The intention-to-treat analysis, or ITT analysis, is the principle of analyzing randomized participants in the treatment group to which they were assigned at randomization. It preserves the comparison of treatment strategies even when participants do not start the study treatment, discontinue it, switch treatments, or deviate from the protocol.
Basic principle and objective
Randomization creates comparable groups at the start of a study. If only participants who fully received a treatment or behaved particularly compliantly with the protocol were analyzed retrospectively, this advantage could be lost. The ITT analysis prevents decisions or events occurring after randomization from changing the group assignment for the primary efficacy analysis. ICH E9 describes the principle as an assessment based on the intended treatment regimen rather than the treatment actually received.
The principle does not mean that all information arising after randomization must be treated identically. Study discontinuation, treatment discontinuation, additional medication, or treatment switching can affect the clinical question and the availability of measurements. Missing data must not be ignored. The analysis must specify in advance which data are required for the estimate, how data will be collected after such events, and which assumptions underlie the handling of missing values.
Relation to the estimand framework
ICH E9(R1) assigns the analysis to a precisely defined estimand. This describes the treatment effect that answers the clinical question, including the strategy for relevant events after treatment initiation. ITT is therefore not a complete substitute for an estimand and is not an automatic instruction for every individual analysis rule. The clinical question determines whether, for example, the effect of assignment to a treatment strategy or another treatment effect should be estimated.
The required data, the primary estimator, and sensitivity analyses follow from the estimand. If the effect of a treatment strategy is being investigated, observations after discontinuation of the study treatment may remain relevant. The protocol and Statistical Analysis Plan must make this possible organizationally. Sensitivity analyses should address the same target of estimation and assess whether the conclusion remains robust to assumptions, for example those concerning missing data.
Distinction from the Full Analysis Set and per-protocol analysis
ITT is an analysis principle; the Full Analysis Set, or FAS, is a specifically defined analysis population. ICH E9 recommends the FAS as the most complete and unbiased subset possible of all randomized participants, one that approximates the ITT principle. The precise definition is specified in advance in the protocol or Statistical Analysis Plan. It must not be equated schematically with the requirement of “at least one dose and one post-baseline measurement,” because such exclusions can impair the randomized comparison.
The per-protocol analysis, by contrast, considers a more narrowly defined population without major protocol deviations. It can provide supplementary evidence for certain questions, but it does not answer the same question as an ITT-oriented primary analysis and is more susceptible to bias from post-randomization exclusions. The Full Analysis Set and per-protocol analysis therefore remain distinct terms. Their role, definition, and analysis must be specified in advance.
Even in an ITT-oriented analysis, the treatment actually received remains important for interpreting efficacy and safety. It is therefore often described additionally, without replacing the primary randomized assignment. Protocol deviations, exposure, treatment discontinuation, and concomitant treatments are recorded systematically. This information helps to contextualize the study results and assess the assumptions of supplementary analyses.
For the ITT analysis, the protocol and Statistical Analysis Plan must specify the randomized analysis population, the handling of treatment discontinuation, and data collection after intercurrent events. This specification ensures that the primary analysis estimates the effect of treatment assignment and that missing data and sensitivity analyses are consistent with the defined estimand.
Relevance for clinical trials
The practical challenge lies not only in the statistical assignment but also in continued data collection. If a participant discontinues the study treatment, information on endpoints, safety events, and subsequent therapies may still be required for the predefined question. An unclear follow-up plan generates missing data and can limit the informative value of the primary analysis. Authorities expect a transparent presentation of analysis populations, attrition, and sensitivity analyses.
Full-service CROs such as Mediconomics support the specification of estimands, analysis populations, and data collection procedures in the protocol and Statistical Analysis Plan. This includes coordinating data management and monitoring with follow-up after treatment discontinuation, statistical programming of primary and sensitivity analyses, and transparent presentation of deviations, missing data, and analysis results in the clinical study report.
Frequently Asked Questions (FAQ)
Are participants excluded from an ITT analysis after treatment discontinuation?
No, the assignment to the randomized group remains unchanged. Which data collected after discontinuation are included in the specific analysis depends on the predefined estimand and analysis plan.
Is the Full Analysis Set always identical to ITT?
No. The FAS is the operationalized, prospectively defined analysis population and is intended to approximate the ITT principle as closely as possible. Its definition must be justified and transparent.
Why is a per-protocol analysis also performed?
It can provide a supplementary perspective, for example for assessing consistency. Because of possible bias from post-randomization exclusions, it does not replace the ITT-oriented primary analysis.
Regulatory references
- ICH E9 “Statistical Principles for Clinical Trials” – describes the ITT principle and the Full Analysis Set.
- ICH E9(R1) “Addendum on Estimands and Sensitivity Analysis” – requires the analysis to be aligned with the estimand.
- EMA Guideline on Missing Data – explains the consequences of missing data for ITT-oriented analyses.
- ICH E8(R1) “General Considerations for Clinical Studies” – requires the statistical analysis to be specified in advance in the protocol.