Multiple imputation in public health research

Area of Science:

• Public Health
• Biostatistics
• Epidemiology

Background:

• Missing data is a significant challenge in public health research.
• Mean or median imputation are common due to ease of use, but may not be optimal.
• Multiple imputation offers superior statistical properties but is underutilized.

Purpose of the Study:

• To compare the performance of multiple imputation against simpler methods.
• To evaluate the impact of covariate selection in multiple imputation models.
• To assess standard deviation estimates from different imputation techniques.

Main Methods:

• Comparison of multiple imputation with mean/median imputation and complete case analysis.
• Analysis of two real-world public health studies.
• A simulation study to explore variations in multiple imputation methods.

Main Results:

• All imputation methods yielded similar results in real studies, except for complete case analysis.
• Simulation revealed significant differences among multiple imputation strategies based on covariate selection.
• Using all available covariates in multiple imputation models reduced coefficient estimate bias compared to using few covariates.
• Multiple imputation provided better standard deviation estimates than single mean imputation.

Conclusions:

• Multiple imputation is a valuable tool for handling missing data in public health research.
• The choice of covariates in multiple imputation models significantly impacts the accuracy of results.
• Further adoption of multiple imputation is recommended for robust public health data analysis.