03 · What You Need to Know
How to Identify the Primary Outcome Without Confusing It With the Main Finding
An outcome is a measured event, condition, characteristic, or response that a study investigates. In intervention research, outcomes are often used to evaluate whether an intervention produced the intended effect. In observational research, they may represent the events or characteristics being explained, predicted, or described.
The term primary outcome is particularly important in confirmatory trials, where the principal outcome often informs sample size planning and the study's main analysis. Other research designs may use different terminology or may not designate one outcome as primary.
What Is a Primary Outcome?
A primary outcome is the outcome designated as central to addressing a study's principal objective. In a well-specified confirmatory trial, it is ordinarily established before researchers examine the outcome results.
For example, a study evaluating AI-assisted feedback might designate students' writing proficiency at the end of an intervention as its primary outcome.
The researchers might also measure revision quality, satisfaction, engagement, and perceived usefulness. These additional measures could be secondary outcomes.
AI should not treat the most interesting or statistically significant of these measures as the primary outcome unless the study's documentation supports that designation.
Primary, Secondary, and Exploratory Outcomes
| Outcome Type |
Meaning |
Common AI Error |
| Primary outcome |
The outcome designated as central to the study's principal objective. |
Selecting the strongest result rather than the prespecified primary measure. |
| Secondary outcome |
An additional outcome addressing other aspects of the research question. |
Promoting a favorable secondary result to primary status. |
| Exploratory outcome |
An outcome investigated to generate hypotheses or examine additional possibilities, with its status depending on the study's analytical plan. |
Presenting exploratory findings as prespecified confirmatory evidence. |
These categories are not interchangeable, although a study may have more than one primary outcome. When several primary outcomes are prespecified, their joint interpretation may require attention to multiplicity and the study's statistical analysis plan.
Nor should AI assume that all qualitative or descriptive studies follow this hierarchy. A qualitative investigation may explore several interconnected phenomena without defining a conventional primary endpoint.
The Primary Outcome Is Not Necessarily the Most Important Result
A study may fail to find a statistically significant difference in its primary outcome while reporting favorable results for secondary measures.
Suppose an educational intervention has no statistically significant effect on overall writing proficiency, but students receiving the intervention report greater satisfaction.
An AI summary might identify satisfaction as the primary outcome because it dominates the discussion. This would misrepresent the original study design if writing proficiency was prespecified as primary.
Identifying the main reported result and identifying the primary outcome are related but distinct tasks. One concerns the evidence reported; the other concerns the outcome's designated role in the investigation.
Where Should AI Look for the Primary Outcome?
The methods section is usually the starting point, particularly subsections describing outcomes, measures, endpoints, and statistical analysis.
For registered trials, the study protocol and trial registry may provide additional evidence about which outcomes were designated before data analysis.
ClinicalTrials.gov, for example, includes structured fields for primary outcome measures. These records may specify the measure, description, and time frame.
However, registry records can be updated. Researchers should inspect the relevant version history or protocol amendments when determining whether an outcome was prespecified.
AI should not assume that the current registry entry necessarily reflects the original prespecification.
What Information Defines an Outcome?
An outcome is not adequately described merely by naming a broad construct such as achievement, engagement, or health.
For example, "academic performance" might refer to examination scores, course grades, standardized assessments, or a composite measure.
A useful extraction should preserve the specific measure and relevant measurement conditions.
In clinical trial methodology, an outcome may require specification of the variable, analysis metric, method of aggregation, and time point. The precise terminology depends on the study design and applicable guidance.
For educational research, an outcome such as writing proficiency might be measured through a rubric-based assessment at eight weeks, while another outcome could be the number of revisions completed during the intervention.
AI should distinguish these measures rather than collapsing them into the broad label "writing improvement."
Can AI Confuse Outcomes With Predictors or Mediators?
Yes. Statistical models may contain several variables with different analytical roles.
Consider a study examining whether teachers' AI literacy predicts their intention to adopt educational technology, with perceived usefulness included as a mediator.
Behavioral intention may be the outcome variable, AI literacy the predictor, and perceived usefulness the proposed mediator.
AI could mistakenly identify perceived usefulness as the primary outcome because it receives substantial theoretical discussion.
Researchers should inspect the research questions, conceptual framework, measurement definitions, and analytical models to establish the role of each variable.
Even then, the outcome variable in a regression model is not automatically a formally designated primary outcome of the entire study.
Can a Study Have More Than One Primary Outcome?
Yes. Some studies prespecify co-primary outcomes, meaning that more than one measure is designated as primary.
For example, an intervention may be evaluated using both reading comprehension and writing proficiency.
The implications depend on the protocol and analysis plan. Some studies require success on all co-primary outcomes, while others use different decision rules or statistical adjustments.
AI should report all designated primary outcomes and avoid choosing one simply because its results are more favorable.
What Is Outcome Switching?
Outcome switching occurs when the outcomes designated or prioritized in a study are changed, omitted, or newly introduced in subsequent reporting without appropriate transparency.
For example, a trial protocol may specify writing proficiency as the primary outcome, while the published article presents student satisfaction as primary without explaining the change.
Not every outcome modification is inappropriate. Changes may be justified by methodological or practical developments, particularly when documented prospectively and transparently.
The concern arises when changes are undisclosed or influenced by knowledge of the results, potentially creating selective outcome reporting.
AI may help compare protocols and publications, but it should not declare misconduct merely because two documents differ. Researchers must establish the chronology, relevant amendments, and possible explanations.
How Do Reporting Guidelines Help?
CONSORT 2025 requires randomized trial reports to describe prespecified primary and secondary outcomes, including how and when they were assessed, and to report changes after trial commencement with reasons.
SPIRIT guidance addresses the specification of outcomes in trial protocols. ClinicalTrials.gov provides structured registration information that can support comparisons between planned and reported outcomes.
These sources are especially useful for clinical and other registered intervention trials. Their requirements should not automatically be imposed on every observational, qualitative, or conceptual paper.
What If the Authors Never Identify a Primary Outcome?
AI should first determine whether the article type and study design would ordinarily require a primary outcome designation.
Some exploratory, descriptive, or qualitative studies may have several objectives without formally ranking their outcomes.
For a confirmatory trial, failure to identify the primary outcome may represent an important reporting concern. However, it is not appropriate to invent one based on the largest effect or first result presented.
A defensible response would be: "The paper reports several outcomes but does not explicitly designate a primary outcome in the available text."
Watch Out
Do not infer the primary outcome from statistical significance, effect size, abstract prominence, or discussion length. These features describe how results are reported, not necessarily which outcome was designated before the findings were known.
06 · What This Means for You
How to Use AI to Extract and Verify Primary Outcomes
Primary outcome extraction should begin with the study's planned objectives rather than its observed findings.
This is particularly important when preparing systematic reviews, evidence tables, or critical appraisals of intervention studies.
A simple decision framework
If the paper explicitly identifies a primary outcome
Extract its exact definition, measurement instrument, time point, and source location.
If a protocol or registration is available
Compare the outcome designation with the relevant prespecified version and documented amendments.
If several primary outcomes are specified
Record all of them and preserve any stated joint decision rule.
If the paper reports outcomes without prioritizing them
Report that no primary outcome is explicitly identified rather than selecting one retrospectively.
If the published outcome differs from the protocol
Check amendment dates and explanations before interpreting the discrepancy.
A Reusable Prompt for Primary Outcome Extraction
Suggested Prompt
"Identify the primary outcome or outcomes of this research study using only the available paper and any supplied protocol or registration. Distinguish outcomes explicitly designated as primary from secondary or exploratory outcomes. For each primary outcome, report its exact definition, measurement method, assessment time point, and supporting source passage. Do not infer primary status from statistical significance, effect size, or prominence in the abstract. If no primary outcome is explicitly designated, state that clearly. If protocol and publication descriptions differ, report the discrepancy without assuming the reason."
Keep Outcome Identity Separate From Outcome Results
For a literature matrix, consider recording the primary outcome, its measurement, its prespecification status, and its observed result in separate fields.
This helps prevent the outcome's definition from changing retrospectively based on the findings.
When interpreting the result, consult the relevant analysis and verify whether the statistical evidence supports the authors' conclusion. Identifying an outcome is not the same as evaluating the statistical methods used to analyze it.
Researchers should also examine whether the AI-generated account preserves the distinction between planned outcomes and the findings emphasized in the published paper.
