Bias and Precision of the "Multiple Imputation, Then Deletion" Method for Dealing With Missing Outcome Data

Area of Science:

• Epidemiologic research
• Statistical analysis
• Biostatistics

Background:

• Multiple imputation (MI) is a common method for handling missing data in epidemiology.
• The multiple imputation, then deletion (MID) method is an alternative when both exposure and outcome data are missing.
• The performance of MID with auxiliary variables for incomplete outcomes requires further investigation.

Purpose of the Study:

• To evaluate the performance of standard MI versus MID.
• To assess their behavior in regression settings with missing exposure and outcome data.
• To investigate the impact of auxiliary variables associated with the incomplete outcome.

Main Methods:

• Simulated data were used for evaluation.
• Regression settings with missing outcomes and exposures were modeled.
• An auxiliary variable associated with the incomplete outcome was included in imputation models.

Main Results:

• When auxiliary variables were unrelated to outcome missingness, both methods showed minimal bias, with standard MI being more efficient.
• When auxiliary variables were associated with outcome missingness, MID produced significantly biased parameter estimates.
• Standard MI consistently provided less biased estimates across tested scenarios.

Conclusions:

• Standard MI is preferable to MID when auxiliary variables are associated with an incomplete outcome.
• Researchers should exercise caution when using MID in the presence of such auxiliary variables.
• Standard MI offers a more robust approach for handling missing data in complex epidemiologic studies.