Research & EvidenceClinical Trial MethodsAdvanced LevelProfessionals
Intention-to-Treat vs Per-Protocol Analysis: What Each Estimate Can and Cannot Tell You
A practical interpretation guide for randomized trials covering allocation, adherence, missing data and post-randomization bias.
13 October 20268–12 min readEvidence sources included
What You Should Be Able to Explain After Reading
- Distinguish ITT, modified ITT and per-protocol approaches.
- Explain why adherence-based exclusions can undermine randomization.
- Recognize missing-data assumptions.
- Interpret trial conclusions in light of analysis populations.
Start with the question
- Intention-to-treat (ITT) generally analyzes randomized participants in the groups to which they were assigned, regardless of adherence or treatment switching, preserving the comparison created by randomization as far as possible.
- A per-protocol analysis estimates outcomes among participants meeting prespecified adherence or eligibility criteria. Because adherence is not randomized, groups can differ in prognosis and behavior after randomization.
- Modified ITT is used inconsistently; authors should state who was included, which exclusions occurred and when decisions were made.
Post-randomization exclusions and missing data
- Excluding participants for non-completion, missed sessions or absent final measures can break baseline comparability. Reasons for dropout may relate to adverse effects, symptom severity, access or prognosis.
- An ITT label does not solve missing outcomes. Reports should show missingness by group, reasons, assumptions, methods and sensitivity analyses where appropriate.
- Per-protocol comparisons can be biased even when criteria were prespecified. More advanced causal methods require explicit assumptions.
Appraisal checklist
- Check the protocol or registry for the planned analysis population, primary outcome, exclusions and missing-data approach. Compare these with the publication and note deviations.
- Look for participant flow by randomized group, numbers analyzed for each outcome, adherence, cross-over, adverse events and reasons for missing observations.
- Interpret effect size and confidence interval alongside clinical importance, risk of bias and applicability to physiotherapy practice. Statistical significance alone does not establish clinical importance.
Reporting and clinical pearl
- CONSORT encourages authors to report how participants were included in each analysis and the numbers analyzed in each group. If ITT and per-protocol estimates differ, do not automatically select the more favorable result; inspect adherence, missingness and assumptions.
- Ask: what population and treatment effect does this analysis estimate, what assumptions are required, and how could post-randomization events have biased it?
Evidence & Further Reading
These sources are provided so learners can trace the clinical recommendations and read the original guidance.