per protocol

An analysis that excludes subjects who were not compliant with the intervention regimen (took prohibited concurrent treatments, for example).

Note: The population of interest is thus highly filtered and not influenced by factors that could dilute the effect of treatment. This form of analysis may therefore produce a larger estimate of treatment effect.

(Related concept: intention-to-treat analysis)

Other definition:

An analysis of the subset of subjects of a randomised controlled trial who have complied with the protocol sufficiently for there to be a good chance that their outcomes demonstrate the effect of the intervention of interest. This subset may be defined on the basis of exposure to the intervention, the availability of measurements and the lack of major departures from the protocol. This analysis strategy may be subject to bias because the non-compliance with the protocol may be due to the intervention.