Statistical Modeling to Adjust for Time Trends in Adaptive Platform Trials Utilizing Non-Concurrent Controls

Area of Science:

• Biostatistics
• Clinical Trial Design
• Pharmaceutical Research

Background:

• Platform trials increasingly use non-concurrent control (NCC) data for efficiency.
• Incorporating NCC data can improve statistical power and reduce sample size requirements.
• Temporal drifts in NCC data can introduce bias into effect estimators.

Purpose of the Study:

• To propose and evaluate frequentist model-based approaches for analyzing late-entering arms using NCC data.
• To mitigate potential bias introduced by temporal drifts when using NCC data.
• To adjust for time-varying effects within platform trial analyses.

Main Methods:

• Proposed two extensions to existing time-adjustment models for NCC data.
• Investigated fixed-length calendar time intervals and alternative model-based adjustments (random effects, polynomial splines).
• Evaluated performance via simulation studies and a case study.

Main Results:

• Spline-based time adjustment controlled type I error for smooth time trends and offered power gains.
• The fixed-effect model with period adjustment demonstrated robustness for arbitrary time trends, assuming equality across arms.
• Period-adjusted models are preferred for trials with sudden time trend changes when using NCC data.

Conclusions:

• New frequentist models effectively leverage NCC data while adjusting for temporal effects in platform trials.
• Model choice depends on the nature of temporal trends and data characteristics.
• The period-adjustment model offers a robust solution for handling time-varying effects with NCC data, particularly in complex scenarios.