Per-protocol analysis restricts the analysis population to participants who adhered to the protocol as designed, excluding those with major protocol deviations, inadequate treatment exposure, or missing primary endpoint data. This approach aims to evaluate treatment efficacy under optimal conditions by focusing on participants who received treatment as intended. While per-protocol analysis may provide cleaner estimates of biological efficacy, it sacrifices the protection against bias that randomization provides.

The criteria defining the per-protocol population must be specified prospectively in the statistical analysis plan before unblinding occurs. Common exclusion criteria include receipt of prohibited concomitant medications, inadequate drug exposure or compliance below a specified threshold, major protocol violations affecting efficacy assessment, and missing primary endpoint data. The criteria should be clinically and scientifically justified and should not be influenced by knowledge of outcomes or treatment assignments.

Per-protocol analysis typically serves as a supportive or sensitivity analysis rather than the primary analysis in regulatory submissions, because exclusion of randomized participants may introduce selection bias. If treatment assignment influences the likelihood of protocol adherence, the per-protocol population may no longer be balanced with respect to prognostic factors, potentially confounding the treatment comparison. However, per-protocol analysis provides valuable complementary information, and concordance between ITT and per-protocol results strengthens confidence in trial findings.