Missing data mechanisms in a dose-finding adaptive trial

Area of Science:

• Biostatistics
• Clinical Trial Design
• Statistical Methods

Background:

• Adaptive clinical trials are increasingly used.
• The impact of missing data on adaptive trial designs is under-researched.
• Crossover study designs are common in clinical research.

Purpose of the Study:

• To investigate the effects of various missing data mechanisms on an adaptive three-period crossover study.
• To compare the performance of a maximizing procedure in adaptive trials with missing data against a traditional nonadaptive design.
• To introduce and evaluate a mixture missing mechanism (MMM) in the context of adaptive crossover trials.

Main Methods:

• Simulated data from a three-period crossover study under different missing data mechanisms: missing completely at random (MCAR), missing at random (MAR), and missing not at random (MNAR).
• Introduction and application of a mixture missing mechanism (MMM).
• Comparison of an adaptive trial using a maximizing procedure with a nonadaptive equal allocation crossover study.

Main Results:

• Missing data mechanisms can lead to biased treatment effect estimates in adaptive trials.
• In equal allocation designs, bias is uniform across treatments, preserving unbiased treatment comparisons.
• The maximizing procedure in adaptive trials exhibits non-uniform bias, resulting in biased treatment comparisons.
• Missing not at random (MNAR) and mixture missing mechanism (MMM) introduce substantial bias in the placebo group, significantly reducing statistical power.

Conclusions:

• Missing data poses a significant challenge to the integrity of adaptive trial results, particularly with the maximizing procedure.
• The choice of adaptive design and the nature of missing data mechanisms critically influence the validity of treatment effect estimates.
• Further research is needed to develop robust methods for handling missing data in adaptive clinical trials to ensure reliable outcomes.