Estimating the effect of multiple imputation on incomplete longitudinal data with application to a randomized...

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Analyzing incomplete longitudinal data is crucial in various scientific fields.
• Mixed effects models and generalized estimating equations are common methods for such analyses.
• Multiple imputation is often used to handle missing data.

Purpose of the Study:

• To compare the performance of multiple imputation combined with mixed effects models and generalized estimating equations versus using these models alone for analyzing incomplete longitudinal data.
• To evaluate the impact of missing data mechanisms (ignorable and nonignorable) on estimation bias and variance.
• To assess these methods in the context of a randomized controlled clinical trial.

Main Methods:

• Comparison of statistical methods for incomplete longitudinal data.
• Application of multiple imputation (MI).
• Use of mixed effects models (MEM) and generalized estimating equations (GEE).
• Assessment under ignorable and nonignorable missing data assumptions.
• Evaluation in a randomized controlled trial (RCT) setting.

Main Results:

• Multiple imputation, when combined with MEM or GEE, often resulted in overestimated variances.
• MI-enhanced methods may produce more biased estimates compared to simpler methods like last observation carried forward (LOCF).
• MEM or GEE applied alone yielded more unbiased estimates, regardless of the missing data mechanism.

Conclusions:

• For incomplete longitudinal data, mixed effects models or generalized estimating equations alone are preferable to using them with multiple imputation.
• Multiple imputation may introduce bias and inflate variance estimates in these contexts.
• The findings have implications for the analysis of clinical trial data with missing observations.