The statistical methods section of a CSR: what ICH E3 requires and how EU CTIS sponsors actually write it

The statistical methods section is one of the most heavily templated parts of a clinical study report, yet it is also where sponsors diverge most in how much they restate versus how much they push into an appendix. ICH E3 sets the expectation; the clinical study reports now visible in the EU Clinical Trials Information System (CTIS), submitted under Regulation (EU) 536/2014, show how sponsors resolve it in practice. This article walks through what E3 asks for, then reads across real CTIS reports to show the recurring structure and, in particular, the cross-reference to Appendix 16.1.9 that lets sponsors avoid restating the full statistical analysis plan.

What ICH E3 requires

E3 places statistical content in three distinct locations and expects each to do a different job: the synopsis, the results body, and the appendices ⁴³. The core statistical requirements live in two body sections plus one appendix.

Section 9.7: statistical methods planned in the protocol and determination of sample size

E3 splits Section 9.7 into two parts ²⁵.

9.7.1 Statistical and analytical plans. This subsection describes what was planned in the protocol, and any changes made before outcome results were available. E3 is explicit that the report should describe the analyses, comparisons, and tests that were planned, not simply the ones that were ultimately run ². The listed elements include:

The planned basis for comparison when a critical measurement is taken more than once (for example, an average across the study, values at particular timepoints, completer-only values, or the last on-therapy value) ².
The planned analytical approach where more than one is plausible (change from baseline, slope analysis, life-table analysis) ².
Whether the primary analysis adjusts for covariates ².
Any planned reasons for excluding patients whose data were available ².
Any subgroups to be examined separately ².
How categorical responses were defined, if used (global scales, severity scores, responses of a given size) ².
Planned monitoring of study results, and, where a data monitoring committee existed, its composition, operating procedures, and blinding-maintenance procedures ².
The frequency and nature of interim analyses, any specified circumstances for terminating the study, and any statistical adjustments made because of interim analyses ²⁵.

9.7.2 Determination of sample size. This subsection gives the planned sample size and the basis for it (statistical or practical), the calculation method with derivation or source reference, the estimates used with an explanation of how they were obtained, the difference the study is designed to detect (for a superiority study), and, for a positive-control non-inferiority study, the difference considered unacceptably large, i.e., the difference the study is designed to exclude ⁵.

Section 11.4: statistical and analytical issues

Section 11.4.2 covers the important features of the analysis as actually performed: the particular methods used; adjustments for demographic or baseline measurements or concomitant therapy; handling of drop-outs and missing data; adjustments for multiple comparisons; special analyses of multicentre studies; and adjustments for interim analyses. Any changes to the analysis made after breaking the blind must be identified ³.

The E3 annex adds a level of statistical detail expected for each primary efficacy variable: the underlying statistical model, the clinical claim stated in precise statistical terms, the methods used to estimate effects and construct confidence intervals, the assumptions behind those methods, any rationale for data transformations, and the test statistic, sampling distribution, computed value, p-value, and intermediate summary data. For multicentre studies analysed by ANOVA, an ANOVA table with centres, treatments, their interaction, error, and total is expected ⁸. The annex also flags that subgroup analyses need care: if they were not preplanned, they ordinarily will not support definitive conclusions ⁸.

Appendix 16.1.9: documentation of statistical methods

Appendix 16.1.9 is titled "Documentation of statistical methods" and is meant to hold the detailed statistical documentation for the study ¹⁰³. E3 draws the division of labour cleanly: the statistical analysis should be described in the text of the report so that clinical and statistical reviewers can follow it, while the detailed documentation of the statistical methods is placed in Appendix 16.1.9 ³. In short, the body carries the summary and rationale; Appendix 16.1.9 carries the full method detail ³.

The synopsis and changes to the plan

E3 treats the synopsis as a stand-alone summary of key efficacy and safety results plus supporting information (population, disposition, important deviations, compliance), and specifically says cross-references to other CSR sections should be avoided there ⁴. That constraint matters: the appendix cross-reference discussed below belongs in the body, not the synopsis.

On changes to the analysis, E3 distinguishes timing. Changes made before outcome results were available are described in the statistical methods section ². Any change in the conduct of the study or the planned analyses after the study started should be described with its timing, reasons, decision process, responsible persons, and the data available at the time ⁵. Changes made before breaking the blind generally carry limited implications; changes after blind-breaking are the ones E3 most wants identified ⁵³.

How EU CTIS clinical study reports actually structure the section

Reading across CSRs and SAPs now available through CTIS, the numbering varies by sponsor but the content is remarkably consistent ⁷⁹⁸⁰⁸¹⁸²⁸⁷⁹²⁹⁸⁹⁹¹⁰²¹⁰³. The section is typically organised as: analysis sets, general conventions, a primary/secondary analysis core, missing-data handling, interim analyses, multiplicity, sample size, and supportive/sensitivity analyses.

Analysis populations. Sponsors define populations under headings such as "Analysis Populations," "Analysis Sets," or "Study Population" ⁹²⁹⁹¹⁰³. The recurring set is the Full Analysis Set (FAS) or ITT population, the safety population, and the per-protocol population, sometimes with additional biomarker, PK, evaluable, completer, or subgroup sets ⁸¹⁸²⁹⁹¹⁰³⁸⁵¹⁴²¹³⁰. The FAS/ITT is usually all randomised subjects analysed by assigned treatment regardless of what was received or of protocol deviations; the per-protocol set excludes major deviations; the safety set is all subjects who received at least one dose, analysed as treated ¹⁴¹⁶¹⁹²⁵¹⁵¹⁷²⁷. Representative wording: "Data will be analyzed in the following populations: Full Analysis set (FAS) ... Safety population ..." ⁹⁹ and "The full analysis set includes all randomized patients ... irrespective of protocol violations and therapy dropouts" ¹⁶.

Estimands and intercurrent events. Some, but not all, CTIS reports adopt ICH E9(R1) estimand language explicitly. Examples include "As detailed in the ICH E9 (R1) addendum, 5 attributes will define the estimand framework in this trial" ²¹ and "Statistical analyses will be performed according to the principles of the ICH guideline E9(R1)" ³⁶, with intercurrent events such as study-drug discontinuation and use of new anti-cancer therapy named directly ¹⁹²⁶²³. Many reports still define only ITT/FAS, PP, and safety sets without estimand framing ²⁶.

General conventions. A short block states the software (frequently SAS 9.4 or higher), how continuous and categorical data are summarised (mean/SD/median/min/max; counts and percentages), how baseline is defined, and the treatment-group and by-cohort presentation rules ⁸⁰⁸¹⁸⁸. Representative wording: "All analyses will be performed using SAS version 9.4 or higher." ⁸⁰.

Primary and secondary analysis core. This is almost always a table linking each endpoint to its statistical method, analysis population, missing-data or censoring approach, and hypothesis ⁷⁹⁸²⁸⁷. Common methods are the stratified log-rank test and stratified Cox proportional hazards model for time-to-event endpoints, and Miettinen and Nurminen or logistic regression for response endpoints ⁷⁹⁸²⁸⁷⁸⁸⁹⁵. Representative wording: "Estimation: Stratified Cox model with Efron's tie handling method" ⁷⁹⁸²⁸⁷.

Missing data. A standard subsection, often titled "Missing Data," states the default (frequently no imputation unless specified), non-responder handling for binary endpoints, censoring rules for time-to-event endpoints, and any LOCF, observed-case, worst-case, or multiple-imputation approaches reserved for sensitivity analyses ⁸⁰¹⁰²¹⁰³⁷⁹⁸²⁸⁵⁹⁸. Representative wording: "In general, missing data will not be imputed for the purpose of data analysis, unless otherwise specified." ⁸⁰ and "ITT missing are considered non-responders" ⁷⁹.

Interim analyses and multiplicity. These are usually explicit, separate subsections and are tied together ⁸⁷¹⁰¹¹⁰²¹⁰³. Sponsors describe whether interims exist, their timing and type (efficacy, futility, interim OS), SAP approval timing, and any alpha-spending or decision rules; the multiplicity subsection describes overall type I error control and the testing hierarchy across endpoints, populations, and looks ⁸⁰⁹⁸⁷⁹⁸²⁸⁷. Representative wording: "A first version of the SAP will be prepared prior to the inclusion of the first study participant and the SAP will be approved prior to the first planned analysis, ie, interim analysis 1." ⁸⁰. Where analyses are descriptive, sponsors state so plainly: "No formal statistical hypothesis testing is planned." ⁸⁰ and "No adjustment for multiplicity is required." ⁸⁰.

Determination of sample size. Sponsors write this as a short assumptions-driven justification: planned N, target power, alpha, the effect to detect, and the variance and dropout assumptions, with the source of the estimates named (a prior phase III study, a published trial or meta-analysis, or a specific earlier study) ¹⁰⁵¹⁰⁶¹⁰⁹¹¹³¹¹⁷¹²³¹²⁸. Representative wording: "The sample size is determined in order to have 90% power for meeting the primary hypothesis." ¹⁰⁶; "Alpha set to 2.5% for one-sided test" ¹⁰⁹; "assuming an estimated 20% dropout rate at 18 months" ¹¹³. Non-inferiority studies state the margin explicitly, e.g., "Non-inferiority margin: 30%" ¹⁰⁷, matching the E3 requirement to state the difference the study is designed to exclude ⁵.

Sensitivity and supportive analyses. Many reports add a dedicated subsection or table for sensitivity, subgroup, and supportive analyses: alternative censoring rules, investigator-versus-BICR comparisons, exclusion of protocol deviations, and treatment-switch adjustments such as RPSFT, IPCW, or two-stage methods ⁸¹⁸³⁹⁵¹⁰⁰. Representative wording: "Exploratory analyses of overall survival adjusting for the impact of treatment switching will be performed" ¹⁰⁰.

The Appendix 16.1.9 cross-reference in practice

The most consistent finding across CTIS reports is that sponsors do not restate the full statistical methodology in the body. Instead they summarise it and point to Appendix 16.1.9, where the SAP and supporting documents live ⁴⁶⁵⁶⁵⁷⁵⁸⁶²⁶⁴. The cross-reference wording is stereotyped, and the following are drawn directly from CTIS reports:

"Full details are provided in the statistical analysis plan (SAP) dated 21 May 2025 (Appendix 16.1.9)..." ⁵⁶
"The statistical methods summarized in this section are those documented in the final statistical analysis plan (SAP) Version 1.0, dated 19 February 2025 (Appendix 16.1.9)." ⁵⁷
"The analysis of the study data was performed according to the statistical considerations and methods described in the SAP, Version 1.0, dated 11 Sep 2023 (Appendix 16.1.9)." ⁵⁸
"Analyses were based on the Statistical Analysis Plan (SAP) Amendment 3, dated 28 Apr 2023 (Appendix 16.1.9)." ⁶⁴
"The statistical analysis used will be described in the text of the report, with detailed documentation of statistical methods presented in appendix 16.1.9 of the study report." ⁶²

Two structural points are worth noting for medical writers. First, the cross-reference is version-and-date specific: sponsors name the SAP version and date (and any amendment number) at the point of reference, which is what makes the summary-plus-pointer approach auditable rather than vague ⁵⁶⁵⁷⁵⁸⁶⁴. Second, sponsors frequently chain the reference across the protocol, the body, and the appendix, e.g., "The planned statistical analysis is described in Section 12 of the clinical study protocol. Please refer to CSR Section 9.7 and Section 9.8 for detailed statistical methods" alongside identification of the specific SAP amendments ⁴⁹.

The appendix itself is labelled to match E3, appearing as "Appendix 16.1.9 Documentation of statistical methods" or, in expanded form, "Appendix 16.1.9 Documentation of Statistical Methods and Supporting Statistical Analysis" ⁴²⁴³⁴⁵⁵²⁵⁴⁵⁹⁶⁵. What sponsors package into it, based on the CTIS reports, is:

The SAP or final statistical analysis plan, often with version and date ⁴²⁴³⁴⁵⁵⁶⁵⁷⁵⁸⁵⁹⁶¹⁶⁵.
Interim analysis plans and interim analysis documentation ⁴⁶⁴⁸⁵¹⁵⁸⁶⁶.
Supporting method detail: analysis-set definitions, missing-data handling, covariate adjustments, multiplicity strategy, and model specifications, with some reports showing explicit sub-sections such as "2.13 Interim analysis" and "5.4 Statistical models" ⁴⁶⁴⁸⁵⁶⁵⁷⁵⁸⁶²⁶⁴⁴⁴.
In some cases, DMC/IDMC-related analysis or review material referenced within the documentation ⁴⁸.

Practical takeaways

For a regulatory writer assembling a CSR under EU CTR, the E3-compliant pattern that CTIS reports converge on is: describe the analysis in the body at a level a clinical and statistical reviewer can follow (analysis sets, primary and secondary methods, missing-data and multiplicity approach, sample size rationale), and place the full SAP and interim analysis documentation in Appendix 16.1.9, referenced by exact version and date ³⁵⁶⁵⁷. Keep the synopsis self-contained and free of the appendix cross-reference, since E3 asks the synopsis to stand alone ⁴. Identify any post-unblinding changes to the analysis explicitly, which is the one place E3 gives the least latitude ³⁵. And where an estimand framework applies, aligning the population and intercurrent-event language to ICH E9(R1) in the body, with the detailed strategy in the SAP, is now an established approach in CTIS filings rather than an outlier ²¹³⁶²³.

ICH E3 Statistical Methods Requirements and EU CTIS Sponsor Practice in Clinical Study Reports