A bias-corrected estimator in multiple imputation for missing data

Area of Science:

• Statistics
• Biostatistics
• Medical Informatics

Background:

• Multiple imputation (MI) is widely used for missing data in medical studies.
• Existing MI methods, especially parametric ones, risk biased estimates if the imputation model is misspecified.
• Nonparametric MI approaches exist but can be complex to implement.

Purpose of the Study:

• To propose a new multiple imputation method that yields consistent estimates.
• To address the challenge of potential misspecification in imputation models.
• To improve the robustness of statistical analyses with missing data.

Main Methods:

• Developed a novel MI technique using an easily implementable imputation model.
• Incorporated a likelihood function correction with density ratio weighting.
• Employed nonparametric estimation for the true conditional density, but not for imputation itself.

Main Results:

• The proposed method provides consistent (asymptotically unbiased) final estimates.
• Theoretical analysis and simulation studies demonstrate the method's effectiveness.
• The approach was validated using a real-world dataset from the Duke Cardiac Catheterization Coronary Artery Disease Diagnostic Dataset.

Conclusions:

• The new MI method offers a robust solution for handling missing data in medical research.
• It mitigates bias issues arising from imputation model misspecification.
• This technique enhances the reliability of statistical inference in the presence of incomplete data.