A critical look at methods for handling missing covariates in epidemiologic regression analyses

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Missing covariate values are common in epidemiologic studies.
• Simple methods like complete-subject analysis or imputation can introduce bias.
• These methods are frequently used despite known limitations.

Purpose of the Study:

• To review and demonstrate the limitations of simple methods for handling missing data in logistic regression.
• To compare simple methods with more sophisticated statistical techniques.
• To provide recommendations for handling missing data in epidemiologic research.

Main Methods:

• Review of existing literature on missing data handling in logistic regression.
• Simulation experiments to illustrate bias in simple methods.
• Comparison of multiple imputation with simple methods using a case-control study.

Main Results:

• The missing-data indicator method can cause severe bias, even with data missing completely at random.
• Regression imputation is sensitive to model misspecification.
• Complete-subject analysis can be less biased than other simple methods.
• Multiple imputation yielded different results than simple methods in a real-world case-control study.

Conclusions:

• Epidemiologists should avoid the missing-indicator method.
• More sophisticated methods (e.g., multiple imputation) are superior for handling missing data.
• Advanced methods are recommended when a significant proportion of data is missing.