Identification and Estimation of Causal Effects Using Non-Concurrent Controls in Platform Trials.

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Epidemiology

Background:

• Platform trials offer flexible multi-arm evaluation of treatments for a single disease.
• They allow dynamic entry/exit of treatment arms, utilizing both concurrent and non-concurrent controls.

Purpose of the Study:

• To address challenges in estimating treatment effects using non-concurrent controls in platform trials.
• To identify optimal estimands and estimation strategies for maximizing efficiency and validity.

Main Methods:

• Discussed identification and estimation assumptions for common estimands in platform trials.
• Proposed a covariate-adjusted doubly robust estimator targeting the concurrent average treatment effect (cATE).
• Utilized simulations and applied the method to the ACTT platform trial.

Main Results:

• Targeting cATE with the proposed estimator provides robust efficiency gains without unwarranted assumptions.
• Collecting prognostic variables is crucial for efficiency, more so than relying on non-concurrent controls.
• The method applied to the ACTT trial showed a 20% precision improvement over naive estimators.

Conclusions:

• The covariate-adjusted doubly robust estimator for cATE is the most robust strategy for platform trials.
• Relying on non-concurrent controls introduces untestable extrapolation assumptions, often invalid.
• Prioritizing prognostic variable collection enhances efficiency more reliably than non-concurrent controls.