ICH E9 (Statistical Principles for Clinical Trials)

ICH E9, formally titled “Statistical Principles for Clinical Trials,” is one of the foundational efficacy guidelines issued by the International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH). Finalized in 1998, it sets out the statistical methodology regulators expect to see underpinning any confirmatory clinical trial submitted in support of a marketing application, and it remains the reference document cited across trial protocols, statistical analysis plans (SAPs), and regulatory review to this day.

What ICH E9 covers

ICH E9 is broad by design — it is not a single statistical technique but a set of principles meant to apply across therapeutic areas and trial designs. Its core areas include:

• Trial design and hypothesis structure. ICH E9 distinguishes superiority, non-inferiority, and equivalence designs, each of which implies a different hypothesis, margin, and statistical testing logic that must be pre-specified before the trial is analyzed.
• Randomization and blinding. The guideline addresses how randomization should be generated, allocated, and concealed, and how blinding should be maintained and, where broken, documented.
• Sample size and power. A trial’s sample size must be justified statistically against a pre-specified effect size, variability assumption, and power target — not arrived at after the fact.
• Analysis populations. ICH E9 defines the Full Analysis Set (FAS) — the population as close as possible to all randomized subjects, analyzed according to the treatment they were assigned, operationalizing the intention-to-treat (ITT) principle — and the Per-Protocol Set (PPS), the subset who received treatment as specified without major protocol deviations. Which population serves as primary, and which as sensitivity/supportive, depends on the trial’s design (see the non-inferiority example below).
• Missing data. The guideline requires a pre-specified strategy for handling missing observations, since ad hoc or outcome-dependent handling can bias results.
• Multiplicity. Where a trial tests multiple endpoints, subgroups, comparisons, or looks at accumulating data via interim analyses, ICH E9 requires that the overall Type I error rate (the chance of a false-positive finding) be controlled across the whole set of tests, not just within each one individually.

The unifying thread across all of these is pre-specification: under ICH E9, the design, populations, and analysis plan are fixed before the trial is unblinded, not adapted afterward to fit whatever pattern the data happen to show.

Why analysis-population choice matters: a non-inferiority example

ICH E9 explicitly notes that the FAS/ITT population and the PPS behave differently depending on trial design. In a superiority trial, non-adherence and protocol deviations tend to dilute a real treatment effect toward the null in both arms, so an ITT/FAS analysis tends to underestimate efficacy relative to a per-protocol analysis, which by discarding non-adherers can inflate the apparent effect — so ICH E9 recommends the FAS as primary for superiority trials, with the PPS as a supportive sensitivity check. In a non-inferiority trial, that same dilution effect can push results toward, not away from, a false finding of non-inferiority, so ICH E9 recommends analyzing both the FAS and PPS as co-primary, with a genuine non-inferiority conclusion generally requiring both to agree.

The ICH E9(R1) addendum and the estimand framework

ICH E9 does not stand alone. In November 2019, ICH finalized ICH E9(R1), an addendum that substantially extends — rather than replaces — the 1998 guideline. E9(R1) was written to close a gap the original guideline left largely implicit: ICH E9 assumed, without fully spelling out, what specific treatment-effect question a trial’s primary analysis was meant to answer, particularly when participants experience intercurrent events after randomization — treatment discontinuation, use of rescue medication, death, or switching to another therapy — that complicate what the originally planned endpoint measurement actually means.

E9(R1) addresses this by introducing the estimand framework: a structured way of defining, before a trial is analyzed, exactly what treatment effect it is designed to estimate, specified through five attributes agreed together — the target population, the treatments compared, the variable/endpoint, the strategy for handling intercurrent events, and the population-level summary measure. Where ICH E9 tells a trial team how to design, randomize, size, and analyze a trial with rigor, ICH E9(R1) tells them, more precisely than the original guideline did, what question that rigor is actually being applied to answer — particularly once real-world complications like treatment discontinuation are accounted for.

Who ICH E9 applies to and where it sits

ICH E9 (and its R1 addendum) are ICH’s own guidelines, adopted by ICH’s regulatory members and observers — including the U.S. FDA, the European Medicines Agency (EMA), and Japan’s PMDA, among others — into their own regional guidance and review practice. A sponsor’s statistical analysis plan referencing “ICH E9 principles” or “an ICH E9(R1)-consistent estimand” is invoking this specific guideline, not a generic claim of statistical rigor.