Intention-to-treat analyses and missing outcome data: a tutorial
![](https://www.evidence4health.org/sites/readit/files/styles/nugget-mobile/public/content/paragraphs/text/images/2024-05-23/KD%20pub_May%2024_image%207.jpg?itok=g3WcLmln×tamp=1716465022)
This tutorial focuses on “intention-to-treat” analyses and missing outcome data in systematic reviews. There is a lack of consensus on the definition of the ITT approach. We will explain the principles of an intention-to-treat analysis, and outline the key issues you need to consider when planning, conducting and writing up your systematic review.
1 WHAT IS AN INTENTION-TO-TREAT ANALYSIS?
The authors of studies included in systematic reviews may use the term “intention-to-treat” or “intent-to-treat” (ITT) to describe the approach taken when reporting and analyzing outcome data. The ITT approach has two principles.
Principle A: Outcome data are reported and/or analysed according to the participant's assigned intervention, regardless of the intervention they actually received or their adherence to their assigned intervention. For randomised controlled trials, this approach is sometimes referred to as an “as-randomised” analysis.
This principle is not met if study authors:
(i) Report and/or analyze outcome data for participants according to the intervention they actually received (this approach is sometimes referred to as an “as-treated analysis”).
(ii) Report and/or analyze outcome data only for participants who adhered sufficiently to their assigned intervention (this approach is sometimes referred to as a “per-protocol analysis”).
Study authors make decisions about which approach to take based on whether they are interested in determining the effect of allocation to an intervention (regardless of whether the intervention was received as intended), the effect of receiving an intervention, or the effect of adhering to an intervention (as specified in the trial protocol).
Principle B: Outcome data are measured for all randomised participants.
If some participants do not contribute data for the outcome of interest at the required follow-up time (i.e., there are missing outcome data), data may be imputed. Various imputation methods are available, from simply assuming that all participants with missing data had a particular outcome (e.g., study authors may assume that all participants with missing data experienced a poor outcome, such as treatment failure), to more complex methods such as multiple imputation.
This principle is not met if study authors report and/or analyze outcome data only for participants with nonmissing outcome data (this approach is sometimes referred to as a “complete-case analysis”)......................