An endpoint describes an outcome measured in a study. A primary endpoint is the main planned outcome used to answer the central research question. Secondary endpoints provide additional information. These labels matter when a headline emphasises one result from a much larger set.
Find the planned question
Look for the protocol or registration linked from the paper. Record the primary outcome and its time point. Then locate the matching result in the published report. A measurement at one time point should not quietly become a claim about a different period.
Preserve the measurement details
Write down the measure, unit, comparison and time frame together. “Improvement” without a stated measure is too vague for an evidence record. If a result uses a score, check what the report says that score represents rather than substituting a familiar health outcome.
Read additional outcomes in context
Secondary and exploratory results can be informative, but they should retain their labels in a summary. Note how many outcomes were examined and whether the authors discuss the analysis plan. Do not elevate a selected favourable result into the study’s original main question.
Use a compact comparison record
Create one line for the planned outcome and a second for the reported result. Include the source location for each. If they differ, retain the difference as an open question until you find an explanation in the paper or supporting documents.
This is a reading aid, not an assessment of any medicine’s benefits or a statistical analysis guide. For a different type of evidence, see our certificate of analysis introduction: batch measurements and clinical outcomes answer different questions.
Sources
NCI: primary endpoint. NCCIH: results.
Describe the endpoint before calculating a result
A usable endpoint description identifies the outcome, unit or scoring rule, time point and analysis population. Consider a fictional trial whose primary endpoint is a change in a named score at 12 weeks. A count of events at 24 weeks is a different outcome and time point, even if both appear in the same paper. Retain the planned distinction when summarising the result.
A primary outcome is selected to address the principal trial question; secondary outcomes add planned information. An exploratory analysis can generate a new question but should keep its exploratory label. The CONSORT 2025 reporting guideline supplies a checklist for reporting randomised trials and links to the full explanation. It is a reporting resource, not evidence that a particular intervention works.
A worked result: absolute and relative differences
Invented data make the denominators visible. Suppose 20 of 200 participants in one group and 30 of 200 in the comparator group experience a specified event by 12 weeks. The observed risks are 10% and 15%. The risk difference, first group minus comparator, is -5 percentage points. The risk ratio is 0.10 / 0.15, approximately 0.67. Expressed as a relative reduction, 1 - 0.67 is approximately 33%. These calculations describe the fictional data; they are not estimates for a medicine or catalogue product.
If the event were beneficial rather than harmful, the direction of interpretation would change. If the groups had different follow-up or missing outcome records, the simple calculation would not resolve those complications. Report the event definition, time point and denominators with the result, and obtain the analysis appropriate to the design. A risk ratio and a hazard ratio are not interchangeable names for the same calculation.
A change within a group is not the treatment contrast
Another invented example: the mean score changes by -8 points in one group and -5 points in the comparator. Subtracting comparator change from the first gives -3 points. Reporting only the -8-point within-group change omits what happened in the comparison group. The subtraction is an unadjusted illustrative contrast; a planned analysis may use a different model or baseline adjustment.
Whether -3 points matters depends on the scale, uncertainty, population and context. A confidence interval around the contrast belongs beside the estimate. The numerical exercise supplies neither a confidence interval nor evidence of a meaningful clinical effect. Do not invent either from the two means alone.
Multiple outcomes need a declared analysis strategy
If a report examines many outcomes and time points, a selected small p-value does not explain how false-positive conclusions were controlled. ICH E9 statistical principles discusses the primary variable and multiplicity. The protocol and statistical analysis plan should identify the strategy actually used; there is no universal adjustment that a reader can assume after seeing the results.
A declared order for testing outcomes, adjusted comparisons and exploratory analyses have different interpretations. Check whether a secondary claim depended on an earlier result meeting its criterion. Record any departure from the planned strategy and the authors’ explanation. Do not reclassify an interesting secondary finding as the prespecified primary success.
An endpoint-results checklist with an audit trail
Create a comparison record with these fields: planned endpoint; source document and version; outcome definition; time point; primary, secondary or exploratory label; participants included; each group’s result; between-group contrast; uncertainty; handling of missing data; and multiplicity statement. Keep “not reported” where a field cannot be found. A blank is not evidence that no missing data or analysis changes existed.
For the fictional event example, the completed arithmetic fields are 20/200 versus 30/200, 12 weeks, risk difference -5 percentage points and risk ratio approximately 0.67. The interval and missing-data method remain unknown because the exercise did not specify them. For a real paper, locate those details in the methods, results or analysis plan before relying on the summary. This is a results-reading exercise, not a complete trial operating procedure, submission standard or analysis plan.
Top comments (0)