On the nondifferential misclassification of a binary confounder.

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Nondifferential misclassification of confounders can bias effect estimates.
• A commonly accepted principle suggests bias correction for misclassified binary confounders moves estimates closer to the true value.

Purpose of the Study:

• To analytically prove the bias correction principle for binary confounders.
• To identify conditions under which this principle fails.
• To extend the findings to other causal measures and covariate adjustment strategies.

Main Methods:

• Analytic mathematical proofs were employed.
• Counterexamples were constructed to demonstrate limitations.
• The study considered binary exposure, outcome, and confounder with nondifferential misclassification.

Main Results:

• The principle holds in the absence of qualitative interaction between exposure and confounder.
• A qualitative interaction can invalidate the bias correction, leading to estimates outside the expected range.
• Analytic proofs were derived for the 'effect of treatment on the treated' and other covariate adjustment methods.

Conclusions:

• The oft-cited result regarding bias correction for misclassified binary confounders is analytically proven under specific conditions.
• Qualitative interactions represent a critical exception to this rule.
• The findings have implications for causal inference and confounding adjustment in observational studies.