Handling Partially Observed Trial Data After Treatment Withdrawal: Introducing Retrieved Dropout Reference-Base...

Area of Science:

• Clinical Trials Methodology
• Statistical Analysis in Medicine
• Bayesian Statistics

Background:

• ICH E9(R1) Addendum recommends treatment-policy strategy for intercurrent events like treatment withdrawal.
• Early patient withdrawal in studies creates missing data, complicating primary outcome analysis.
• Existing multiple imputation methods for retrieved dropout can face estimation challenges.

Purpose of the Study:

• To introduce a novel multiple imputation method for handling missing data due to treatment withdrawal.
• To address challenges in parameter estimation within retrieved dropout imputation models.
• To improve the reliability of statistical analyses in clinical trials with early treatment termination.

Main Methods:

• Developed a novel retrieved dropout reference-base centred multiple imputation approach.
• Combined a core reference-based model with a retrieved dropout compliance model.
• Utilized mildly informative Bayesian priors for parameters that are difficult to estimate.
• Incorporated both on- and off-treatment data for imputation model parameterization.

Main Results:

• The novel approach alleviates the need for complex analysis rules for non-estimable or poorly estimated parameters.
• Reduces unrealistically large standard errors in the resulting statistical analyses.
• Provides a more robust method for handling missing data in treatment-policy estimands.

Conclusions:

• Retrieved dropout reference-base centred multiple imputation offers a robust solution for missing data in clinical trials.
• This Bayesian approach enhances the accuracy and stability of treatment effect estimations.
• Facilitates more reliable analysis when patients withdraw from randomized treatment early.