Multiple imputation methods for handling incomplete longitudinal and clustered data where the target analysis is a...

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Multiple imputation (MI) is widely used for missing data.
• Two main MI approaches exist: joint modeling and fully conditional specification (FCS).
• The compatibility of MI methods with linear mixed-effects models (LMMs) for longitudinal/clustered data with random effects is unclear.

Purpose of the Study:

• To compare the performance of seven MI methods for missing data in longitudinal and clustered settings.
• To assess the theoretical compatibility between imputation models and LMMs.
• To evaluate parameter estimation consistency for LMMs with random intercepts and slopes.

Main Methods:

• Compared seven MI methods under joint modeling and FCS approaches.
• Assessed theoretical compatibility of imputation and LMMs.
• Conducted simulation studies for longitudinal and clustered data.
• Motivated simulations by the Longitudinal Study of Australian Children (LSAC) data.

Main Results:

• MI method performance varied based on whether covariates had fixed or random effects and if the outcome had missing values.
• Compatible imputation and analysis models led to consistent estimation of regression parameters and variance components.
• Demonstrated findings using LSAC data analysis.

Conclusions:

• Compatible imputation and analysis models are crucial for consistent LMM parameter estimation.
• The choice of MI method should consider the structure of missingness and the LMM specification.
• Findings provide guidance for handling missing data in complex longitudinal and clustered analyses.