An oncology endpoint is the result of several linked decisions, not simply a date selected from a dataset. Assessment schedules, lesion measurements, response rules, review methods, and censoring choices can all influence interpretation. These dependencies deserve attention before the final survival curve is produced.
EORTC’s RECIST publications provide the primary response-criteria references. The FDA’s oncology endpoint guidance supplies regulatory context. Neither should be reduced to a generic algorithm applied without considering the study design.
1. Preserve assessment history
Maintain the relationship between lesions, assessments, reader decisions, and participant-level response. Identify which criteria and review process apply. A response derived from investigator assessment should not be presented as if it came from an independent central review.
For a hypothetical participant with a missed scan followed by progression, retain the actual assessment sequence. Do not collapse it into one endpoint record before reviewers can inspect the evidence and applicable rules.
2. Make event and censoring rules explicit
A progression-free survival analysis depends on more than progression dates. Specify how death, missed assessments, new anticancer treatment, and incomplete follow-up are handled according to the approved plan. The correct rule is study-specific and must not be inferred from a convenient programming template.
Review the event definition and censoring reason together. Check whether the selected date corresponds to the specified assessment and whether the supporting records are available. A consistent-looking event code can conceal different underlying circumstances.
3. Examine the measurement process
Differences in assessment timing and follow-up may affect the opportunity to observe progression. Summaries of visit completion and scan timing can support interpretation of the primary analysis. They are diagnostic information, not an automatic adjustment for every potential bias.
ICH E9(R1) is relevant to the treatment-effect question and intercurrent events. Keep those decisions connected to the endpoint derivation, especially where treatment changes affect what outcome is being estimated.
4. Review difficult participant cases
Build an endpoint review set containing missed assessments, discordant reads, incomplete dates, and new therapy before documented progression. Ask statistical and clinical reviewers to agree on expected behavior before implementation is finalized.
Retain an explanation for each selected endpoint date and censoring reason. Compare the output with the source assessment history and the governing plan. When differences are resolved, assess whether the same issue affects other participants.
The goal is a defensible endpoint package: a reviewer can identify the event definition, understand the assessment process, inspect the participant-level derivation, and interpret the result with its limitations intact. A polished curve is only the final presentation of that evidence.
