Clinical research & EBM · Reference
What is per-protocol analysis?
Per-protocol analysis evaluates only the participants who completed a trial in accordance with the protocol, excluding those who deviated. It estimates the effect under ideal adherence, but by breaking the randomised comparison it risks bias, so it is usually reported alongside intention-to-treat.
Analysing only the adherent
Where intention-to-treat analysis keeps every participant in the group to which they were randomised, per-protocol analysis restricts attention to those who actually received and completed their assigned intervention as the protocol specified. Participants who deviated substantially — by stopping early, switching arms or breaching eligibility — are set aside. The intention is to estimate the effect of the treatment when it is taken correctly, sometimes described as the efficacy or "as-treated under ideal conditions" estimate, in contrast to the pragmatic, real-world estimate that intention-to-treat provides.
Why it risks bias
The weakness of per-protocol analysis is that adherence is rarely random. People who complete a protocol often differ systematically from those who do not — they may be healthier, more motivated or experiencing fewer side effects. Excluding non-adherers therefore selects a non-comparable subset and can break the balance that randomisation created, reintroducing confounding. For this reason a per-protocol result tends to flatter the intervention, and on its own it is regarded as a weaker basis for judging effectiveness than the intention-to-treat result.
When and how it is used
Per-protocol analysis is not discarded — it is reported as a secondary or sensitivity analysis alongside intention-to-treat, and agreement between the two strengthens confidence in the findings. It has a special role in non-inferiority trials, where intention-to-treat can bias towards non-inferiority and the per-protocol analysis acts as a stricter check. Good practice is to pre-specify in the protocol exactly which deviations cause exclusion, so the analysis population is defined before the data are examined rather than chosen to suit the results.
Key facts
At a glance
- Definition: Analyses only protocol-compliant participants
- Contrast: Differs from intention-to-treat (analyse as randomised)
- Estimates: Effect under ideal adherence (efficacy)
- Main risk: Selection bias from non-random adherence
- Role: Secondary or sensitivity analysis
- Important in: Non-inferiority trials, as a stricter check
Common questions
FAQ
What is the difference between per-protocol and intention-to-treat analysis?+
Intention-to-treat analyses every participant in the group they were randomised to, preserving randomisation and estimating real-world effectiveness. Per-protocol analyses only those who followed the protocol, estimating the effect under ideal adherence but risking bias because adherence is rarely random.
Why can per-protocol analysis be biased?+
Because people who adhere to a protocol often differ systematically from those who do not — being healthier or more motivated — excluding non-adherers selects a non-comparable subset. This can break the balance randomisation created and reintroduce confounding, usually flattering the intervention.
When is per-protocol analysis useful?+
It is reported as a secondary or sensitivity analysis alongside intention-to-treat, where agreement between the two builds confidence. It is particularly important in non-inferiority trials, where intention-to-treat can bias towards a non-inferiority conclusion and per-protocol provides a stricter check.
Going deeper
Related on CASRAI
- Intention-to-treat →
- Non-inferiority trial →
- Randomized controlled trial →
- Confounding →
- Clinical research hub →
Sources
The step most authors miss
Doing CRediT right? Don’t stop at the statement.
A CRediT statement credits you inside one paper. The recognition CRediT was built for happens when those roles are tied to you, persistently. Sign in with your ORCID — free — and claim your CRediT contributions on casrai.org, the home of the standard. They become a verified, portable part of your identity, not a line that disappears into one PDF.
Free: claim your contributions, then export a journal-ready CRediT statement, schema.org structured data, JATS XML, CSV or BibTeX — and preview your public profile. A membership publishes that profile publicly and verifies the journals you serve.
