Statistical Principles for Clinical Trials, incl. Addendum on Estimands and Sensitivity Analysis
ICH E9(R1) sets out the statistical principles that govern the design, conduct, analysis, and evaluation of clinical trials, and adds the estimand framework through its 2019 addendum. The original E9 (1998) covers trial design, pre-specification of analyses, control of bias, missing data, and the roles of confirmatory and exploratory objectives. The R1 addendum introduces a structured way to define precisely what a trial sets out to estimate — the estimand — and requires sensitivity analyses that test the assumptions behind the main result. Together they align a trial's objective, design, conduct, and analysis.
What this does not cover
stated in the document's own scope- It is not a good-clinical-practice standard; the ethical and operational conduct of trials follows ICH E6.
- It does not set the general framework for planning clinical studies, which is the subject of ICH E8.
- It does not prescribe the format of the study report; ICH E3 governs how results are presented.
- It is method-agnostic and does not mandate specific statistical software, models, or a single missing-data method.
Always verify against the current published text before relying on it for a submission or inspection.
Overview
ICH E9(R1) is the reference for how statistics support clinical development. The 1998 core guideline established that the credibility of a trial rests on decisions made before the data are seen — the design, the pre-specified analysis, and the control of bias. The 2019 addendum answers a question the original left implicit: what, precisely, is the trial trying to estimate? The estimand framework makes that target explicit by naming five attributes, including how events such as treatment discontinuation or rescue medication are handled. Sensitivity analysis then tests whether the conclusion holds when those assumptions are relaxed.
Legal basis & how it acquires force
E9(R1) is a harmonised scientific guideline rather than a statute; it takes effect through adoption in each ICH region. The EMA adopted the addendum as a CHMP scientific guideline, and the FDA issued it as guidance for industry. It supports the statistical expectations of marketing applications assessed under each region's medicines law, and operates alongside the good-clinical-practice framework of ICH E6, which governs how the trial generating the data is conducted.
Document structure
| Part | Covers |
|---|---|
| Considerations for overall clinical development | Trial context within a development programme, confirmatory versus exploratory trials, and scope of statistical planning |
| Trial design considerations | Design types, blinding, randomisation, comparator and endpoint selection, and multiplicity |
| Trial conduct considerations | Monitoring, interim analysis, and handling of changes during the trial |
| Data analysis considerations | Analysis sets, missing data, and evaluation of the primary and secondary questions |
| Addendum on estimands | The five estimand attributes and the alignment of objective, design, conduct, and analysis |
| Addendum on sensitivity analysis | Structured analyses that test the robustness of the estimate to its assumptions |
Key requirements
- Pre-specify the principal analyses, primary endpoints, and statistical methods in the protocol before unblinding.
- Define the estimand for each trial objective using its five attributes: treatment, population, variable, intercurrent-event strategy, and population-level summary.
- Choose an intercurrent-event handling strategy explicitly and align the trial design, data collection, and analysis to it.
- Plan sensitivity analyses that probe the robustness of the main estimate to its assumptions, keeping them distinct from supplementary analyses.
- Control the overall type I error and justify multiplicity adjustments across endpoints, comparisons, and interim looks.
- Prevent and minimise bias through blinding, randomisation, and pre-specified handling of missing data.
- Document the analysis populations, such as the full analysis set and per-protocol set, and the rationale for each.
Implementation tips
- Draft the estimand before the design is fixed — the five attributes should drive endpoint choice and data collection, not follow them.
- Agree the intercurrent-event strategy with clinical and regulatory input early, since it changes what data you must capture.
- Keep sensitivity analyses tied to specific assumptions in the main estimand rather than adding generic secondary analyses.
- Write the statistical analysis plan to trace each objective through its estimand to its primary and sensitivity analysis.
International alignment
E9(R1) applies across the EU, Japan, and the United States and is widely referenced by other regulators and by academic trialists. It sits within the ICH Efficacy series: it depends on ICH E6 for good clinical practice, connects to ICH E8 on general considerations for clinical studies, and feeds ICH E3, which reports the analyses in the clinical study report. Regional statistical review practices have converged on the estimand vocabulary since 2019.
ICH E9(R1): frequently asked questions
Quick answers to common questions about ICH E9(R1).
What is an estimand?
A precise description of what a trial sets out to estimate, defined by five attributes: the treatment, the population, the variable or endpoint, the strategy for intercurrent events, and the population-level summary.
How does sensitivity analysis differ from supplementary analysis?
Sensitivity analysis tests whether the main estimate is robust to the assumptions behind it. Supplementary analyses explore additional questions and are not a substitute for pre-specified sensitivity analysis.
Did the addendum replace the 1998 guideline?
No. The R1 addendum supplements the original E9; the two are read together, with the addendum adding the estimand and sensitivity-analysis framework.
When did the addendum reach Step 4?
The E9(R1) addendum was adopted at ICH Step 4 on 20 November 2019.
What are intercurrent events?
Events occurring after treatment starts that affect the interpretation or existence of the endpoint, such as discontinuation, rescue medication, or death — the estimand must state how each is handled.