Multiple Imputation for General Missing Data Patterns in the Presence of High-dimensional Data.

Area of Science:

• Biostatistics
• Data Science
• Biomedical Research

Background:

• Missing data is a common challenge in biomedical research.
• High-dimensional data necessitates advanced imputation techniques.
• Existing methods struggle with general missing data patterns.

Purpose of the Study:

• To develop and evaluate multiple imputation by chained equations (MICE) methods for high-dimensional data with general missing patterns.
• To leverage regularized regression within the MICE framework for improved imputation accuracy.
• To compare the proposed methods against existing imputation techniques.

Main Methods:

• Investigated two novel MICE approaches incorporating regularized regression for imputation.
• Conducted simulation studies to assess performance against established methods.
• Applied the proposed methods to real-world biomedical datasets.

Main Results:

• The proposed MICE approach using an indirect application of regularized regression demonstrated superior performance in minimizing bias.
• Simulation results indicated improved accuracy for handling complex missing data patterns.
• The methods were successfully illustrated on two distinct biomedical datasets.

Conclusions:

• The novel MICE methods offer a robust solution for imputing missing values in high-dimensional biomedical data.
• Regularized regression integrated into MICE effectively addresses general missing data patterns.
• The proposed techniques provide a valuable advancement for statistical analysis in research with missing data.