before an editor or reviewer has to find it.