Interpreting the Estimand Framework From a Causal Inference Perspective

Area of Science:

• Pharmaceutical Sciences
• Biostatistics
• Clinical Trials

Background:

• The International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH) published the estimand framework in 2019 to standardize treatment effect definitions in clinical trials.
• The estimand framework defines an estimand using five attributes: treatments, variables, target populations, population-level summaries, and intercurrent events.
• Intercurrent events are handled using strategies like treatment policy, hypothetical, composite variable, while-on-treatment, and principal stratum.

Purpose of the Study:

• To interpret the ICH estimand framework through the lens of causal inference.
• To elucidate the distinctions between estimands and genuine causal treatment effects.
• To propose a novel method for integrating causal inference principles into the estimand framework.

Main Methods:

• Utilized a causal inference framework based on potential outcomes to define individual treatment effects (ITE) and average treatment effects (ATE).
• Compared the statistical presentation of ATE with the attributes of an estimand.
• Proposed mapping estimand attributes onto the statistical presentation of ATE, with intercurrent events acting as mediation mechanisms.

Main Results:

• The statistical presentation of ATE is not equivalent to an estimand, primarily due to the handling of intercurrent events.
• Intercurrent events influence the statistical presentation of ATE by modifying treatments, variables, and target populations.
• The proposed mapping provides a novel way to integrate causal inference into the estimand framework, enhancing its theoretical foundation.

Conclusions:

• Interpreting the estimand framework using causal inference offers a stronger theoretical basis for its application in clinical trials.
• This causal perspective is valuable for both academic and pharmaceutical industry clinical trials.
• The insights may also benefit observational studies seeking to understand causal inference theories.