.jpg){kind=link}
- Published
- Updated
Comparing only participants who fully follow the protocol may appear to produce a cleaner analysis. If adherence relates to post-treatment experience, however, the selected groups can differ importantly from the original randomised comparison and answer a different question.
Define the question and assumptions
Prespecify the events and thresholds defining the per-protocol set. Apply exclusion rules independently of treatment outcome preferences. Reliability and missingness of adherence measurements also matter. Explain the analysis question, its relationship to the main analysis and its limitations rather than assuming adherence automatically restores an unbiased comparison.
Hypothetical study example
In a hypothetical trial, participants experiencing discomfort stop treatment more often. An analysis restricted to completers can remove these experiences from the comparison. The team does not interpret a more favourable result as the true treatment effect without examining selection reasons and the characteristics of those retained.
Design review checks
- Record the timing and scientific rationale of each analytical exclusion rule.
- Assess how adherence was measured and where that information is missing.
- Show participant flow and reasons for differences between analysis populations.
Record the decision and its limits
Accepted reporting does not label a per-protocol estimate as bias-free. Explain the roles of different populations, especially in designs such as noninferiority, with appropriate expertise. Do not alter exclusions after outcomes are known to create an attractive comparison. Report planned and additional analyses alongside their target questions and selection limitations.
Frequently asked questions
- Are adherent participants always a better comparison?
- No. Post-treatment selection can affect comparability between groups.
- Is agreement with the main analysis sufficient?
- Consistency helps, but shared assumptions and limitations still require assessment.