People in a trial may stop an assigned intervention, change what they receive or use it differently from the plan. How researchers analyse those participants affects the question the results can answer.
The Cochrane Handbook’s risk-of-bias chapter discusses the effect of assignment to an intervention and analyses based on the original randomised groups. This is the central idea behind intention-to-treat analysis.
Imagine a fictional trial assigning people to receive reminder messages or usual information. Some assigned to reminders never read them. Analysing participants according to their original assignment addresses the effect of offering that strategy in the trial, rather than only its effect among people who followed it fully.
Removing everyone who did not follow the plan can change the comparability created by randomisation. Those who remain may differ in ways related to the outcome. A per-protocol analysis therefore needs its own justification and careful methods; it is not automatically the more truthful version.
Look beyond the label
Check who was included, how missing outcomes were handled and whether participants stayed in their assigned analysis groups. Merely writing “intention to treat” does not resolve missing data or guarantee that every methodological choice was appropriate.
Keep the research question explicit in your notes: the effect of assignment, or an effect under adherence to a specified intervention? Those questions can both matter, but they are not interchangeable.
This distinction helps explain why two analyses of the same trial may report different estimates without one being a simple arithmetic mistake. It is a guide to reading research methods, not an instruction for changing a treatment plan or interpreting an individual’s response without clinical advice.
Image: a thematic illustration, not a photograph of this example.

