Estimand Strategies

[bms63]

Goal

Create a vignette showcasing how to use admiral with Estimands.

Background

This discussion was created to look at creating a vignette for how to use admiral and R code for Estimands. It came from Discussion: censoring using first date rather than last. It looks like admiral functions are enough to satisfy the needs for Estimands.

The estimand framework, as outlined in the ICH E9(R1) addendum, provides a structured approach to defining the treatment effect of interest in clinical trials, particularly in the presence of intercurrent events (ICEs). Leading industry organizations such as PHUSE, CDISC, and PharmaSUG have been actively involved in developing and disseminating strategies and frameworks for its implementation.

Here's a compilation of estimand strategies and frameworks from these organizations:

Common Estimand Strategies (from ICH E9(R1) and referenced by all organizations)

The core estimand strategies for handling intercurrent events (ICEs), as defined by the ICH E9(R1) guideline and discussed across PHUSE, CDISC, and PharmaSUG, include:

Treatment-Policy Strategy: This strategy considers clinical outcomes regardless of any intercurrent events occurring. It assesses the effect of assignment to a treatment regimen, regardless of whether the treatment was fully adhered to or modified due to ICEs.

Hypothetical Strategy: This strategy defines a scenario where certain intercurrent events are assumed not to have occurred. For example, it might envisage a scenario where participants would not need rescue medication or would not stop treatment prematurely.

Composite Strategy: With this approach, the intercurrent event itself is incorporated into the definition of the variable or endpoint being measured.

While-on-Treatment Strategy: This strategy focuses on the outcome observed only while the participant is actively receiving the assigned treatment, or more broadly, the response prior to the occurrence of an intercurrent event.

Principal Stratum Strategy: This strategy defines the population of interest based on whether an intercurrent event would or would not have occurred for an individual, regardless of their assigned treatment

Links:

• Implementation of the ICH E9(R1) Estimands Framework Using Data Standards
• PHUSE Whitepaper: Implementation of ICH E9(R1) Estimands Framework Using Data Standards

PHUSE (Pharmaceutical Users Software Exchange):

PHUSE has been instrumental in the practical implementation aspects of the estimand framework:

PHUSE Estimand White Paper: PHUSE initiated a project titled "Implementation of ICH E9(R1) Estimands Framework Using Data Standards." This ongoing project aims to develop a White Paper, in collaboration with CDISC and regulatory agencies like the FDA, to offer recommendations and best practices for implementing the estimands framework within data standards. The White Paper, expected to be published in May 2025, specifically addresses the technical implementation of the estimands framework within the data flow, which was not the primary focus of the original ICH E9(R1) guideline.

Focus Areas: The PHUSE initiative tackles challenges in data collection, tabulation, analysis, and reporting in the context of estimands and intercurrent events. It provides guidance on how to document intercurrent event occurrences and data handling.

Links: Implementation of ICH E9(R1) Estimands Framework using Data Standards - Example Document

CDISC (Clinical Data Interchange Standards Consortium)

CDISC focuses on integrating the estimand framework into its data standards to ensure clarity and consistency:

Collaboration with PHUSE: CDISC is a key collaborator in the PHUSE White Paper project, contributing expertise on data standards to help implement the estimand framework effectively.

Data Standards for Estimands: CDISC's efforts involve providing recommendations for documenting estimands implementation strategies within analysis datasets (ADaM). This includes establishing clear links between intercurrent events (ICEs) and specific data points.

Intercurrent Event Variables: CDISC suggests specific variables like ICEzzS (to identify the intercurrent event impacting a data point for a given estimand based on priority rules) and ICCyS (to identify intercurrent events impacting the interpretation or usage of a data point) within ADaM datasets.

Documentation Guidance: CDISC recommends new sections in documentation, such as "Estimands and Estimators" for a synthetic view of their implementation, and advises describing intercurrent event datasets (ADICE) and handling strategies within analysis datasets.

General Frameworks: CDISC has established critical frameworks for data management in clinical research, and "Estimands & Handling Intercurrent Events" is a recognized topic within their scope.

Links: Estimands implementation using data standards

PharmaSUG

PharmaSUG serves as a forum for discussing practical applications and considerations of estimands in clinical trials:

Practical Considerations: PharmaSUG emphasizes practical considerations for handling intercurrent events and their impact on estimand estimation, often discussing how to align multiple imputation processes with chosen ICE strategies.

Input from Clinicians and Statisticians: PharmaSUG highlights the importance of integrating both clinical and statistical input when defining estimands and intercurrent events, stressing that strategies other than 'treatment policy' should be well-justified.

Application of ICH E9(R1) Strategies: PharmaSUG presentations and papers delve into the application of the five ICH E9(R1) strategies (Treatment-Policy, Hypothetical, Composite, While-on-Treatment, and Principal Stratum) and even propose using a mix of these strategies to handle complex ICEs

Links:

• Defining estimands using a mix of strategies to handle intercurrent events in clinical trials
• Practical considerations for Intercurrent Events and Multiple Imputation

[bms63]

A Simple Vignette has been created here - #2956

Some discussion happening in examples site on extension - pharmaverse/examples#142 driven from interest from the admiral-adam-ig working group.