Per-protocol analysis
From TrialTree Wiki
Per-Protocol (PP) Analysis
Per-Protocol (PP) Analysis evaluates the effect of an intervention among participants who fully adhered to the assigned treatment protocol. In contrast to Intention-to-Treat (ITT) analysis, which includes all randomized participants regardless of adherence, PP analysis estimates the efficacy of an intervention under ideal conditions.
1. When to Use Per-Protocol Analysis
PP analysis is particularly useful in specific scenarios:
- When assessing the true biological or mechanistic effect of an intervention.
- In studies where adherence is essential to achieving the treatment effect (e.g., dietary or behavioral trials).
- As a complement to ITT analysis for sensitivity or subgroup analysis.
Example: In a diabetes drug trial, the ITT population includes all randomized participants, while the PP analysis includes only those who took ≥80% of their prescribed doses.
2. Key Considerations
- Higher risk of bias: Excluding non-adherent participants may introduce selection bias.
- Loss of randomization benefits: Treatment groups may become unbalanced after exclusions.
- Limited generalizability: Findings apply only to those who followed the protocol and may not reflect real-world effectiveness.
3. Steps for PP Analysis
1. Define adherence criteria: This should be outlined in the study protocol. Examples include minimum medication adherence (e.g., ≥80%), attendance at a specified number of therapy sessions, or completion of follow-up assessments. 2. Exclude non-adherent participants: Participants who crossed over to another group, missed major follow-up points, or deviated from the protocol are excluded. 3. Analyze the per-protocol population: Conduct statistical analysis using only those who met the adherence threshold.
4. Statistical Methods for PP Analysis
- Continuous outcomes: Use t-tests, linear regression, or mixed-effects models.
- Binary outcomes: Apply logistic regression or estimate relative risk.
- Time-to-event outcomes: Use Kaplan–Meier survival analysis or Cox proportional hazards models.
5. PP vs. ITT Analysis
| Aspect | Per-Protocol (PP) | Intention-to-Treat (ITT) |
|---|---|---|
| Population | Only adherent participants | All randomized participants |
| Effect Estimate | Efficacy (ideal conditions) | Effectiveness (real-world) |
| Bias Risk | Higher (due to exclusions) | Lower (preserves randomization) |
| Clinical Relevance | Lower | Higher |
6. Reporting PP Analysis in RCTs
- Clearly define the adherence criteria used.
- Report the number and proportion of participants excluded from the ITT population.
- Compare PP and ITT results as part of a sensitivity analysis.
7. Conclusion
Per-Protocol analysis estimates the ideal efficacy of an intervention but comes with a higher risk of bias. It should be interpreted cautiously and reported alongside ITT analyses as part of a comprehensive trial analysis. Pre-specifying adherence criteria is critical to maintain transparency and minimize bias.
Bibliography
- Gupta SK. Intention-to-treat concept: a review. Perspectives in Clinical Research. 2011;2(3):109–112. Discusses both intention-to-treat and per-protocol approaches.
- Montori VM, Guyatt GH. Intention-to-treat principle. CMAJ. 2001;165(10):1339–1341. Includes discussion of when per-protocol analysis may be considered.
- White IR, Horton NJ, Carpenter J, Pocock SJ. Strategy for intention-to-treat analysis in randomised trials with missing outcome data. BMJ. 2011;342:d40. Reviews limitations and alternatives including per-protocol.
- Hernán MA, Hernández-Díaz S, Robins JM. Randomized trials analyzed as observational studies. Annals of Internal Medicine. 2013;159(8):560–562. Critiques naive per-protocol analysis and suggests causal modeling approaches.
- Piantadosi S. Clinical Trials: A Methodologic Perspective. 3rd ed. Wiley; 2017. Chapter on protocol adherence and analysis strategies, including per-protocol and as-treated analysis.
Adapted for educational use. Please cite relevant trial methodology sources when using this material in research or teaching.