On the other hand, per-protocol analysis does the first few steps, but when it's time to analyze the data, anyone noncompliant or no longer in the study has their data excluded from the final analysis. Per-protocol analysis has been described as a way to estimate the effect of receiving a healthcare treatment, whereas intention-to-treat measures the effect of assigning a patient to treatment.
Problems with per-protocol analysis
There is no question whether per-protocol analysis is an important component of clinical trial data. However, sometimes it's used alone, and that's where the trouble starts. When the only data you have is based on per-protocol analysis, there can be several potential issues:
Violating randomization
In per-protocol analysis, randomization is thrown out the window as participants from the randomized groups are disqualified for various reasons. Randomization isn't totally random.
It's meant to be a controlled and balanced randomization that allows you to compare two mixed groups of people. When people start getting kicked out of the study data for any reason, different biases can crop up as the groups become increasingly less balanced.