Consider a study with binary exposure, outcome, and confounder, where the confounder is nondifferentially misclassified. Epidemiologists have long accepted the unproven but oft-cited result that, if the confounder is binary, then odds ratios, risk ratios, and risk differences that control for the mismeasured confounder will lie between the crude and the true measures. In this paper, we provide an analytic proof of the result in the absence of a qualitative interaction between treatment and confounder, and we demonstrate via counterexample that the result need not hold when there is such a qualitative interaction. We also present an analytic proof of the result for the effect of treatment among the treated and describe extensions to measures conditional on or standardized over other covariates.
From the Departments of aBiostatistics and bEpidemiology, Harvard University, Boston MA.
Submitted 15 September 2011; accepted 27 December 2011.
Supported by NIH training grant 5T32 AI 7358–22 (to E.L.O.) and NIH grant ES017876 (to T.J.V.). The authors reported no other financial interests related to this research.
Supplemental digital content is available through direct URL citations in the HTML and PDF versions of this article (www.epidem.com). This content is not peer-reviewed or copy-edited; it is the sole responsibility of the author.
Editors' note:Related articles appear on pages 440 and 443.
Correspondence: Elizabeth L. Ogburn, HSPH Program on Causal Inference, Department of Epidemiology, 677 Huntington Ave, Boston, MA 02115. E-mail: email@example.com.