Adaptive Weight Selection for Time-To-Event Data Under Non-Proportional Hazards

Area of Science:

• Biostatistics
• Clinical Trial Design
• Survival Analysis

Background:

• Standard clinical trials for time-to-event endpoints often assume proportional hazards, using a single-stage log-rank test.
• This rigid approach is problematic when the proportional hazards assumption is violated or effect sizes are unknown.
• Existing methods lack flexibility, potentially leading to inconclusive trial results.

Purpose of the Study:

• To introduce a more flexible and robust procedure for clinical trial planning with time-to-event endpoints.
• To address the limitations of assuming proportional hazards and the lack of prior knowledge on effect sizes.
• To improve the success rate of clinical trials by offering a more adaptable design.

Main Methods:

• Employs an adaptive multi-stage design instead of a traditional single-stage approach.
• Utilizes combination-type tests in the initial stage for robustness under uncertain deviation patterns.
• Incorporates Royston-Parmar spline models for survival curve extrapolation to inform subsequent stages.

Main Results:

• Demonstrates through a real-world example that the proposed approach can salvage trials that would otherwise be inconclusive.
• Simulation studies confirm sufficient statistical power performance.
• The method maintains greater flexibility compared to standard procedures.

Conclusions:

• The proposed adaptive multi-stage procedure offers a more flexible and robust alternative for clinical trials with time-to-event endpoints.
• This approach is particularly beneficial when prior knowledge is limited or the proportional hazards assumption is questionable.
• The methodology enhances trial success probability and statistical power.