Treatment Comparisons in Adaptive Platform Trials Adjusting for Temporal Drift.

Area of Science:

• Biostatistics
• Clinical Trial Design
• Causal Inference

Background:

• Adaptive platform trials (APTs) allow dynamic entry/exit of treatment arms.
• Non-concurrent controls and temporal drift pose challenges in long-term APTs.
• Adjusting for temporal changes is crucial for valid treatment comparisons in APTs.

Purpose of the Study:

• To propose novel causal inference methods for treatment comparisons in APTs.
• To develop approaches that specifically address temporal drift and unmeasured confounders.
• To evaluate the performance of proposed methods via simulation.

Main Methods:

• Propensity score weighting methods are utilized for treatment effect estimation.
• Two distinct approaches are proposed, one being doubly robust.
• The doubly robust method ensures validity if either propensity score or time effect models are correct.

Main Results:

• Proposed methods demonstrate well-controlled Type I error rates.
• High statistical power is achieved in simulations, even with unmeasured confounders.
• The approaches effectively adjust for temporal drift in APTs.

Conclusions:

• The proposed propensity score weighting methods offer robust solutions for treatment comparisons in APTs.
• These methods enhance the validity and reliability of findings from long-duration adaptive trials.
• The doubly robust approach provides an added layer of protection against model misspecification.