A DAG-based comparison of interventional effect underestimation between composite endpoint and multi-state analysis...

Area of Science:

• Cardiovascular clinical trials
• Biostatistics
• Health research methodology

Background:

• Composite endpoints (hospital admissions, death) are key in cardiovascular trials.
• Cox proportional hazards models for time to first event are standard but debated for recurrent events.
• Potential biases of analyzing only first events are often overlooked.

Purpose of the Study:

• Investigate bias in treatment effect estimates when considering first vs. multiple events.
• Utilize directed acyclic graphs (DAGs) and simulation studies for bias analysis.
• Motivated by a heart failure randomized controlled trial.

Main Methods:

• Directed acyclic graphs (DAGs) to identify bias sources.
• Simplified examples and simulation studies to investigate bias.
• Comparison of first-event-only analysis versus recurrent event analysis.

Main Results:

• Cox models restricting to first events are prone to selection and direct effect bias, causing underestimation.
• Incorporating recurrent events reduces these biases.
• Proportional hazards-based multi-state models decrease bias and increase power for recurrent composite endpoints.

Conclusions:

• Including multiple episodes per individual in primary analysis reduces bias in treatment effect estimates.
• Findings support moving beyond the first-event-only paradigm.
• This approach enhances information use and interpretability in cardiovascular research.