Estimation of Treatment Policy Estimands for Continuous Outcomes Using Off-Treatment Sequential Multiple Imputation.

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Pharmaceutical Research

Background:

• The International Council for Harmonisation E9 (R1) guideline emphasizes precise estimand definition, including handling of intercurrent events (IEs) like treatment discontinuation.
• Traditional methods for continuous repeated measures, such as mixed models for repeated measures (MMRM) and multiple imputation (MI), often use a treatment policy strategy that ignores discontinuation.
• This approach may introduce bias if outcomes differ post-discontinuation or if missing data is more prevalent among discontinuers.

Purpose of the Study:

• To propose and evaluate novel multiple imputation (MI) models designed to accommodate differences in patient outcomes before and after treatment discontinuation.
• To assess the performance of these MI models in the context of planning a Phase 3 clinical trial for a respiratory disease.
• To compare the proposed MI models against traditional analyses that ignore treatment discontinuation.

Main Methods:

• Development of a set of multiple imputation (MI) models capable of handling intercurrent events, specifically treatment discontinuation.
• Evaluation of these MI models using simulated data within the framework of a planned Phase 3 respiratory disease trial.
• Comparison of bias and variance introduced by traditional methods versus the proposed MI models.

Main Results:

• Analyses that ignore treatment discontinuation can lead to substantial bias and underestimation of variability in treatment effects.
• The proposed multiple imputation (MI) models demonstrate the ability to correct for bias introduced by treatment discontinuation.
• While correcting bias, the proposed MI models inevitably result in an increase in the variance of the estimated treatment effect.

Conclusions:

• The proposed multiple imputation (MI) models offer an improvement over traditional analyses that ignore treatment discontinuation in clinical trials.
• The choice of the optimal MI model depends on specific trial characteristics, including design, disease context, and the pattern of observed and missing data post-discontinuation.
• Careful consideration of MI model selection is crucial for accurate estimation of treatment effects in the presence of intercurrent events.