Invited commentary on Pearl and principal stratification.

Area of Science:

• Clinical Trials Methodology
• Causal Inference
• Biostatistics

Background:

• The principal stratum framework, proposed by Frangakis and Rubin (2002), offers a method for analyzing clinical trial data involving post-randomization variables.
• This framework has been applied to topics such as treatment compliance, mediation, and surrogacy.

Purpose of the Study:

• To evaluate the utility of the principal stratum framework in the context of the Women's Health Initiative (WHI) randomized trials.
• To determine if confining analyses to the principal stratum framework restricts the scope of scientifically relevant questions, particularly in postmenopausal hormone therapy research.

Main Methods:

• The study considers the application of the principal stratum framework and the potential outcomes formulation in the WHI trials.
• It examines the relationship between the framework and issues of study reliability and causal interpretation.

Main Results:

• The essential issues for causal interpretation and study reliability are primarily related to avoiding context-specific biases.
• These biases are often not closely linked to whether treatment effects are represented by potential outcomes contrasts within the principal stratum framework.

Conclusions:

• While principal strata questions can be scientifically interesting, restricting analyses solely to this framework may prevent addressing crucial questions in certain research settings, such as postmenopausal hormone therapy.
• The study suggests that the principal stratum framework might unduly restrict the scientific questions addressable in clinical trial analyses involving post-randomization variables.