Estimation Methods for Estimands Using the Treatment Policy Strategy; a Simulation Study Based on the PIONEER 1 Trial

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Pharmaceutical Research

Background:

• Treatment policy strategy is common in Phase III trials for handling intercurrent events.
• Retrieved dropout methods, while used, face challenges like variance inflation and model fitting due to data sparsity.

Purpose of the Study:

• Introduce and evaluate likelihood-based retrieved dropout methods.
• Compare these new methods against existing retrieved dropout approaches, simpler models, and reference-based multiple imputation.
• Investigate statistical properties and performance in complex clinical trial scenarios.

Main Methods:

• Developed likelihood-based versions of retrieved dropout approaches.
• Conducted simulations using data from the PIONEER 1 Phase III clinical trial (Type II diabetes).
• Compared statistical properties including bias and variance inflation against alternative imputation and analysis methods.

Main Results:

• Likelihood-based methods showed similar statistical properties to multiple imputation equivalents.
• All retrieved dropout approaches exhibited high variance, a significant issue.
• Retrieved dropout methods were less biased than reference-based approaches, indicating a bias-variance trade-off.

Conclusions:

• High variance inflation in retrieved dropout methods is often more problematic than bias.
• Simpler retrieved dropout models may be suitable for primary analysis if post-event data is largely observed.
• Jump-to-reference shows promise for symptomatic treatments due to power and handling missing data, despite bias concerns.