Comprehensive implementations of multiple imputation using retrieved dropouts for continuous endpoints.

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Metabolic Disease Research

Background:

• Mixed Models for Repeated Measures (MMRM) face challenges with missing data in metabolic diseases.
• Multiple Imputation based on Retrieved Dropouts (MIRD) is emerging as a standard for longitudinal endpoints.
• Existing MIRD methods often rely on last on-treatment data for imputation.

Purpose of the Study:

• To introduce and evaluate a novel class of MIRD approaches utilizing all available data from retrieved dropouts (RDs).
• To compare the performance of new MIRD methods against established MIRD and other statistical approaches like MMRM.
• To assess type-I error and power rates under various missing data scenarios.

Main Methods:

• Proposed new MIRD methods using all available RD data, implemented via one-step MCMC or two-step regression.
• Applied ANCOVA post-imputation and Rubin's rule for combining estimates.
• Conducted extensive simulation studies and analyzed two real-world Phase 3 clinical trial datasets.

Main Results:

• The new MIRD class demonstrated performance comparable to the established MIRD approach.
• The one-step MCMC MIRD method showed better type-I error control and increased power in specific scenarios.
• Real-world data analysis confirmed the enhanced power of the new MIRD class, especially in larger datasets.

Conclusions:

• The proposed MIRD approach using all observed RD data is a robust and powerful alternative for longitudinal continuous endpoints.
• The one-step MCMC implementation offers advantages in statistical power and type-I error control.
• This new MIRD class is anticipated to be more easily implemented in clinical trial reporting due to simpler programming.