Scenario Resolution: Before visit windows and outcome collection schedules are locked in an electronic Data Capture (EDC) or eCOA platform, clinical development and biostatistics teams must formulate the exact treatment-effect question as an ICH E9(R1) estimand. This requires translating the trial objective into five explicit attributes: the treatment condition and comparator, the target patient population, the variable (endpoint) obtained per participant, the pre-specified handling strategy for every foreseeable intercurrent event, and the population-level summary metric. Locking this architecture in the protocol and Statistical Analysis Plan (SAP) prevents the fatal operational error of letting post-hoc imputation methods dictate the clinical question.
What ICH E9(R1) is, and what it is not
The International Council for Harmonisation (ICH) formally adopted the ICH E9(R1) Addendum on Estimands and Sensitivity Analysis in Clinical Trials at Step 4 on 20 November 2019 (with the formal guideline document dated 17 November 2019). On the central ICH efficacy guidelines status page, it is classified under Step 5 implementation. Crucially, ICH E9(R1) is an addendum to the foundational ICH E9 guideline (Statistical Principles for Clinical Trials, published in 1998), rather than a replacement. The principles of randomization, blinding, and trial design outlined in the 1998 text remain active regulatory standards across major global health authorities.
In the United States, the Food and Drug Administration (FDA) issued the addendum as final Guidance for Industry in May 2021 under docket FDA-2017-D-6113 (with page content current as of 11 May 2021). The corresponding Federal Register notice published on 12 May 2021 confirmed that this final text superseded the preliminary draft issued on 31 October 2017. Clinical trial teams must ensure they reference the May 2021 final guidance rather than archived draft documents. In the European Union, the European Medicines Agency (EMA) published the text as a Step 5 scientific guideline (EMA/CHMP/ICH/436221/2017), adopted by the Committee for Medicinal Products for Human Use (CHMP) on 30 January 2020 with a formal effective date of 30 July 2020.
Write the estimand before you write the visit schedule
In historical clinical development practice, protocol teams frequently selected an efficacy variable (such as change from baseline in disease activity score at week 24) and immediately proceeded to build the schedule of assessments, visit windows, and case report forms (CRFs). Biostatisticians were subsequently left to resolve treatment discontinuations, rescue medication use, and missed visits during post-hoc statistical modeling. ICH E9(R1) systematically dismantles this disconnected workflow by enforcing a structured hierarchy:
graph TD
A["Trial Objective: Clinical Question"] --> B["Estimand: 5 Core Attributes Defined"]
B --> C["Trial Design & Data Collection Schedule"]
C --> D["Main Estimator: Aligned to Primary Estimand"]
D --> E["Sensitivity Analysis: Same Estimand, Robustness Tested"]To construct a compliant estimand, the multidisciplinary team (clinicians, biostatisticians, clinical operations leaders, and regulatory strategists) must define five core attributes in advance of trial initiation:
Treatment Condition: The specific investigative treatment regimen of interest and the comparator condition (e.g., active drug 50 mg daily versus placebo, including background standard of care).
Target Population: The broader population of patients targeted by the clinical trial question, defined by inclusion and exclusion criteria, baseline disease characteristics, and geographical/demographic parameters.
Variable (Endpoint): The individual-level measure obtained for each patient (e.g., change from baseline in HbA1c at 24 weeks, progression-free survival time, or 6-minute walk distance).
Intercurrent Event Strategies: The explicit clinical strategies chosen to account for each foreseeable post-baseline event that affects either the interpretation or existence of the endpoint measurements.
Population-Level Summary: The mathematical summary measure used to compare treatment groups (e.g., difference in mean change, hazard ratio, odds ratio, or difference in responder proportions).
The table below illustrates how two different clinical development programs translate distinct trial objectives into fully specified estimand attributes:
| Estimand Attribute | Chronic Pain Development Program | Oncology Second-Line Solid Tumor |
|---|---|---|
| Trial Objective | Evaluate pain relief in moderate-to-severe OA | Assess progression delay in refractory metastatic cancer |
| Treatment Condition | Oral Investigational Agent 100 mg vs Placebo | Monoclonal Antibody IV vs Standard Chemotherapy |
| Target Population | Adults with symptomatic knee osteoarthritis | Adults with EGFR-mutated metastatic NSCLC post-progression |
| Variable (Endpoint) | Weekly average numerical pain rating score at Week 12 | Progression-Free Survival (PFS) according to RECIST v1.1 |
| Intercurrent Events | Rescue analgesics (Hypothetical); Discontinuation (Treatment Policy) | Crossover therapy (Hypothetical); Toxic death (Composite) |
| Population Summary | Difference in adjusted mean pain score at Week 12 | Hazard Ratio for PFS over 24 months of follow-up |
Intercurrent events are not missing data
A central conceptual breakthrough of ICH E9(R1) is the rigorous boundary established between intercurrent events (ICEs) and missing data. Historically, trial protocols routinely conflated patient treatment discontinuation with study withdrawal, treating both as missing observations to be resolved through statistical imputation algorithms like Last Observation Carried Forward (LOCF) or Mixed Models for Repeated Measures (MMRM).
"Intercurrent events are events occurring after treatment initiation that affect either the interpretation or the existence of the measurements associated with the clinical question of interest... Missing data are data that would be meaningful for the analysis of a given estimand but were not collected."
— ICH E9(R1) Addendum, Glossary (intercurrent events; missing data)
Under ICH E9(R1), an intercurrent event is recognized as an intrinsic part of the clinical reality of administering therapy. For instance, when a patient discontinues their assigned study medication due to an adverse event or initiates unblinded rescue medication, their subsequent health status is not a blank data cell; it is an observed clinical reality. Conversely, missing data represent values that were intended to be collected under the chosen estimand framework but were omitted due to operational failures, missed clinic visits, or loss to follow-up.
Five strategies, five different clinical claims
ICH E9(R1) defines five distinct methodological strategies to address intercurrent events. Each strategy corresponds to a fundamentally different clinical question, changes a specific attribute of the trial, and carries distinct operational and analytical requirements:
1. Treatment Policy Strategy: The variable of interest is evaluated regardless of whether the intercurrent event occurs. This strategy addresses the clinical question of treatment effectiveness in routine practice where patients may discontinue or switch medications. It requires continuous data collection following treatment discontinuation. Limitation: Treatment policy cannot be applied to terminal events (death) because physical values after death do not exist.
2. Hypothetical Strategy: A scenario is envisaged in which the intercurrent event would not occur (e.g., what would the treatment effect be if emergency rescue medication had not been administered?). This strategy is widely used to isolate biological drug efficacy, but it relies on strong, verifiable statistical modeling assumptions regarding counterfactual outcomes.
3. Composite Variable Strategy: The intercurrent event is integrated directly into the definition of the endpoint variable. For example, a successful treatment response may be defined as achieving a 50% symptom reduction without requiring rescue medication or discontinuing study drug. A patient experiencing the ICE is categorized as a treatment failure.
4. While on Treatment Strategy: The response is evaluated across the time period prior to the occurrence of the intercurrent event. This strategy is frequently applied to chronic maintenance therapies, measuring annualized event rates or patient-reported outcomes while participants adhere to the assigned regimen.
5. Principal Stratum Strategy: The clinical question is restricted to the subpopulation of patients in whom the intercurrent event would not occur under either treatment assignment. Critical Warning: A principal stratum estimand is fundamentally distinct from a naive per-protocol completer analysis. Comparing observed completers on active drug against observed completers on placebo confounds treatment effect with post-randomization patient characteristics, introducing severe selection bias.
The comparative matrix below details how each strategy modifies the estimand structure and what data collection mandates follow:
| Strategy Name | Attribute Modified | Data Follow-Up Mandate | Analytical Considerations |
|---|---|---|---|
| Treatment Policy | Treatment Condition definition | Mandatory continuous follow-up after ICE | Estimates real-world effectiveness; inapplicable to death |
| Hypothetical | Variable (Counterfactual scenario) | Follow-up depends on imputation model | Requires robust validation of missingness assumptions |
| Composite Variable | Variable definition (Binary/Categorical) | Follow-up until ICE occurrence | Combines clinical outcomes; may mask component nuances |
| While on Treatment | Variable observation window | Follow-up restricted to on-treatment phase | Measures on-drug performance; ignores post-stop rebound |
| Principal Stratum | Target Population definition | Baseline profiling + post-ICE verification | Requires specialized causal inference modeling |
What the protocol and SAP must lock
To ensure audit readiness and regulatory alignment, the clinical study protocol and Statistical Analysis Plan must explicitly document the estimand architecture before trial commencement. Clinical operations teams must implement explicit EDC and eCOA tracking mechanisms to record the exact date, timing, dosage, and underlying clinical reason for every intercurrent event (such as differentiating between discontinuation due to disease progression versus intolerable toxicity).
Furthermore, the SAP must specify the main estimator aligned to the primary estimand and pre-specify a comprehensive sensitivity analysis plan. Under ICH E9(R1), sensitivity analyses must target the exact same estimand as the primary analysis, systematically probing the stability of the conclusions against departures from core statistical assumptions (such as non-random dropout or missingness mechanisms). Supplementary analyses, by contrast, address alternative clinical questions and should be designated as secondary or supportive.
By embedding the ICH E9(R1) framework into early protocol design, clinical development sponsors ensure that statistical analyses remain truthful to the trial's therapeutic intent, regulatory submissions withstand international scrutiny, and endpoint evidence reflects genuine patient benefit.