Adjusting for incomplete baseline covariates in randomized controlled trials: a cross-world imputation framework

Area of Science:

• Biostatistics
• Clinical Trials
• Statistical Methods

Background:

• Adjusting for baseline covariates in randomized controlled trials (RCTs) enhances treatment effect precision.
• Missing covariate data is a common challenge in RCTs, potentially biasing results.
• Existing methods like single imputation and the missingness-indicator method (MIM) offer efficiency gains over ignoring covariates.

Purpose of the Study:

• To introduce a novel theoretical imputation framework, cross-world imputation (CWI), for handling missing covariates.
• To compare the efficiency of single imputation and MIM within the CWI framework.
• To identify conditions for optimal efficiency in covariate adjustment with missing data.

Main Methods:

• Development of the cross-world imputation (CWI) theoretical framework.
• Analysis of single imputation and MIM as special cases within CWI.
• Derivation of theoretical conditions for efficiency equivalence between methods.
• Simulation studies and real-world data analysis (Childhood Adenotonsillectomy Trial).

Main Results:

• The missingness-indicator method (MIM) implicitly optimizes CWI values, achieving maximal efficiency.
• Single imputation can attain MIM's efficiency under specific derived conditions.
• Both methods demonstrate efficiency gains compared to unadjusted analyses.

Conclusions:

• The CWI framework provides a unified view for understanding and comparing imputation strategies for missing covariates.
• MIM is shown to be optimally efficient, while single imputation's efficiency is condition-dependent.
• Findings have practical implications for statistical analysis in clinical trials with missing covariate data.