Intention-to-treat analysis (revision 16)
Old revision·16:52, 22 Jul 2025·CiteBot
| Intention-to-treat analysisTrial methodology | |
|---|---|
| Abbreviation | ITT |
| Principle | Analyse as randomised, not as treated |
| Preserves | The balance randomisation created |
| Contrast with | Per-protocol analysis |
| Topic infobox · conventions | |
Intention-to-treat analysis includes every randomised participant in the group to which they were assigned, regardless of whether they received the assigned treatment, adhered to it, or completed the trial. Its purpose is to preserve the comparability that randomisation created.[1]
The alternative — analysing only those who completed as assigned — breaks that comparability, because completion is not random. Participants who stop because a treatment is not working, or because it is poorly tolerated, differ systematically from those who continue.[2]
Intention-to-treat generally produces a more conservative estimate than a per-protocol analysis, and it answers a different question: what happens when treatment is offered, rather than what happens when it is taken as directed.[2] Which question a trial should answer depends on the comparison it was designed to make.[3]
Why analysing as treated fails
[edit]Randomisation makes the groups comparable on everything, measured and unmeasured. Any post-randomisation selection — dropping non-completers, excluding non-adherers, analysing by treatment actually received — reintroduces confounding by the very factors that led to the selection.[2]
The classic demonstration is that adherence itself predicts outcome even in a placebo arm: participants who take placebo reliably do better than those who do not, because adherence is a marker for other health-related behaviour. An analysis conditioning on adherence therefore finds an effect where none exists.[2]
The practical consequence is that a striking difference between an intention-to-treat and a per-protocol result is informative about differential dropout rather than about the treatment.[1]
Missing data
[edit]Strict intention-to-treat requires outcome data on every randomised participant, which trials rarely have. How missing data are handled determines what an "intention-to-treat" analysis actually is, and the label alone does not say.[1]
| Approach | Assumption |
|---|---|
| Complete-case | Missingness unrelated to outcome — usually implausible |
| Last observation carried forward | Outcome froze at withdrawal — usually implausible |
| Multiple imputation | Missingness explained by observed variables |
| Reference-based imputation | Withdrawers behave like the control group thereafter |
Obesity trials in this field commonly use a reference-based approach for the treatment-policy estimand, which assumes that participants who stop revert towards placebo behaviour — a conservative and reasonably defensible assumption given the regain observed after discontinuation.[1]
Estimands
[edit]Current guidance frames the question as choosing an estimand — a precise statement of what quantity is being estimated, including how intercurrent events such as discontinuation and rescue medication are handled.[1]
Two are common in this field. The treatment policy estimand asks what happens when treatment is initiated, counting all data regardless of adherence; the trial product estimand asks what happens if treatment is taken as directed throughout. The second gives larger effects, and the two are frequently reported side by side.[4]
Neither is more correct. They answer different questions, and a secondary source quoting one without saying which is quoting a number without its definition. This is why STEP and SURMOUNT figures sometimes differ between sources reporting the same trial.[2]
See also
References
- ^ a b c d e International Council for Harmonisation, E9(R1): Estimands and Sensitivity Analysis in Clinical Trials (2019).
- ^ a b c d e Hollis S, Campbell F. "What is meant by intention to treat analysis? Survey of published randomised controlled trials." BMJ 319(7211):670–674 (1999). PMID 10480822.
- ^ International Council for Harmonisation, E10: Choice of Control Group and Related Issues in Clinical Trials (2000).
- ^ Wilding JPH, Batterham RL, Calanna S, et al. "Once-weekly semaglutide in adults with overweight or obesity." New England Journal of Medicine 384(11):989–1002 (2021). PMID 33567185.